| [ |
| { |
| "task_id": "chipseq-peak-calling", |
| "name": "ChIP-seq Peak Calling: TAL1 Binding Site Comparison", |
| "description": "This task analyzes ChIP-seq data for the TAL1 transcription factor in two mouse hematopoietic cell types: G1E erythroid progenitor cells and megakaryocytes. Eight single-end FASTQ files (4 TAL1 ChIP samples + 4 input controls, all subset to chromosome 19) are provided along with a mm10 chr19 reference genome. The data is from Wu et al. 2014 (GEO GSE51338).", |
| "task_prompt": "Identify TAL1 transcription factor binding sites in G1E cells and Megakaryocytes from ChIP-seq data, then compare peaks between cell types to find shared and cell-type-specific binding. A mm10 chr19 reference genome is provided in the reference/ directory. The output should be a CSV file with the following columns: 'chrom','start','end','name','score','strand','signal_value','pvalue','qvalue','peak','cell_type','status'.\n<example>chrom,start,end,name,score,strand,signal_value,pvalue,qvalue,peak,cell_type,status\nchr19,5798729,5798998,G1E_TAL1_peak_12,885,.,17.5082,91.8514,88.5255,277,G1E,shared</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/chipseq-peak-calling/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/chipseq-peak-calling/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/chipseq-peak-calling/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "bacterial-assembly", |
| "name": "Bacterial Genome Assembly and Annotation: MRSA Characterization", |
| "description": "This task analyzes a methicillin-resistant Staphylococcus aureus (MRSA) clinical isolate from paired-end Illumina MiSeq sequencing. The goal is to assemble the genome de novo and produce a comprehensive characterization report covering assembly quality, gene content, sequence typing, and antimicrobial resistance. The data is from Hikichi et al. 2019 (BioSample SAMD00180470), estimated genome size ~2.9 Mbp.", |
| "task_prompt": "Assemble and characterize an MRSA bacterial genome from paired-end Illumina reads. Report assembly quality, completeness, sequence type, gene content, and antimicrobial resistance. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_length,2911349\nnum_contigs,44\nn50,276459\ngc_content,32.77\nlargest_contig,589438\ncompleteness,C:100.0%[S:100.0%,D:0.0%],F:0.0%,M:0.0%,n:124\nmlst_scheme,saureus\nmlst_sequence_type,764\ncds_count,2717\ntrna_count,57\nrrna_count,9</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/bacterial-assembly/data.tar.gz" |
| } |
| ], |
| "reference_data": [], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/bacterial-assembly/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "mobile-elements", |
| "name": "Bacterial Mobile Genetic Element Characterization: MRSA Genomic Islands and AMR", |
| "description": "This task analyzes a methicillin-resistant Staphylococcus aureus (MRSA) clinical isolate from paired-end Illumina MiSeq sequencing. The goal is to assemble the genome and perform comprehensive characterization of mobile genetic elements, antimicrobial resistance determinants, and virulence factors. The report should cover assembly quality, gene content, insertion sequence elements, AMR genes identified by multiple methods, and virulence factors. The data is from Hikichi et al. 2019 (BioSample SAMD00180470), S. aureus estimated genome ~2.9 Mbp.", |
| "task_prompt": "Assemble an MRSA genome from paired-end Illumina reads and characterize its mobile genetic elements, antimicrobial resistance determinants, and virulence factors. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_length,2911349\nnum_contigs,44\nn50,276459\ngc_content,32.77\nlargest_contig,589438\ncompleteness,C:100.0%[S:100.0%,D:0.0%],F:0.0%,M:0.0%,n:124\ncds_count,2717\ntrna_count,57\nrrna_count,9\nis_elements_found,0\nintegrons_found,0\namr_genes_protein_method,45\namr_genes_nucleotide_method,21\namr_genes_confirmed_both,2\nvirulence_factors,83\nplasmid_contigs,0\nreplicon_types,0</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mobile-elements/data.tar.gz" |
| } |
| ], |
| "reference_data": [], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mobile-elements/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "outbreak-investigation", |
| "name": "Foodborne Pathogen Outbreak Investigation via WGS Phylogenomics", |
| "description": "This task analyzes whole-genome sequencing data from three Mycobacterium tuberculosis clinical isolates to determine outbreak clustering. The workflow combines reference-based SNP calling with de novo assembly, producing both SNP-based and core-gene phylogenies, pairwise SNP distance matrices, sequence typing, antimicrobial resistance profiling, and pan-genome analysis. The data consists of paired-end Illumina reads from three TB isolates and the M. tuberculosis H37Rv reference genome.", |
| "task_prompt": "Determine outbreak relatedness among three M. tuberculosis isolates using whole-genome sequencing data. A reference genome is provided in the reference/ directory. The output should include: outbreak_report.csv (metric,value format), assembly_stats.csv, mlst_results.csv, amr_summary.csv, snp_distance_matrix.tsv, snp_phylogeny.nwk, and gene_phylogeny.nwk.\n<example>metric,value\nnum_isolates,3\nreference_genome,M. tuberculosis H37Rv\nsnp_alignment_length,2703\ncompleteness,C:97.6%[S:96.0%,D:1.6%],F:1.6%,M:0.8%,n:124\npangenome_core_genes,3867\npangenome_total_genes,4066\nsnp_tree_available,yes\ngene_tree_available,yes</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/outbreak-investigation/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/outbreak-investigation/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/outbreak-investigation/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "atacseq-accessibility", |
| "name": "ATAC-seq Chromatin Accessibility Profiling", |
| "description": "This task analyzes ATAC-seq data from GM12878 cells to identify open chromatin regions and characterize transcription factor binding motifs. Paired-end reads enriched for chr22 are provided along with the hg38 chr22 reference, ENCODE blacklist regions, and gene annotations. The data is from Buenrostro et al. 2013 (SRR891268).", |
| "task_prompt": "Profile chromatin accessibility in GM12878 cells from ATAC-seq data. Identify open chromatin regions and characterize transcription factor binding motifs. A chr22 reference genome, blacklist regions, and gene annotations are provided in the reference/ directory. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\nsample,SRR891268\ntotal_reads,138630\nalignment_rate,32.33%\nmapped_reads,150244\nduplication_rate,0.048854\nchrM_reads,0\nnum_peaks,110\nfrip,0.0153\nknown_motifs_enriched,59</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/atacseq-accessibility/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/atacseq-accessibility/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/atacseq-accessibility/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "longread-assembly", |
| "name": "Nanopore Long-read Bacterial Genome Assembly", |
| "description": "This task assembles and characterizes a Salmonella enterica genome from Oxford Nanopore long-read sequencing data. A reference genome is provided for quality assessment. The data is from a Nanopore MinION run.", |
| "task_prompt": "Assemble a Salmonella enterica genome from Nanopore long reads. Assess read quality, assemble, polish, and characterize the resulting genome. A reference genome is provided in the reference/ directory. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\nnum_reads,114986.0\nmean_read_length,938.8\nmean_read_quality,11.3\ntotal_bases,107951527.0\nassembly_length,8456926\nnum_contigs,170\nn50,123189\ngc_content,51.88\nlargest_contig,418268\ngenome_fraction,90.728\nmisassemblies,7\ncompleteness,C:42.7%[S:38.7%,D:4.0%],F:41.9%,M:15.3%,n:124\ncds_count,16755\ntrna_count,116\nrrna_count,14\namr_genes,56</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/longread-assembly/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/longread-assembly/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/longread-assembly/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "hybrid-assembly", |
| "name": "Hybrid Genome Assembly from Illumina and Nanopore Data", |
| "description": "This task performs hybrid genome assembly combining Illumina short reads and Oxford Nanopore long reads from a Bacillus subtilis isolate. A reference genome is provided. Both paired-end Illumina FASTQ and Nanopore FASTQ files are in the data/ directory.", |
| "task_prompt": "Assemble a bacterial genome using both Illumina paired-end reads and Nanopore long reads. Polish the assembly and characterize it. A reference genome is in the reference/ directory. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\nassembly_length,3989415\nnum_contigs,39\nn50,984413\ngc_content,43.81\nlargest_contig,1072107\ngenome_fraction,98.519\ncircular_contigs,0\ncompleteness,C:100.0%[S:100.0%,D:0.0%],F:0.0%,M:0.0%,n:124\nmlst_scheme,bsubtilis\nmlst_sequence_type,7\ncds_count,3975\ntrna_count,37\nrrna_count,2\namr_genes,9</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/hybrid-assembly/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/hybrid-assembly/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/hybrid-assembly/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "sv-detection", |
| "name": "Bacterial Structural Variant and SNP Detection", |
| "description": "This task detects structural variants (deletions, duplications, inversions) and SNPs/indels in an MRSA genome from Illumina paired-end reads aligned to an assembled reference. The reference assembly is in the reference/ directory.", |
| "task_prompt": "Detect structural variants and small variants (SNPs, indels) in an MRSA genome from paired-end Illumina reads. A reference assembly is provided in the reference/ directory. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_reads,911977\nmapped_reads,840453\nsnps,33870\nindels,650\nstructural_variants,331\nsv_deletions,136\nsv_duplications,81\nsv_inversions,110\nsv_breakends,0</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/sv-detection/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/sv-detection/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/sv-detection/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "pangenome-evolution", |
| "name": "E. coli Pan-genome and Core Phylogeny", |
| "description": "This task performs pan-genome analysis of five Escherichia coli strains (K-12, O157:H7, CFT073, UTI89, APEC O1) to identify core, shell, and cloud genes, reconstruct core-gene phylogeny, and profile AMR and sequence types across strains.", |
| "task_prompt": "Analyze five E. coli genome assemblies to characterize the pan-genome, reconstruct a core-gene phylogeny, and profile antimicrobial resistance and sequence types. Genome assemblies are in the data/ directory. The output should include pangenome_report.csv (metric,value format), mlst_all.csv, amr_summary.csv, and core_phylogeny.nwk.\n<example>metric,value\nnum_genomes,5\ncore_genes,3432\nshell_genes,4269\ncloud_genes,0\ntotal_genes,7701\ntree_available,yes</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/pangenome-evolution/data.tar.gz" |
| } |
| ], |
| "reference_data": [], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/pangenome-evolution/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "metagenomic-profiling", |
| "name": "Metagenomic Assembly and Functional Profiling", |
| "description": "This task performs metagenomic analysis of paired-end sequencing data to assemble contigs, predict genes, detect antimicrobial resistance, and profile taxonomic composition.", |
| "task_prompt": "Analyze paired-end metagenomic sequencing reads. Assemble contigs, predict genes, classify taxonomy, and detect antimicrobial resistance genes. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\nassembly_length,2903797\nnum_contigs,98\nn50,58516\npredicted_genes,2828\nspecies_detected,84\namr_genes,20</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/metagenomic-profiling/data.tar.gz" |
| } |
| ], |
| "reference_data": [], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/metagenomic-profiling/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "phage-characterization", |
| "name": "Bacteriophage Genome Assembly and Functional Characterization", |
| "description": "This task assembles and characterizes a bacteriophage genome from paired-end Illumina sequencing data. The data contains reads from a known temperate phage. A reference genome is provided in the reference/ directory for comparison.", |
| "task_prompt": "Assemble a bacteriophage genome from paired-end Illumina reads and produce a comprehensive functional characterization including genome quality assessment, gene annotation with functional categories (lysis, lysogeny, replication, structural), and taxonomic classification. A reference genome is provided in the reference/ directory. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ngenome_length,174257\nnum_contigs,5\ngc_content,35.60\nclosest_hit,Enterobacteria phage RB59\ncds_count,305\ntrna_count,0\nlysis_genes,8\nlysogeny_genes,1\nreplication_genes,14\nstructural_genes,48\ncheckv_quality,Complete\ncheckv_completeness,100.0\nvirulence_factors,0\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/phage-characterization/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/phage-characterization/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/phage-characterization/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "genome-comparison", |
| "name": "Pairwise Bacterial Genome Comparison", |
| "description": "This task compares two E. coli genomes (K-12 laboratory strain vs CFT073 uropathogenic clinical isolate) to identify SNPs, structural rearrangements, and unique gene content between them.", |
| "task_prompt": "Compare two bacterial genomes to identify sequence differences and gene content variation. Report alignment statistics, SNPs, structural rearrangements, and shared vs unique genes. The two genome assemblies are in the data/ directory. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ngenome_a_length,4641652\ngenome_b_length,5231428\naligned_bases,3967944\naverage_identity,97.17\ntotal_snps,106261\ntotal_indels,4611\nbreakpoints,1037\nrelocations,44\ninversions,20\ncds_genome_a,4305\ncds_genome_b,4894\nshared_orthologs,3610\nunique_genes,1927\ntotal_pangenome,5537</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/genome-comparison/data.tar.gz" |
| } |
| ], |
| "reference_data": [], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/genome-comparison/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "mapping-qc", |
| "name": "Genome Mapping and Coverage Quality Assessment", |
| "description": "This task assesses the quality of whole-genome sequencing read alignment to a bacterial reference genome. Paired-end reads and a reference genome are provided.", |
| "task_prompt": "Align paired-end bacterial WGS reads to the provided reference genome and produce a comprehensive mapping quality report including alignment statistics, coverage metrics, and base quality. The reference genome is in the reference/ directory. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_reads,903562\nmapped_reads,832038\nmapping_rate,92.16\nproperly_paired_rate,90.06\naverage_base_quality,34.1\naverage_insert_size,372.0\nerror_rate,2.579704e-02\nmean_coverage,45.96\nmin_coverage,0\nmax_coverage,824\ncoverage_breadth_pct,93.6867</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mapping-qc/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mapping-qc/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mapping-qc/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "multisample-variants", |
| "name": "Multi-sample Variant Calling and Comparison", |
| "description": "This task calls variants in two closely related M. tuberculosis clinical isolates and compares them to identify shared and sample-specific mutations. Paired-end reads for two samples and a reference genome are provided.", |
| "task_prompt": "Call variants in two bacterial isolates relative to a reference genome, then compare to identify shared and unique mutations. Paired-end reads for sampleA and sampleB are in the data/ directory, reference in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_variants_sampleA,1475\ntotal_variants_sampleB,1593\nsnps_sampleA,1220\nsnps_sampleB,1291\nunique_to_sampleA,423\nunique_to_sampleB,541\nshared_variants,1052</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/multisample-variants/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/multisample-variants/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/multisample-variants/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "consensus-genome", |
| "name": "Bacterial Consensus Genome Generation", |
| "description": "This task generates a consensus genome sequence from M. tuberculosis Illumina reads by mapping to a reference, calling variants, and masking low-coverage regions. Paired-end reads and a reference genome are provided.", |
| "task_prompt": "Generate a consensus genome from bacterial sequencing reads by aligning to a reference, calling variants, masking low-coverage regions, and comparing the consensus back to the reference. Reads are in data/, reference in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\nreference_length,4411532\nconsensus_length,4411828\nmasked_bases_n,186284\nmasked_pct,4.22\nmean_coverage,53.61\nlow_coverage_regions,2435\nidentity_to_reference,99.88\naligned_bases,4223731\nsnps_vs_reference,1696</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/consensus-genome/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/consensus-genome/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/consensus-genome/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "gene-prediction", |
| "name": "Gene Prediction Method Comparison", |
| "description": "This task compares gene predictions from three different approaches on an E. coli K-12 genome to quantify agreement and identify method-specific predictions.", |
| "task_prompt": "Predict genes in a bacterial genome using three different approaches and compare their predictions. Identify overlapping and unique gene calls across methods. The genome assembly is in data/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ngenome_length,4641652\ngc_content,50.79\npredictor_a_genes,4319\npredictor_b_genes,4305\npredictor_c_genes,4319\noverlap_a_b,4305\noverlap_a_c,4313\noverlap_b_c,4300\nconsensus_all_three,4300</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/gene-prediction/data.tar.gz" |
| } |
| ], |
| "reference_data": [], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/gene-prediction/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "downsampling-analysis", |
| "name": "Read Downsampling and Assembly Quality Titration", |
| "description": "This task evaluates how sequencing depth affects assembly quality by subsampling reads at different fractions and assembling each subset. A reference genome is provided for quality assessment.", |
| "task_prompt": "Subsample bacterial sequencing reads at 25%%, 50%%, and 100%% fractions, assemble each subset, and compare assembly quality against a reference genome. Reads are in data/, reference in reference/. The output should be a CSV file with the following columns: 'fraction','total_length','num_contigs','n50','largest_contig','genome_fraction','num_reads'.\n<example>fraction,total_length,num_contigs,n50,largest_contig,genome_fraction,num_reads\n0.25,4320588,254,42834,109463,97.285,364597\n0.50,4351979,154,76641,204577,98.035,729819\n1.00,4367594,136,85401,228985,98.260,1459854</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/downsampling-analysis/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/downsampling-analysis/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/downsampling-analysis/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "plasmid-typing", |
| "name": "Plasmid Detection and Replicon Typing", |
| "description": "This task identifies plasmid sequences in an E. coli O157:H7 genome assembly, classifies replicon types, and profiles plasmid-borne antimicrobial resistance and virulence factors.", |
| "task_prompt": "Identify plasmid sequences in a bacterial genome assembly, classify their replicon types, and determine which antimicrobial resistance and virulence genes are plasmid-borne vs chromosomal. The genome assembly is in data/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ngenome_length,5594605\ntotal_contigs,3\nplasmid_clusters,2\nplasmid_total_length,96027\nchromosome_contigs,1\nreplicon_types_detected,2\namr_genes_genome,43\namr_genes_plasmid,0\nvirulence_factors_plasmid,7</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/plasmid-typing/data.tar.gz" |
| } |
| ], |
| "reference_data": [], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/plasmid-typing/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "genome-completeness", |
| "name": "Genome Completeness and Quality Assessment", |
| "description": "This task assesses the completeness and quality of a fragmented bacterial genome assembly using multiple metrics. A reference genome is provided for comparison.", |
| "task_prompt": "Assess the completeness and quality of a bacterial genome assembly using reference alignment, single-copy orthologs, and gene content analysis. The assembly is in data/, reference in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_length,8497890\nnum_contigs,170\nn50,123948\ngc_content,52.11\nlargest_contig,420634\ngenome_fraction,91.066\nmisassemblies,9\nduplication_ratio,1.038\ncompleteness_pct,37.9\nfragmented_pct,48.4\nmissing_pct,13.7\ncds_count,16767\ntrna_count,115\nrrna_count,14\ncoding_density_pct,72.9\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/genome-completeness/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/genome-completeness/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/genome-completeness/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "species-identification", |
| "name": "Multi-reference Bacterial Species Identification", |
| "description": "This task identifies an unknown bacterial isolate by comparing its genome against three reference genomes from different species (E. coli, Salmonella, S. aureus).", |
| "task_prompt": "Identify an unknown bacterial genome by comparing it against multiple reference genomes from different species. Report alignment identity and coverage to each reference, and determine the best match. The unknown genome is in data/, references in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ngenome_length,5179971\nnum_contigs,2\ngc_content,50.61\ntyping_scheme,ecoli_achtman_4\nsequence_type,95\nidentity_to_ecoli,97.15\naligned_to_ecoli_pct,85.93\nidentity_to_salmonella,84.76\naligned_to_salmonella_pct,35.77\nidentity_to_saureus,0.00\naligned_to_saureus_pct,0.00\nbest_match_species,Escherichia coli\nbest_match_identity,97.15\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/species-identification/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/species-identification/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/species-identification/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "viral-amplicon", |
| "name": "Viral Amplicon Surveillance Analysis", |
| "description": "This task performs SARS-CoV-2 amplicon sequencing analysis from paired-end reads. A reference genome and primer scheme are provided in the reference/ directory.", |
| "task_prompt": "Analyze viral amplicon sequencing data to call variants, generate a consensus genome, and determine the viral lineage and clade. A reference genome and primer BED file are in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ngenome_length,29906\nmasked_bases,82\nmasked_pct,0.27\nalignment_rate,97.04%\nmapped_reads,975406\nmean_coverage,4699.34\ntotal_variants_called,31\npass_variants,4\nlineage,Unassigned\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/viral-amplicon/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/viral-amplicon/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/viral-amplicon/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "bisulfite-methylation", |
| "name": "Bisulfite Sequencing DNA Methylation Analysis", |
| "description": "This task analyzes whole-genome bisulfite sequencing data to quantify DNA methylation at CpG, CHG, and CHH contexts. Paired-end reads and a reference genome are provided.", |
| "task_prompt": "Analyze bisulfite sequencing reads to map DNA methylation levels genome-wide. Report alignment statistics, methylation percentages in CpG/CHG/CHH contexts, and CpG site coverage. Reads are in data/, reference genome in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_read_pairs,20000\naligned_pairs,9890\nmapping_efficiency,49.5\nduplication_rate,0\ncpg_methylation_pct,100.0\nchg_methylation_pct,100.0\nchh_methylation_pct,100.0\ntotal_cpg_sites,6224\nmethylated_cpg_sites,6217</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/bisulfite-methylation/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/bisulfite-methylation/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/bisulfite-methylation/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "rnaseq-isoform", |
| "name": "RNA-seq Isoform Assembly and Quantification", |
| "description": "This task performs RNA-seq isoform-level analysis from paired-end reads aligned to a chr22 reference. Gene expression quantification, transcript assembly, and comparison to reference annotation are required.", |
| "task_prompt": "Align RNA-seq reads to a reference genome, assemble transcripts, quantify gene expression, and compare assembled isoforms to the reference annotation. Reads in data/, genome and gene annotation in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_reads,47605\nuniquely_mapped,15633\nunique_mapping_pct,32.84\nmultimapped_pct,23.29\nsplice_junctions,1271\nassembled_transcripts,14035\ngenes_expressed,130\ntranscript_sensitivity,95.9\ntranscript_precision,100.0\nassigned_read_pairs,10566\ngenes_with_counts,186\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/rnaseq-isoform/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/rnaseq-isoform/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/rnaseq-isoform/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "ancient-dna", |
| "name": "Ancient DNA Authentication and Damage Assessment", |
| "description": "This task processes degraded sequencing reads to assess DNA authenticity through damage patterns, endogenous content, and coverage metrics. A reference genome is provided.", |
| "task_prompt": "Process degraded DNA sequencing reads to assess authenticity and damage. Report alignment statistics, endogenous content, coverage, and deamination damage patterns at read termini. Reads in data/, reference in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_input_reads,2296\nmapped_reads,2296\nendogenous_pct,100.00\nduplication_rate,0.023529\nreads_after_dedup,2242\nmean_read_length,80.2\nmean_coverage,10.71\ncoverage_breadth_pct,96.73\ndamage_5prime_ct,0\ndamage_3prime_ga,0\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/ancient-dna/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/ancient-dna/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/ancient-dna/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "mirna-seq", |
| "name": "Small RNA-seq miRNA Discovery and Quantification", |
| "description": "This task analyzes small RNA sequencing data from human placenta to identify and quantify microRNAs. Single-end reads and miRNA reference databases are provided.", |
| "task_prompt": "Identify and quantify microRNAs from small RNA sequencing data. Trim adapters, align to miRNA references, count unique miRNAs, and identify the most abundant species. Reads in data/, miRNA references in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_reads,500000\npassed_filter,238018\nmature_mapped_reads,154932\nhairpin_mapped_reads,183608\nunique_mirnas_detected,394\ntop_mirna,hsa-miR-10a-5p\ntop_mirna_count,41120</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mirna-seq/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mirna-seq/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mirna-seq/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "gcms-metabolomics", |
| "name": "GC-MS Metabolomics Profiling: Brown Algae Salinity Adaptation", |
| "description": "This task analyzes GC-MS metabolomics data from Ectocarpus brown algae adapted to different salinity conditions (seawater, 5% NaCl freshwater, 100% NaCl freshwater). Six mzML files from low-resolution GC-MS are provided along with sample metadata, a spectral reference library (8 standard compounds), and alkane retention time standards for Kovats retention index calculation. The data is from Dittami et al. 2012 (Zenodo 16538501). The goal is to perform untargeted metabolomics profiling: peak detection, retention time alignment, feature grouping, spectral deconvolution into pseudospectra, retention index computation, and compound identification via spectral library matching.", |
| "task_prompt": "Profile metabolites from GC-MS data of brown algae (Ectocarpus) samples grown under different salinity conditions. Six mzML files are provided in data/ along with a sample metadata file, a spectral reference library, and alkane retention time standards in reference/. Detect chromatographic peaks across all samples, align retention times, group corresponding features, fill missing values, annotate isotope patterns and adduct groups, compute Kovats retention indices from the alkane standards, and identify metabolites by matching against the provided spectral library. The output should be a CSV file with columns: 'metric','value'.\n<example>metric,value\ntotal_peaks_detected,9492\nfeatures_before_alignment,1325\nfeatures_after_alignment,1333\nsamples_processed,6\nfill_rate_percent,100\npseudospectra_count,124\nisotope_annotations,200\nadduct_annotations,576\nretention_index_range,992-2700\nmean_retention_index,1917.5\nlibrary_spectra_count,8\ncompounds_matched,15\nunique_compounds_identified,8\nmean_match_score,0.733\nidentified_compounds,\"Alanine, 3TMS;Aspartic acid, 2TMS;Citric acid, 4TMS;D-Mannitol, 6TMS;Glycine, 3TMS;Pyroglutamic acid, 2TMS;Ribitol, 5TMS;Tryptamine, 2TMS\"\nmean_sample_correlation,0.567\nmean_feature_cv_percent,71.8</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/gcms-metabolomics/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/gcms-metabolomics/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/gcms-metabolomics/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "cutandrun", |
| "name": "CUT&RUN Epigenomic Profiling", |
| "description": "This task analyzes CUT&RUN sequencing data to identify histone modification peaks. Paired-end reads, a chr22 reference genome, and blacklist regions are provided.", |
| "task_prompt": "Identify histone modification enrichment peaks from CUT&RUN sequencing data using two independent peak calling approaches, then compute a consensus peak set. A reference genome and blacklist are in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_reads,33914\nmapped_reads,33914\nmapping_rate,100.00\nduplication_rate,0.046353\npeaks_caller_a,198\npeaks_caller_b,38\nconsensus_peaks,64\nfraction_reads_in_peaks,0.0238</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/cutandrun/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/cutandrun/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/cutandrun/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "scrna-full-pipeline", |
| "name": "Single-Cell RNA-seq Full Pipeline: Multi-Quantifier Analysis", |
| "description": "This task processes single-cell RNA-seq data from two mouse samples sequenced on the 10x Chromium v2 platform. Paired-end FASTQ files are provided for each sample (R1 = cell barcode + UMI, R2 = cDNA), along with a mouse chromosome 19 reference genome and gene annotation. The goal is to quantify gene expression per cell using multiple approaches, compare them, then perform quality control, doublet detection, batch correction, dimensionality reduction, clustering, marker gene identification, and differential expression between samples. The data is from the nf-core/scrnaseq test dataset.", |
| "task_prompt": "Process two single-cell RNA-seq samples from paired-end FASTQ files. The data/ directory contains subdirectories Sample_X/ and Sample_Y/ with 10x Chromium v2 FASTQ files (R1 has cell barcode + UMI; R2 has cDNA). The reference/ directory contains a mouse chr19 genome FASTA (genome.fa), gene annotation (genes.gtf), and a cell barcode whitelist (10xv2_whitelist.txt). Quantify gene expression per cell using at least two independent methods, compare the cell counts from each. Then merge samples, filter empty droplets and low-quality cells, detect and remove doublets, normalize, find highly variable genes, reduce dimensions, correct for batch effects between the two samples, cluster cells, find marker genes per cluster, and test for differentially expressed genes between the two samples. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>metric,value\ntotal_input_reads,225000\nnum_samples,2\ncells_before_filtering,8678\ncells_after_filtering,2914\ngenes_detected,267\nmedian_genes_per_cell,2\nmedian_counts_per_cell,2\npct_mitochondrial_median,0.0\ndoublet_fraction,0.0055\nnum_hvg,107\nnum_pcs_used,20\npca_variance_ratio_pc1,0.0357\nnum_clusters,12\ncluster_0_size,1323\ncluster_0_marker,Malat1\ncluster_1_size,1169\ncluster_1_marker,Malat1\ndominant_cell_type,unassigned\nde_genes_between_samples,0\ntop_de_gene,Gm32256\nquantifier_a_cells,8678\nquantifier_b_cells,10207</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/scrna-full-pipeline/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/scrna-full-pipeline/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/scrna-full-pipeline/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "crispr-screen", |
| "name": "CRISPR Screen Analysis: Drug Sensitivity Gene Discovery", |
| "description": "This task analyzes a pooled CRISPR knockout screen to identify genes that confer sensitivity to a cancer drug. Three FASTQ files represent different conditions of a Brunello sgRNA library screen: baseline (day 0), drug-treated (day 8), and vehicle control (day 8). A tab-separated guide library file maps 77,441 sgRNAs to ~19,000 human genes. The goal is to count sgRNA representation across conditions, apply multiple statistical methods to rank gene-level essentiality, compare drug-treated vs vehicle to identify drug-specific hits, and produce a comprehensive screen quality and hit report.", |
| "task_prompt": "Analyze a pooled CRISPR knockout screen from three conditions (baseline T0, drug-treated T8, vehicle control T8) to identify genes whose knockout confers drug sensitivity. A guide RNA library file is provided in reference/library.tsv (tab-separated: sgRNA_name, sequence, gene). Three FASTQ files in data/ contain sequencing reads from each condition. Reads contain vector backbone flanking the 20bp guide sequence. Count guide representation across all conditions, then apply at least two independent statistical methods to rank genes by depletion significance. Compare drug-treated results against vehicle control to distinguish drug-specific effects from general essentiality. The output should be a CSV file with columns: 'metric','value'.\n<example>metric,value\ntotal_counts_t0,172264\nzero_count_sgrnas_t0,16800\ngini_index_t0,0.495\nmedian_count_t0,2\ntotal_counts_drug,146594\nzero_count_sgrnas_drug,23444\ngini_index_drug,0.5575\nmedian_count_drug,1\ntotal_counts_vehicle,149774\nzero_count_sgrnas_vehicle,22403\ngini_index_vehicle,0.5466\nmedian_count_vehicle,1\ntotal_sgrnas,77441\nmapped_reads_t0,172264\nmapping_pct_t0,0.8201\nmapped_reads_drug,146594\nmapping_pct_drug,0.8125\nmapped_reads_vehicle,149774\nmapping_pct_vehicle,0.8238\ntop_depleted_gene_rra,MGME1\ntop_depleted_fdr_rra,0.506188\ntop_depleted_lfc_rra,-0.48677\ntop_enriched_gene_rra,SNTB1\ntop_enriched_fdr_rra,0.670792\nnum_depleted_genes_fdr25_rra,0\nnum_enriched_genes_fdr25_rra,0\ntop_gene_mle,SNTB1\ntop_gene_mle_fdr,0\ntop_gene_mle_beta,1.7188\nnum_significant_mle_fdr25,1\nnum_depleted_genes_vehicle_fdr25,0\nhigh_confidence_hits,0\nmoderate_confidence_hits,1\ndrug_specific_depleted_genes,0</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/crispr-screen/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/crispr-screen/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/crispr-screen/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "amplicon-microbiome", |
| "name": "16S Amplicon Microbiome: Community Profiling and Functional Prediction", |
| "description": "This task analyzes 16S rRNA amplicon sequencing data from four paired-end samples representing two treatment groups (control and treated) at two sites. The V4 region was amplified with primers 515F (GTGYCAGCMGCCGCGGTAA) and 806R (GGACTACNVGGGTWTCTAAT). A SILVA 138.1 taxonomy reference database is provided in the reference/ directory. Sample metadata including treatment group and site information is in data/metadata.tsv. The goal is to perform a complete microbiome analysis: remove primers, denoise reads to identify sequence variants, assign taxonomy, build a phylogenetic tree, predict functional potential, calculate diversity metrics, and test for differentially abundant taxa between treatment groups.", |
| "task_prompt": "Analyze 16S rRNA amplicon sequencing data from four paired-end samples. Remove primer sequences (forward: GTGYCAGCMGCCGCGGTAA, reverse: GGACTACNVGGGTWTCTAAT), denoise to identify sequence variants, assign taxonomy using the reference database in reference/, construct a phylogenetic tree from the sequence variants, predict functional pathways, calculate alpha diversity (Shannon, Simpson, observed richness, phylogenetic diversity) and beta diversity, and test for differentially abundant taxa between treatment groups defined in data/metadata.tsv. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>metric,value\ntotal_samples,4\ntotal_input_reads,8704\ntotal_filtered_reads,7528\ntotal_sequence_variants,365\nchimeras_removed,0\nclassified_variants,337\ntop_phylum,Patescibacteria\ntop_phylum_pct,27\nmean_shannon,3.9849\nmean_simpson,0.9585\nmean_observed_richness,115\nmean_phylogenetic_diversity,17.8772\npredicted_pathways,483\ndifferentially_abundant_features,0</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/amplicon-microbiome/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/amplicon-microbiome/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/amplicon-microbiome/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "rna-fusion", |
| "name": "RNA Fusion Detection from RNA-seq", |
| "description": "This task analyzes paired-end RNA-seq data to detect gene fusion events and quantify gene expression. A chr22 reference genome and gene annotation are provided.", |
| "task_prompt": "Detect gene fusions from RNA-seq data using chimeric read analysis, and quantify gene expression and transcript assembly. Reference genome and annotation in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_reads,185012\nuniquely_mapped,1986\nmapping_pct,1.07\nchimeric_reads,745\nsplice_junctions,166\nfusions_detected,0\nhigh_confidence_fusions,0\nmedium_confidence_fusions,0\ngenes_expressed,26\nassembled_transcripts,20588</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/rna-fusion/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/rna-fusion/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/rna-fusion/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "spatial-transcriptomics", |
| "name": "Spatial Transcriptomics: Visium FFPE Brain Cancer Analysis", |
| "description": "This task analyzes Visium spatial transcriptomics data from a human FFPE brain cancer sample processed with the CytAssist platform. The data includes raw sequencing reads (FASTQ), pre-aligned reads (BAM), and quantified spot-level expression matrices with spatial coordinates and tissue images from a standard processing pipeline. The input consists of a chr22-subset reference to keep the task lightweight. The goal is to assess read quality, alignment statistics, load the spatial expression data, perform quality control, normalize, cluster spots both spatially and non-spatially, identify spatially variable genes and marker genes per cluster, compute neighborhood enrichment and co-occurrence statistics, and produce a summary report.", |
| "task_prompt": "Analyze Visium spatial transcriptomics data from an FFPE brain cancer sample. The data/ directory contains: raw FASTQ reads (R1/R2 for two lanes), pre-processed outputs in data/outs/ (count matrices in raw_feature_bc_matrix/, spatial coordinates and tissue images in spatial/, and a pre-aligned BAM file). The reference/ directory contains a chr22 genome reference. Assess read quality, compute alignment statistics from the provided BAM, load the spatial expression data with coordinates, filter low-quality spots, normalize, identify highly variable genes, reduce dimensions, cluster spots, compute spatial autocorrelation (identify spatially variable genes), perform neighborhood enrichment analysis between clusters, compute co-occurrence statistics, find cluster marker genes, and aggregate QC reports. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>metric,value\ntotal_reads,20000\nmapped_reads,19759\nmapping_rate_pct,98.8\nduplicate_reads,2376\ntotal_spots,11397\nin_tissue_spots,10881\nspots_with_counts,7089\ngenes_detected,7709\nmedian_genes_per_spot,2\nmedian_counts_per_spot,2\nnum_hvg,100\nnum_pcs,15\npca_variance_ratio_pc1,0.0359\nnum_spatial_clusters,12\ncluster_0_size,6785\ncluster_0_marker,ENSG00000120885\ntop_spatial_gene,ENSG00000014824\ntop_morans_I,-0.0003\nspatially_variable_genes,0\nnhood_enrichment_max_z,3.9837\nco_occurrence_max,7.9322\nspatial_extent_x,2439.0\nspatial_extent_y,2399.0</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/spatial-transcriptomics/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/spatial-transcriptomics/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/spatial-transcriptomics/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "taxonomic-profiling", |
| "name": "Multi-classifier Taxonomic Profiling of Metagenomic Reads", |
| "description": "This task performs taxonomic classification of metagenomic shotgun sequencing data using multiple independent classification approaches. Two paired-end Illumina samples from an ancient metagenome study (Maixner et al. 2021) are provided. Pre-built classification databases for three distinct algorithmic approaches (k-mer matching, Burrows-Wheeler transform alignment, and protein-level translation) are provided in the reference/ directory. The goal is to run all three classifiers independently, merge their results to identify consensus taxa, and compute diversity metrics.", |
| "task_prompt": "Classify metagenomic reads from two paired-end samples using three independent taxonomic classifiers provided in reference/. Each subdirectory in reference/ contains a pre-built database for a different classification approach. Quality-filter the reads first, then run each classifier independently on both samples, merge the per-sample results to identify taxa detected by multiple classifiers (consensus), and calculate alpha diversity metrics from the consensus profiles. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>metric,value\ntotal_samples,2\ntotal_input_reads,2339208\ntotal_qc_reads,1882572\nclassifiers_used,3\nclassified_reads_classifier1,216\nclassified_reads_classifier2,893\nclassified_reads_classifier3,0\nunique_species_classifier1,2\nunique_species_classifier2,4\nunique_species_classifier3,0\ntotal_unique_taxa,5\nconsensus_taxa,2\ntop_taxon,Homo sapiens\ntop_taxon_abundance,33.35\nmean_shannon,0.3164\nmean_simpson,0.1974\nmean_richness,4.5</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/taxonomic-profiling/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/taxonomic-profiling/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/taxonomic-profiling/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "lcms-metabolomics", |
| "name": "LC-MS Untargeted Metabolomics: Urine Feature Discovery", |
| "description": "This task performs untargeted metabolomics analysis on LC-MS data from human urine samples acquired on an LTQ-Orbitrap in negative ionization mode. Twelve mzML files represent three sample types: 6 study samples (3 male, 3 female), 3 pooled quality control (QC) samples, and 3 blank controls. A sample metadata file provides sample type, gender, age, BMI, injection order, and batch information. The goal is to detect chromatographic peaks, align retention times, group features, apply QC-based and blank-subtraction filters, annotate isotopes and adducts, perform statistical comparison between groups, and produce a comprehensive quality and discovery report.", |
| "task_prompt": "Analyze LC-MS untargeted metabolomics data from human urine. The data/ directory contains 12 mzML files: 6 study samples (HU_neg_*), 3 QC pool samples (QC1_*), and 3 blank controls (Blanc*), plus a sampleMetadata.tsv with sample type, gender, age, BMI, injection order, and batch. Detect chromatographic peaks across all samples, align retention times, group features, fill missing values, then apply quality filters: remove features dominated by blank signal and features with high coefficient of variation in QC samples. Annotate isotope patterns and adducts. Compare feature intensities between male and female subjects. The output should be a CSV file with columns: 'metric','value'.\n<example>metric,value\nnum_samples,6\nnum_qc_pools,3\nnum_blanks,3\ntotal_scans,21878\nmz_range_min,50.0321\nmz_range_max,691.3176\nrt_range_min_sec,5.29\nrt_range_max_sec,1208.69\npeaks_detected,57325\nfeatures_grouped,4951\nfeatures_after_fill,4951\nfeatures_blank_removed,684\nfeatures_qc_removed,3265\nfeatures_final,1580\nisotope_annotations,1135\nadduct_annotations,1937\nputative_identifications,9\nmean_qc_cv_pct,21.75\nmedian_qc_cv_pct,23.86\nsignificant_features_fdr05,0\nsignificant_features_fdr25,0\nsignificant_annotated_features,0\ntop_feature_mz,355.1949\ntop_feature_pvalue,3.33e-04\ntop_feature_fdr,3.51e-01\ntop_feature_log2fc,2.7334\ntop_feature_annotation,unknown</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/lcms-metabolomics/data.tar.gz" |
| } |
| ], |
| "reference_data": [], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/lcms-metabolomics/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "somatic-variant-calling", |
| "name": "Somatic Variant Calling: Tumor-Normal Paired Analysis", |
| "description": "This task performs somatic variant calling from paired tumor-normal whole-genome sequencing data. The data consists of multi-lane paired-end Illumina reads from a tumor sample (6 lanes) and a matched normal sample (5 lanes), aligned to a small GRCh37 reference (chromosomes 1-3, 8, 11, X subsets). Reference files including a genome FASTA, known variant sites (dbSNP, known indels, population frequencies), and target intervals are provided in reference/. The goal is to preprocess both samples through alignment and quality recalibration, call somatic variants using multiple independent approaches, filter the results, compute coverage statistics, and annotate variants.", |
| "task_prompt": "Call somatic variants from paired tumor-normal sequencing data. Multi-lane paired-end FASTQ files are in data/tumor/ and data/normal/. A reference genome with indexes, known variant databases (dbSNP, known indels, population frequencies), and target intervals are in reference/. For each sample: quality-filter reads, align to reference with read groups, mark duplicates, and recalibrate base quality scores using the known sites. Then call somatic variants using at least three independent approaches, filter the calls, compute coverage statistics for both samples, and annotate variants with the known variant database. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>metric,value\ntumor_input_reads,5544\ntumor_qc_reads,4654\nnormal_input_reads,5522\nnormal_qc_reads,4602\ntumor_duplication_rate,0.0615\nnormal_duplication_rate,0.0356\ntumor_mean_coverage,2.14\nnormal_mean_coverage,2.19\ncaller1_raw_variants,12\ncaller1_pass_variants,5\ncaller2_raw_variants,171\ncaller3_raw_variants,35\ncallers_used,3\nannotated_variants_caller1,5\nannotated_variants_caller2,180\nannotated_variants_caller3,35\nestimated_contamination,0.0</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/somatic-variant-calling/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/somatic-variant-calling/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/somatic-variant-calling/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "amr-bgc-screening", |
| "name": "Antimicrobial Resistance and Biosynthetic Gene Cluster Screening", |
| "description": "This task screens an E. coli O157:H7 genome for antimicrobial resistance genes, biosynthetic gene clusters, and virulence factors using multiple detection methods.", |
| "task_prompt": "Screen a bacterial genome for antimicrobial resistance genes using nucleotide and protein-based methods, detect biosynthetic gene clusters, and identify virulence factors. The assembled genome is in data/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ngenome_length,5594605\npredicted_genes,5388\namr_genes_nucleotide,50\namr_genes_protein,0\nstress_response_genes,0\nvirulence_amr_genes,0\nbiosynthetic_gene_clusters,0\nvirulence_factors,127</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/amr-bgc-screening/data.tar.gz" |
| } |
| ], |
| "reference_data": [], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/amr-bgc-screening/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "variant-trio", |
| "name": "Variant Annotation Trio: Clinical Interpretation of Ashkenazi Family", |
| "description": "This task performs clinical variant annotation and inheritance analysis on a trio (proband, father, mother) from the GIAB Ashkenazi family. Three VCF files contain benchmark variant calls for chromosome 22 of GRCh38 from HG002 (son/proband), HG003 (father), and HG004 (mother). A chr22 reference FASTA, a clinical variant significance database, and a PED pedigree file are provided. The goal is to merge trio variants, filter by variant type (SNPs and indels separately), annotate with clinical databases, determine inheritance patterns, and identify clinically relevant variants.", |
| "task_prompt": "Perform clinical variant annotation and inheritance analysis on a trio of VCF files. The data/ directory contains three VCF files with variant calls for a family trio (son as proband, plus father and mother) on chromosome 22. The reference/ directory contains a chr22 FASTA reference, a clinical variant significance database (VCF format), and a pedigree (PED) file defining family relationships. Merge the trio VCFs, separate SNPs from indels, apply quality filters to each type independently, annotate variants with clinical significance from the database, determine inheritance patterns using the pedigree, and classify variants. The output should be a CSV file with columns: 'metric','value'.\n<example>metric,value\ntotal_variants,69655\nsnps,58016\ninsertions,5836\ndeletions,5803\ntitv_ratio,2.40\nhet_hg002,32826\nhom_alt_hg002,17341\nhet_hg003,33322\nhom_alt_hg003,17248\nhet_hg004,30304\nhom_alt_hg004,16624\nautosomal_dominant,1132\nautosomal_dominant_denovo,1132\nautosomal_recessive_hom,21578\nautosomal_recessive_comp,0\nx_linked_recessive,0\nx_linked_dominant,0\nclinvar_pathogenic,0\nclinvar_likely_pathogenic,0\nclinvar_uncertain,17\nclinvar_benign,1181\nclinvar_likely_benign,28\ntotal_clinvar_annotated,1248\ntop_pathogenic_gene,none\ntop_pathogenic_pos,none\ntop_pathogenic_change,none\nmendelian_errors,0\nde_novo_candidates,1132\ncompound_het_candidates,0</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/variant-trio/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/variant-trio/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/variant-trio/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "clinical-metaproteomics", |
| "name": "Clinical Metaproteomics: Multi-Engine Marine Microbiome Analysis", |
| "description": "This task analyzes metaproteomics data from a marine microbiome sample collected at the Bering Strait chlorophyll maximum layer. Three replicate mass spectrometry runs (MGF format) are provided along with a protein sequence database containing trimmed metapeptides and common contaminants. A Gene Ontology annotation file is also included. The goal is to identify peptides and proteins using multiple search engine approaches, control false discovery rate, merge results across engines, perform functional annotation using GO terms, and summarize identifications across replicates.", |
| "task_prompt": "Analyze metaproteomics data from three replicate mass spectrometry runs of a marine microbiome sample. The data/ directory contains three MGF spectrum files (sample1.mgf, sample2.mgf, sample3.mgf). The reference/ directory contains a protein sequence database (proteins.fasta with metapeptides and contaminants) and Gene Ontology annotations (go_terms.tsv). Create a target-decoy database for false discovery estimation. Search spectra against the database using at least two independent search engines, merge their results, estimate and filter at 1% FDR, identify unique peptides and proteins, map GO functional annotations, and compare identifications across the three replicates. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>metric,value\ntotal_spectra,1813\ntotal_psms_fdr1pct,2996\nunique_peptides,315\nunique_proteins,287\nsample1_psms,1003\nsample2_psms,937\nsample3_psms,1056\nsearch_engines_used,2\nfdr_threshold,0.01\ngo_terms_total,530\ngo_biological_process,286\ngo_molecular_function,181\ngo_cellular_component,63\nproteins_per_sample_avg,221</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/clinical-metaproteomics/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/clinical-metaproteomics/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/clinical-metaproteomics/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "mhc-immunopeptidomics", |
| "name": "MHC Immunopeptidomics: Peptide Identification and Quantification", |
| "description": "This task analyzes mass spectrometry immunopeptidomics data to identify and quantify MHC class I-presented peptides from HepG2 cell line replicates. Two mzML files containing tandem mass spectrometry data are provided along with a human protein reference database in reference/. MHC-presented peptides have a characteristic length distribution of 8-12 amino acids with no enzymatic cleavage specificity, requiring specialized search strategies. The goal is to process raw spectra, search against the protein database without enzyme specificity, statistically validate identifications, align retention times across runs, and quantify peptide features.", |
| "task_prompt": "Identify and quantify MHC-presented peptides from two replicate mass spectrometry experiments. Two mzML spectral files are in data/ and a human protein reference database is in reference/. Process the spectra (centroid if needed), search against the protein database using unspecific cleavage to find peptides of 8-12 amino acids, add decoy sequences for statistical validation, rescore peptide-spectrum matches, filter at 5% FDR, align identifications across runs, and quantify peptide features. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>metric,value\ntotal_samples,2\ntotal_spectra,5200\ntotal_psms_searched,2277\ntotal_psms_filtered,887\nunique_peptides_raw,1292\nunique_peptides_filtered,288\nmodal_peptide_length,9\npct_peptides_8_12mer,83.0\ntotal_quantified_features,51\nsample1_spectra,2600\nsample1_psms,441\nsample2_spectra,2600\nsample2_psms,446</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mhc-immunopeptidomics/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mhc-immunopeptidomics/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mhc-immunopeptidomics/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "riboseq", |
| "name": "Ribosome Profiling Translation Analysis", |
| "description": "This task analyzes ribosome profiling (Ribo-seq) data from yeast to quantify active translation. Single-end reads, a reference genome, and gene annotation are provided.", |
| "task_prompt": "Analyze ribosome profiling sequencing data to quantify translational activity. Remove rRNA contamination, align ribosome-protected fragments, determine the characteristic RPF length profile, and quantify gene-level translation. Reads in data/, genome and annotation in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_reads,500000\npassed_length_filter,418491\nrrna_pct,0.0\nnon_rrna_reads,418491\nuniquely_mapped,150640\nmapping_pct,36.00\npeak_rpf_length,28\ngenes_with_reads,4652</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/riboseq/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/riboseq/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/riboseq/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "neoantigen-prediction", |
| "name": "Neoantigen Prediction: Tumor-Normal Somatic Analysis", |
| "description": "This task performs tumor-normal paired somatic variant analysis on human exome sequencing data. Paired-end FASTQ files are provided for both a tumor and matched normal sample, along with a chr22 reference genome. The goal is to align reads, mark duplicates, assess coverage, call somatic variants from the tumor-normal pair, call germline variants from the normal sample, filter variants, and produce a summary report with alignment QC, coverage, variant counts, and mutation types.", |
| "task_prompt": "Perform somatic and germline variant analysis from paired tumor-normal sequencing data. The data/ directory contains paired-end FASTQ files for a tumor sample (tumor_R1.fastq.gz, tumor_R2.fastq.gz) and a matched normal sample (normal_R1.fastq.gz, normal_R2.fastq.gz). The reference/ directory contains a chr22 reference genome (genome.fa) with index files. Align both samples to the reference, mark duplicates, compute coverage statistics, call somatic variants using the tumor-normal pair, call germline variants from the normal sample, filter variants, and report alignment quality, coverage, duplicate rates, and variant counts. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>metric,value\ntumor_input_reads,266736\nnormal_input_reads,267721\ntumor_mapped_reads,3882\nnormal_mapped_reads,3610\ntumor_mean_coverage,9.4\nnormal_mean_coverage,9.4\ntumor_duplicate_pct,38.51\nnormal_duplicate_pct,36.98\nsomatic_variants_raw,0\nsomatic_variants_pass,0\ngermline_variants,55\ngermline_snps,54\ngermline_indels,1\nvariant_calling_mode,paired_tumor_normal</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/neoantigen-prediction/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/neoantigen-prediction/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/neoantigen-prediction/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "somatic-germline-dual", |
| "name": "Somatic+Germline Dual Analysis: Hereditary Cancer Variants", |
| "description": "This task performs dual somatic and germline variant analysis from paired tumor-normal exome sequencing data. Paired-end FASTQ files are provided for both a tumor and matched normal sample along with a chr22 reference genome. The goal is to independently call somatic variants (from the tumor-normal pair) and germline variants (from each sample separately), then intersect the variant sets to identify positions that are shared or unique to each call set. This dual-analysis approach reveals both inherited variants and acquired somatic mutations, with the intersection distinguishing germline contamination from true somatic events.", |
| "task_prompt": "Perform dual somatic and germline variant analysis from paired tumor-normal sequencing data. The data/ directory contains paired-end FASTQs for a tumor sample (tumor_R1/R2.fastq.gz) and matched normal sample (normal_R1/R2.fastq.gz). The reference/ directory contains a chr22 reference genome with index files. Align both samples, mark duplicates, compute coverage. Call somatic variants using the tumor-normal pair. Call germline variants from both samples independently. Intersect the germline call sets to find shared and sample-specific variants. Report alignment QC, coverage, duplicate rates, variant counts for somatic and germline calls, and the intersection statistics. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>metric,value\ntumor_input_reads,266736\nnormal_input_reads,267721\ntumor_mapped_reads,3882\nnormal_mapped_reads,3610\ntumor_mean_coverage,9.4\nnormal_mean_coverage,9.4\ntumor_duplicate_pct,38.51\nnormal_duplicate_pct,36.98\nsomatic_variants,0\ngermline_variants_normal,55\ngermline_variants_tumor,70\ngermline_snps,54\ngermline_indels,1\nshared_germline_positions,18\nnormal_only_germline,37\ntumor_only_germline,52</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/somatic-germline-dual/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/somatic-germline-dual/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/somatic-germline-dual/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "hicar-chromatin", |
| "name": "HiCAR Chromatin Interaction: Proximity Ligation and Accessibility", |
| "description": "This task analyzes HiCAR (Hi-C with Accessibility Readout) data, a multi-omic assay that simultaneously captures chromatin interactions and chromatin accessibility from the same sequencing library. Four paired-end samples (2 KD replicates, 2 WT replicates) targeting chromosome 22 are provided. In HiCAR, R1 reads represent proximity ligation junctions (like Hi-C) while R2 reads capture local accessibility (like ATAC-seq). A chr22 reference genome and gene annotations are provided in reference/. The goal is to process both read types, extract proximity ligation pairs and accessibility signals, build contact matrices, call accessibility peaks, and report interaction and accessibility statistics.", |
| "task_prompt": "Analyze HiCAR multi-omic chromatin data from four paired-end samples (two conditions: KD and WT, two replicates each). In HiCAR, R1 captures proximity ligation contacts and R2 captures chromatin accessibility. A chr22 reference genome and annotations are in reference/. Trim adapters, align reads, parse proximity ligation pairs from the alignments, call accessibility peaks from the R2 reads (convert to single-end format), build contact matrices at 10kb resolution, and compute statistics for both contact and accessibility data. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>metric,value\ntotal_samples,4\nconditions,2\ntotal_read_pairs,255490\ntrimmed_read_pairs,255490\ntotal_valid_pairs,187129\ncis_contacts,187129\ntrans_contacts,0\ncis_contact_pct,100.0\ntotal_accessibility_peaks,2135\nkd_rep1_peaks,289\nkd_rep2_peaks,93\nwt_rep1_peaks,824\nwt_rep2_peaks,929\ncontact_matrices_created,4\nmatrix_resolution,10000</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/hicar-chromatin/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/hicar-chromatin/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/hicar-chromatin/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "radseq-popgen", |
| "name": "RADseq Population Genetics: Stickleback Freshwater-Marine Divergence", |
| "description": "This task analyzes RADseq (restriction-site associated DNA sequencing) data from a study of threespine stickleback (Gasterosteus aculeatus). Sixteen individuals from two groups \u2014 8 from Bear Paw Lake (freshwater) and 8 from Rabbit Slough (oceanic/anadromous) \u2014 are multiplexed in a single FASTQ file with inline barcodes. The data uses SbfI restriction enzyme, single-end 36bp reads. The goal is to perform de novo RADseq analysis: demultiplex reads, build per-sample loci, construct a catalog of loci across all samples, genotype individuals, and compute divergence statistics. Data from Hohenlohe et al. 2010 (SRA SRR034310).", |
| "task_prompt": "Analyze RADseq data from 16 threespine stickleback individuals (8 freshwater Bear Paw, 8 oceanic Rabbit Slough) multiplexed in a single FASTQ file. A barcode file and population map are provided in reference/. Demultiplex the reads using the barcodes, build per-sample loci de novo, create a shared catalog, genotype all individuals, and compute genetic divergence metrics including allele frequency differentiation between the two groups. Export variant calls and structure data. The output should be a CSV file with columns: 'metric','value'.\n<example>metric,value\ntotal_reads,8895289\nretained_reads,7799884\nsamples_demultiplexed,16\ncatalog_loci,45492\nloci_passing_filters,8741\ntotal_sites,279712\nvariant_sites,2470\nsamples_in_vcf,16\nstructure_loci,2470\nmean_fst,0.1219\npi_bear_paw,0.14806\npi_rabbit_slough,0.22022\nnum_groups,2</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/radseq-popgen/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/radseq-popgen/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/radseq-popgen/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "mag-recovery", |
| "name": "MAG Recovery: Metagenome-Assembled Genomes from Environmental Sample", |
| "description": "This task recovers metagenome-assembled genomes (MAGs) from a coffee fermentation metagenome. Paired-end Illumina MiSeq reads (300bp) from a microbial community are provided. The goal is to quality-filter reads, assemble the metagenome, map reads back to contigs for coverage estimation, apply multiple independent binning algorithms, merge bins using a consensus approach, predict genes, and assess bin quality. This is a standard environmental metagenomics workflow for recovering draft genomes from complex communities.", |
| "task_prompt": "Recover metagenome-assembled genomes (MAGs) from paired-end metagenomic reads. The data/ directory contains paired FASTQ files (reads_R1.fastq.gz and reads_R2.fastq.gz) from a microbial community. Quality-filter the reads, assemble the metagenome, map reads back to assembled contigs to calculate coverage depth, then apply at least two independent binning methods to group contigs into draft genomes. Refine bins by comparing results across methods. Predict genes on assembled contigs. Report assembly quality, binning results, and per-bin statistics. The output should be a CSV file with columns: 'metric','value'.\n<example>metric,value\ntotal_reads_before,2052218\ntotal_reads_after,2027938\nq30_rate_before,84.6\nq30_rate_after,85.06\ntotal_contigs,7187\ntotal_assembly_length,21672139\nlargest_contig,85618\nassembly_gc_pct,35.18\nassembly_n50,3172\nmapped_reads,570589\nmapping_pct,28.14\npredicted_genes,19285\nmetabat2_bins,4\nmaxbin2_bins,5\nrefined_bins,1\nlargest_bin_name,bin.4\nlargest_bin_length,1645295\nlargest_bin_contigs,397\nlargest_bin_n50,4922\nlargest_bin_gc_pct,39.26\ntotal_binned_length,1645295\nmean_bin_size,1645295\nmean_bin_gc_pct,39.26\ndastool_score_bin.4,0</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mag-recovery/data.tar.gz" |
| } |
| ], |
| "reference_data": [], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/mag-recovery/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "viral-phylodynamics", |
| "name": "Viral Phylodynamics (Molecular Clock Analysis)", |
| "description": "Molecular clock analysis estimates evolutionary rates, divergence times, and epidemic dynamics from time-stamped viral sequences using Bayesian and maximum-likelihood methods. This task involves multiple sequence alignment, phylogenetic model selection, temporal signal assessment, molecular clock estimation, geographic migration reconstruction (phylogeography), effective population size inference (skyline analysis), and ancestral sequence reconstruction. Key challenges include handling ambiguous sampling dates, detecting recombination, choosing appropriate clock models (strict vs relaxed), and correctly rooting the phylogeny for time calibration.", |
| "task_prompt": "Perform a molecular clock and phylodynamic analysis on viral sequences. Sequences are in data/sequences.fasta and sampling metadata (dates, locations) in data/metadata.tsv. A reference genome is in reference/reference.fasta. Align the sequences, select the best-fit substitution model, build a maximum-likelihood phylogeny with bootstrap support, perform temporal signal assessment, estimate evolutionary rates and the time to most recent common ancestor (TMRCA), reconstruct geographic migration history, estimate effective population size over time (skyline), reconstruct ancestral sequences, identify geographic transmission clusters, and export the results for interactive visualization. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>\nmetric,value\nnum_sequences,34\nalignment_length,10611\ngap_fraction,0.0218\nbest_model,TN+F+I\nmean_bootstrap,83.8\nhigh_support_nodes,23\ntotal_internal_nodes,31\nclock_rate,1.162e-03\nclock_r_squared,0.80\nnum_dated_tips,34\ntmrca,2012.38\nmugration_model,fitted\nnum_migration_countries,15\nskyline_intervals,10\nmax_effective_pop_size,359.7\nmin_effective_pop_size,46.7\nancestral_sequences,61\nnum_geographic_clusters,7\nlargest_cluster_size,5\nvisualization_export,success\ndate_range_start,2013.87\ndate_range_end,2016.96\nnum_countries,15\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/viral-phylodynamics/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/viral-phylodynamics/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/viral-phylodynamics/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "edna-metabarcoding", |
| "name": "eDNA Aquatic Metabarcoding Biodiversity Assessment", |
| "description": "Environmental DNA (eDNA) metabarcoding enables non-invasive biodiversity monitoring of aquatic ecosystems by analyzing DNA shed by organisms into their environment. This task involves processing amplicon sequencing data targeting the 12S mitochondrial rRNA gene of fish. The pipeline requires primer removal, read merging, quality filtering, sequence dereplication, multiple clustering approaches (OTU and ASV-based), chimera detection, multi-method taxonomic assignment with consensus resolution, and community ecology analysis including alpha diversity, beta diversity, and detection probability estimation.", |
| "task_prompt": "Analyze environmental DNA (eDNA) metabarcoding data from aquatic samples to assess fish biodiversity. Six paired-end amplicon sequencing samples from coral reef water samples are provided in data/. A 12S rRNA reference database for fish species is in reference/. The amplicon data was generated with standard fish eDNA primers (MiFish-U). Remove primers, merge paired reads, quality-filter, dereplicate, cluster sequences using multiple methods (OTU clustering at 97%, network-based clustering, and denoising), remove chimeric sequences, assign taxonomy using multiple database search approaches with lowest common ancestor consensus, compute community ecology metrics (alpha diversity, beta diversity, species detection rates across samples), and generate a comprehensive biodiversity report.\nThe output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>\nmetric,value\ntotal_raw_reads,1407956\ntotal_merged_reads,699972\nmerge_rate,99.43\nunique_sequences,9199\nclusters_otu97,309\nclusters_swarm,371\nclusters_denoised,389\nclean_sequence_count,385\nchimera_count,4\nassigned_sequences,299\nunassigned_sequences,39\nspecies_count,118\ngenus_count,126\nfamily_count,54\norder_count,32\nmean_shannon_diversity,3.7519\nmin_species_richness,114\nmax_species_richness,187\nmean_beta_diversity,0.6965\nmin_beta_diversity,0.533\nmax_beta_diversity,0.8741\nmean_detection_rate,0.3418\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/edna-metabarcoding/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/edna-metabarcoding/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/edna-metabarcoding/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "metatranscriptomics", |
| "name": "Metatranscriptomics: Active Microbial Community Profiling", |
| "description": "This task analyzes metatranscriptomic data from a gut microbiome sample (HMP2, SRR6820516). Paired-end Illumina reads (150bp, ~260K pairs) capture both ribosomal RNA and messenger RNA from the active microbial community. The goal is to separate rRNA from mRNA, assemble transcripts de novo, predict and annotate genes, quantify expression, and compute diversity metrics. This workflow requires integrating quality control, rRNA removal, de novo assembly, gene prediction, functional annotation, read mapping, and ecological statistics.", |
| "task_prompt": "Analyze metatranscriptomic paired-end reads to profile active microbial community composition and functional gene expression. The data/ directory contains paired FASTQ files (reads_R1.fastq.gz and reads_R2.fastq.gz) from a gut microbiome. The reference/ directory contains rRNA sequence databases (in rrna_db/) and a protein database (uniprot_sprot.fasta.gz) for functional annotation. Quality-filter the reads, separate ribosomal RNA reads from messenger RNA reads, assemble the non-rRNA fraction into transcripts, predict genes, annotate them functionally against the protein database, quantify gene expression by mapping reads back to the assembly, and compute community diversity metrics from the taxonomic composition of annotated genes. The output should be a CSV file with columns: 'metric','value'.\n<example>metric,value\ntotal_reads_before,521108\ntotal_reads_after,497272\nq30_rate_before,91.49\nq30_rate_after,93.44\nrrna_reads,105662\nrrna_pct,21.17\nnonrrna_reads,393396\ntotal_contigs,2468\nassembly_length,1948243\nlargest_contig,12392\nassembly_gc_pct,41.77\nassembly_n50,982\npredicted_genes,3538\nannotated_genes,2308\nmapped_reads,353506\nmapping_pct,89.86\nexpressed_genes,3337\nhighly_expressed_genes,100\nshannon_diversity,2.1974\nsimpson_diversity,0.8112\nobserved_richness,20\nchao1_estimate,20.0\ntop_organism,Acetivibrio thermocellus (strain ATCC 27405 / DSM 1237 / JCM 9322 / NBRC 103400 / NCIMB 10682 / NRRL B-4536 / VPI 7372)\ntop_organism_pct,30.33\ntop_function,Elongation factor G\nfunctional_categories,9</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/metatranscriptomics/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/metatranscriptomics/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/metatranscriptomics/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "nascent-transcription", |
| "name": "Nascent Transcription: GRO-seq Polymerase Activity Profiling", |
| "description": "Nascent transcription profiling captures RNA polymerase activity genome-wide, revealing active transcription start sites, enhancers, and gene body transcription in real-time. This task involves analyzing GRO-seq data from two human cell types (CD4+ T-cells and Jurkat) mapped to chromosome 21. The analysis requires ribosomal RNA depletion before genome alignment, strand-specific signal separation, peak detection for transcription initiation sites and transcribed regions, TSS profiling, and genomic annotation of detected features.", |
| "task_prompt": "Analyze GRO-seq nascent transcription data from two conditions (CD4+ T-cells and Jurkat). For each condition, merge the replicate FASTQ files, perform read quality control, remove ribosomal RNA contamination, align to the chromosome 21 reference genome (single-end, no intron spanning), deduplicate, and compute alignment statistics. For the CD4+ condition, generate strand-specific genome coverage tracks (plus and minus strand separately), convert to signal tracks, detect transcription start site peaks and transcribed region peaks, create a TSS-centered signal profile heatmap, annotate peaks with nearby genes, and classify transcribed regions as genic or intergenic. The output should be a CSV file with columns: 'metric','value'.\n<example>\nmetric,value\ncd4_raw_reads,99020\ncd4_clean_reads,98318\ncd4_q30_rate,95.81\ncd4_rrna_rate,21.78\ncd4_non_rrna_reads,76901\ncd4_uniquely_mapped_reads,33350\ncd4_unique_mapping_rate,43.37\ncd4_dedup_reads,30797\njurkat_raw_reads,99020\njurkat_clean_reads,97944\njurkat_q30_rate,95.94\njurkat_rrna_rate,47.09\njurkat_non_rrna_reads,51827\njurkat_uniquely_mapped_reads,21664\njurkat_unique_mapping_rate,41.80\njurkat_dedup_reads,18379\ntss_peaks,583\ntranscription_unit_peaks,5\ngenic_peaks,3\nintergenic_peaks,2\nplus_strand_coverage_regions,17710\nminus_strand_coverage_regions,19556\nannotated_peaks,583\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/nascent-transcription/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/nascent-transcription/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/nascent-transcription/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "circrna-discovery", |
| "name": "Circular RNA Discovery: C. elegans Wild-type vs fust-1 Mutant", |
| "description": "This task analyzes paired-end RNA-seq data to discover circular RNAs (circRNAs) in C. elegans. Four samples from two conditions (wild-type N2 and fust-1 mutant, 2 replicates each) are provided as paired-end FASTQ files, subset to chromosome I only. A reference genome (chrI.fa), gene annotation (chrI.gtf), and gene prediction file (chrI.txt) are included. The goal is to detect back-splice junctions characteristic of circular RNA molecules, quantify them across samples, and characterize the detected circRNAs. Data from nf-core/circrna test dataset (SRA SRX9313375-SRX9313380).", |
| "task_prompt": "Detect circular RNAs from paired-end RNA-seq data of C. elegans chromosome I. Four paired-end FASTQ samples (2 wild-type N2, 2 fust-1 mutant) are in data/. A chromosome I reference genome, gene annotation (GTF), and gene prediction file are in reference/. Trim adapters, align reads to detect chimeric back-splice junctions, identify circular RNA candidates, quantify back-splice junction read counts, and characterize the detected circular RNAs by length and genomic distribution. The output should be a CSV file with columns: 'metric','value'.\n<example>metric,value\nsamples_analyzed,4\ntotal_circular_rnas,134\nshared_by_multiple_tools,0\ntotal_back_splice_reads,765\nmean_circrna_length,1852\nmin_circrna_length,192\nmax_circrna_length,15586\ndetection_tools_used,2\nchromosomes_with_circrnas,1\nn2_samples,2\nfust1_samples,2</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/circrna-discovery/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/circrna-discovery/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/circrna-discovery/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "hic-3d-conformation", |
| "name": "Hi-C 3D Genome Conformation Analysis", |
| "description": "Analyze Hi-C chromatin interaction data to map 3D genome organization. Hi-C captures chromosome conformation by proximity ligation of cross-linked chromatin, producing paired-end reads that represent physical contacts between distant genomic loci. The analysis involves quality-filtering paired reads, aligning to a reference genome, extracting valid contact pairs, building interaction matrices at fixed resolution, and calling structural features including chromatin compartments (A/B), topological domain boundaries, and significant loop interactions. The input is paired-end sequencing reads from a yeast Hi-C experiment with a matching reference genome.", |
| "task_prompt": "Analyze Hi-C paired-end reads to characterize 3D genome organization. The input reads are in data/ and the reference genome is in reference/. Process the reads through quality trimming, alignment in Hi-C mode, contact pair extraction, deduplication, and contact matrix generation. From the balanced contact matrix, call chromatin compartments (A/B type), identify topological domain boundaries at multiple window sizes, detect chromatin loops, and identify statistically significant chromatin interactions. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>\nmetric,value\ntotal_read_pairs,500000\nreads_after_trim,488672\ntrim_rate_pct,2.27\nmapped_pairs,381220\nunique_pairs,381220\ncis_contacts,323321\ntrans_contacts,57899\ncis_ratio,0.8481\ncis_gt_1kb,24336\ncis_gt_10kb,8259\nduplicate_pairs,5036\ncompartment_switches,1024\ncompartment_a_bins,1080\ncompartment_b_bins,1030\nboundaries_25kb,284\nboundaries_50kb,166\ndetected_loops,0\nsignificant_interactions,30\nresolution_bp,5000\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/hic-3d-conformation/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/hic-3d-conformation/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/hic-3d-conformation/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "genome-scaffolding", |
| "name": "Long-Read Genome Scaffolding of Fragmented Assembly", |
| "description": "Genome scaffolding improves assembly contiguity by ordering and orienting contigs using long-range information from long reads. This task involves scaffolding a fragmented short-read E. coli K-12 assembly (~116 contigs) using Oxford Nanopore long reads through multiple approaches: reference-guided scaffolding, long-read linkage-based scaffolding, and minimizer-based scaffolding. The pipeline compares results, selects the best scaffolder, and performs comprehensive quality assessment including gene completeness, read mapping, gap analysis, and improvement quantification.", |
| "task_prompt": "Scaffold a fragmented short-read bacterial assembly using long reads. The data/ directory contains contigs.fasta (a fragmented short-read assembly of E. coli K-12, ~116 contigs) and long_reads.fastq.gz (Oxford Nanopore long reads, ~50x coverage). The reference/ directory contains the complete E. coli K-12 MG1655 reference genome for comparison. Apply at least three different scaffolding approaches (reference-guided, long-read linkage-based, and minimizer-based), compare their results, select the best by contiguity, then assess the scaffold quality including completeness, mapping rate, gap analysis, and improvement over the initial assembly.\nThe output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>\nmetric,value\ninitial_num_contigs,116\ninitial_total_length,4511552\ninitial_n50,117600\ninitial_l50,13\ninitial_largest_contig,268093\ninitial_num_misassemblies,5\ninitial_genome_fraction_pct,95.915\nscaffold_num_contigs,39\nscaffold_total_length,4519252\nscaffold_n50,4490464\nscaffold_l50,1\nscaffold_largest_contig,4490464\nscaffold_num_misassemblies,5\nscaffold_genome_fraction_pct,95.957\nscaffolder_ragtag_n50,4490464\nscaffolder_ragtag_sequences,39\nscaffolder_links_n50,178122\nscaffolder_links_sequences,80\nscaffolder_ntlink_n50,454565\nscaffolder_ntlink_sequences,82\nmapped_reads,115474\nmapping_rate_pct,97.34\ninitial_gap_count,0\ninitial_n_bases,0\nscaffold_gap_count,77\nscaffold_n_bases,7700\nn50_fold_improvement,38.2\nsequence_count_reduction_pct,66.4\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/genome-scaffolding/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/genome-scaffolding/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/genome-scaffolding/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "longread-rna-isoform", |
| "name": "Long-Read RNA Isoform Discovery from Direct RNA Sequencing", |
| "description": "This task discovers and characterizes transcript isoforms from long-read direct RNA sequencing. Data is from the SG-NEx project (A549 cell line, 80K reads, chr22). The workflow involves read QC, splice-aware alignment, transcript assembly, comparison with reference annotation for known/novel isoform classification, expression quantification, and coverage analysis.", |
| "task_prompt": "Discover and characterize transcript isoforms from long-read direct RNA sequencing data. The data/ directory contains a FASTQ file (reads.fastq.gz) with long single-molecule RNA reads from a human lung carcinoma cell line. The reference/ directory contains chromosome 22 sequence (chr22.fa) and gene annotation (chr22.gtf). Assess read quality, perform splice-aware long-read alignment, assemble transcripts, compare assembled transcripts against the reference annotation to classify known vs novel isoforms, quantify transcript expression, and compute coverage statistics. The output should be a CSV file with columns: 'metric','value'.\n<example>metric,value\ntotal_reads,80000.0\nmean_read_length,1030.7\nmedian_read_length,826.0\nread_n50,1350.0\nmean_quality,10.4\ntotal_bases,82457048.0\nmapped_reads,9406\nmapping_pct,11.76\nassembled_transcripts,340\nassembled_genes,297\nmean_transcript_length,12759\ntranscript_n50,36578\nmean_coverage,18.12\nbase_sensitivity,15.8\nbase_precision,95.4\nknown_isoforms,298\nnovel_isoforms,42\nmulti_exon_transcripts,219\nexact_match_transcripts,281\nnovel_junction_transcripts,5\nintergenic_transcripts,4\nexpressed_transcripts,340\nhighly_expressed_transcripts,97\nnovel_loci,30</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/longread-rna-isoform/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/longread-rna-isoform/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/longread-rna-isoform/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "circrna-detection", |
| "name": "Circular RNA Detection and Quantification", |
| "description": "Detect and quantify circular RNAs from RNA-seq data. Circular RNAs (circRNAs) are covalently closed single-stranded RNA molecules formed by back-splicing, where a downstream splice donor joins to an upstream splice acceptor. They are detected by identifying back-splice junction (BSJ) reads in RNA-seq data. Multiple detection methods should be used and their results merged by consensus to reduce false positives. Downstream analysis includes extracting circRNA sequences, computing back-splice junction read counts, and predicting potential microRNA binding sites on circRNA sequences. The input is paired-end RNA-seq reads from a C. elegans experiment with a matching reference genome, gene annotation, and mature miRNA sequences.", |
| "task_prompt": "Detect circular RNAs from paired-end RNA-seq data. The input reads are in data/ and the reference genome, gene annotation (GTF and refFlat format), and mature miRNA sequences are in reference/. Process reads through quality trimming, alignment with chimeric read detection, and run at least two independent back-splice junction detection methods. Merge calls by consensus (supported by >= 2 methods), extract circRNA sequences, count supporting BSJ reads, and predict miRNA binding sites on the circRNA sequences. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>\nmetric,value\ntotal_reads,444124\nadapter_pct,14.5\nunique_map_pct,98.59\nchimeric_reads,1004\ntotal_alignments,896302\ncaller1_count,47\ncaller2_count,92\ncaller3_count,86\nconsensus_count,33\ntotal_bsj_reads,149\nmax_bsj_reads,41\nmean_bsj_reads,4.52\nextracted_sequences,33\nmean_length,2336\nmiranda_targets,388\nrnahybrid_targets,19682\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/circrna-detection/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/circrna-detection/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/circrna-detection/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "pharmacogenomics", |
| "name": "CYP2D6 Pharmacogenomic Star Allele Calling", |
| "description": "Pharmacogenomics uses genetic variation to predict drug response. CYP2D6 is one of the most clinically important pharmacogenes, metabolizing ~25% of commonly prescribed drugs. This task involves calling CYP2D6 star alleles from WGS data (GIAB HG002), including copy number estimation from the CYP2D6/CYP2D7 depth ratio, variant-based star allele assignment, activity score calculation, metabolizer phenotype prediction, and drug interaction reporting using CPIC guidelines.", |
| "task_prompt": "Determine pharmacogenomic star alleles from whole-genome sequencing data, focusing on the CYP2D6 gene and its clinical implications for drug metabolism. Paired-end WGS reads covering chromosome 22 are in data/. The reference/ directory contains the chr22 reference genome (GRCh38), star allele variant definitions, and drug-gene interaction databases from CPIC. Perform read QC, align to the reference, mark duplicates, call variants in the CYP2D6/CYP2D7 region, estimate gene copy number from depth ratios, determine the star allele diplotype, calculate the metabolizer activity score, predict the metabolizer phenotype, and identify affected drug-gene interactions.\nThe output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>\nmetric,value\ntotal_reads_before_qc,1648406\ntotal_reads_after_qc,1566772\nq30_rate_pct,89.93\nmapped_reads,1569886\nmapping_rate_pct,99.99\nduplicate_rate_pct,0.85\ncyp2d6_mean_depth,46.6\ncyp2d7_mean_depth,46.5\ncyp2d6_cyp2d7_depth_ratio,1.003\ncopy_number_estimate,2\nvariants_in_cyp2d6_region,36\nsnps_in_cyp2d6_region,34\ndiplotype,*4/*17\ndefining_variants_found,3\nactivity_score,0.5\nmetabolizer_phenotype,Intermediate Metabolizer\ncpic_drug_pairs_total,80\ncpic_level_ab_drugs,22\nmosdepth_cyp2d6_coverage,45.9\nmosdepth_cyp2d7_coverage,43.4\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/pharmacogenomics/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/pharmacogenomics/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/pharmacogenomics/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "rna-editing-detection", |
| "name": "RNA Editing Detection: A-to-I Editing from Matched RNA/DNA Sequencing", |
| "description": "RNA editing is a post-transcriptional modification where adenosine is converted to inosine (A-to-I) by ADAR enzymes, which is read as guanosine by the sequencing machinery. This task detects A-to-I editing events from matched RNA-seq and DNA-seq (whole exome) data from a human cancer sample (chromosome 7 subset). The analysis requires independent preprocessing of RNA and DNA reads with different alignment strategies, multiple variant callers to identify RNA-specific variants, consensus filtering, strand-aware base change classification, known SNP removal, and repeat region annotation to distinguish true editing from genomic variation.", |
| "task_prompt": "Detect RNA editing events by comparing matched RNA-seq and DNA sequencing data. The data/ directory contains paired-end FASTQ files for RNA-seq (rna_R1.fastq.gz, rna_R2.fastq.gz) and whole exome DNA-seq (dna_R1.fastq.gz, dna_R2.fastq.gz) from the same patient. The reference/ directory contains the genome reference (genome.fa), gene annotation (genes.gtf), known variant sites (dbsnp.vcf.gz), population frequency data (gnomad.vcf.gz), and repeat region annotations (repeat_regions.bed). For RNA: quality-filter reads, perform splice-aware alignment with 2-pass mode, mark duplicates, split reads at N-cigar positions, and recalibrate base qualities. For DNA: quality-filter reads, align to the reference, mark duplicates, and recalibrate base qualities. Call variants using at least three independent approaches, then identify variants supported by multiple callers. Filter for A-to-G and T-to-C base changes (which represent editing on opposite strands), remove known genomic SNPs, and annotate overlap with repeat regions. The output should be a CSV file with columns: 'metric','value'.\n<example>\nmetric,value\nrna_raw_reads,500000\nrna_clean_reads,492236\nrna_q30_rate,94.3\nrna_unique_mapping_rate,3.62\nrna_mapped_reads,32593\nrna_duplication_rate,0.87\ndna_raw_reads,500000\ndna_clean_reads,491876\ndna_q30_rate,94.4\ndna_mapped_reads,84510\ndna_mapping_rate,17.11\ndna_duplication_rate,0.16\ncaller1_variants,0\ncaller2_variants,2512\ncaller3_variants,18\nintersection_variants,10\nag_tc_candidates,0\nnovel_variants,10\nrepeat_region_variants,4\ncandidate_editing_sites,0\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/rna-editing-detection/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/rna-editing-detection/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/rna-editing-detection/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "cnv-detection-wes", |
| "name": "CNV Detection from Whole-Exome Sequencing", |
| "description": "Detect copy number variants from whole-exome sequencing data using multiple CNV callers, merge calls, and assess concordance. Copy number variants (CNVs) are genomic segments that differ in copy number from the diploid reference. WES-based CNV detection requires paired tumor-normal analysis with specialized callers that account for the non-uniform capture efficiency across exome targets. The input includes paired-end reads for both tumor and normal samples, a reference genome (chr20), and an exome target intervals BED file.", |
| "task_prompt": "Detect copy number variants from paired tumor-normal whole-exome sequencing data. The input reads for tumor and normal samples are in data/ and the reference genome with exome target intervals are in reference/. Process both samples through quality trimming, alignment, duplicate marking, and coverage calculation. Run at least two independent CNV detection methods on the paired tumor-normal BAMs using the target intervals, merge their results, and compute summary statistics. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>\nmetric,value\ntumor_reads,4319\ntumor_reads_after_trim,4164\nnormal_reads,435\nnormal_reads_after_trim,408\ntumor_mean_coverage,0.0\nnormal_mean_coverage,0.0\ntumor_dup_pct,14.63\nnormal_dup_pct,0.0\ncnvkit_cnv_count,0\nfreec_cnv_count,0\ngatk_cnv_count,0\ntotal_callers_run,3\ntotal_merged_cnvs,0\ntarget_intervals,1000\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/cnv-detection-wes/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/cnv-detection-wes/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/cnv-detection-wes/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "haplotype-phasing", |
| "name": "Haplotype Phasing and Genotype Refinement", |
| "description": "Phase genotypes and refine variant calls from 1000 Genomes Phase 3 chr22 data. 50 study samples with ~10K biallelic SNPs, 200-sample reference panel.", |
| "task_prompt": "Phase genotypes and refine variant calls from a genotype study using a reference panel. The data/ directory contains a VCF file (study.vcf.gz) with 50 individuals genotyped at biallelic SNP positions on chromosome 22. The reference/ directory contains a reference panel VCF (ref_panel.vcf.gz) with 200 individuals, and genetic recombination maps for chr22. Normalize variants, perform quality control (filter by missingness, MAF, and HWE), find shared sites between study and reference, phase haplotypes using at least two statistical phasing methods, compare phasing concordance, perform genotype imputation using the reference panel, and assess post-imputation quality including LD pruning. The output should be a CSV file with columns: 'metric','value'.\n<example>metric,value\nstudy_samples,50\nstudy_variants_before_qc,10050\nstudy_variants_after_qc,9936\nvariants_removed_by_qc,114\nreference_samples,200\nreference_variants,21751\nshared_variants,8972\nshapeit4_phased_variants,8972\nbeagle_phased_variants,8972\nphasing_concordance_pct,88.29\nimputed_total_variants,8972\nimputed_filtered_variants,8972\nnewly_imputed_variants,0\nld_pruned_variants,2662\nld_removed_variants,6310</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/haplotype-phasing/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/haplotype-phasing/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/haplotype-phasing/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "dda-proteomics-simple", |
| "name": "DDA Proteomics: Single-Engine BSA Identification", |
| "description": "This task performs basic DDA mass spectrometry protein identification on a BSA standard using a single search engine. An mzML spectral file and a bovine protein database are provided.", |
| "task_prompt": "Identify proteins from mass spectrometry data by searching spectra against a protein database, controlling false discovery rate, and counting identified peptides and proteins. Spectral data in data/, protein database in reference/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntotal_spectra,1690\ndatabase_size_with_decoys,12104\nsearch_engine_hits,1604\npsms_after_fdr,16\nunique_peptides,16\nproteins_identified,9</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/dda-proteomics-simple/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/dda-proteomics-simple/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/dda-proteomics-simple/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "immune-repertoire", |
| "name": "Immune Repertoire Analysis (BCR-seq)", |
| "description": "Analyze B cell receptor sequencing data to characterize the adaptive immune repertoire. BCR-seq captures immunoglobulin heavy chain sequences including variable (V), diversity (D), and joining (J) gene segments, along with the complementarity-determining region 3 (CDR3) that determines antigen specificity. The analysis involves quality filtering, paired-end assembly, V(D)J gene annotation against germline reference databases, and downstream characterization including gene usage profiling, CDR3 length distribution, somatic hypermutation rates, clonal diversity estimation, and clone identification. The input includes paired-end BCR sequencing reads with UMI barcodes (I1 file) and germline reference sequences.", |
| "task_prompt": "Analyze B cell receptor sequencing data to characterize the immune repertoire. The input reads (R1, R2, and I1 UMI barcode files) are in data/ and germline V/D/J reference sequences are in reference/. Process reads through quality trimming, paired-end assembly, V(D)J gene annotation, and compute repertoire metrics including gene usage, CDR3 statistics, mutation rates, diversity indices, and clone estimates. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>\nmetric,value\ntotal_read_pairs,2500\nreads_after_trim,2272\nassembled_sequences,2292\ntotal_annotated,2272\nproductive_count,1795\nproductive_pct,79.01\nunique_v_genes,154\nunique_j_genes,21\ntop_v_gene,IGHV5-51*01\ntop_v_count,180\nmean_cdr3_length,45.1\nmedian_cdr3_length,42\nmean_mutation_pct,8.3\nv_gene_shannon_diversity,4.051\nestimated_clones,1256\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/immune-repertoire/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/immune-repertoire/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/immune-repertoire/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "germline-wes-gatk", |
| "name": "Germline WES Variant Calling: Clinical Exome Analysis", |
| "description": "Germline variant calling from whole-exome sequencing (WES) data following clinical best practices. This task involves processing paired-end WES reads from a human sample (chromosome 7 subset), including quality control, alignment to the reference genome, duplicate marking, base quality score recalibration using known variant sites, variant calling with multiple approaches, separate SNP and indel filtering, variant normalization, transition/transversion ratio calculation, coverage analysis, and clinical annotation against a pathogenicity database.", |
| "task_prompt": "Call germline variants from whole-exome sequencing data following clinical best practices. The data/ directory contains paired-end FASTQ files (sample_R1.fastq.gz, sample_R2.fastq.gz) from a human exome capture experiment. The reference/ directory contains the genome reference (genome.fa), gene annotation (genes.gtf), known variant sites for base quality recalibration (dbsnp.vcf.gz), a clinical pathogenicity database (clinvar.vcf.gz), and exome target regions (exome_targets.bed, exome_targets_padded.bed). Quality-filter the reads, align to the reference with proper read group information, sort, mark duplicates, and recalibrate base qualities using known sites. Compute coverage statistics. Call variants using both a genotype likelihood model and a pileup-based approach, normalize variants, then separate SNPs and indels for independent quality filtering. Merge filtered variants, compute summary statistics (transition/transversion ratio, heterozygous/homozygous counts), and annotate against the pathogenicity database. The output should be a CSV file with columns: 'metric','value'.\n<example>\nmetric,value\nraw_reads,1000000\nclean_reads,972846\nq30_rate,94.49\nmapped_reads,169419\nmapping_rate,17.34\nduplication_rate,0.28\nmean_coverage,0.22\ntotal_variants,416\nsnp_count,313\nindel_count,103\ngatk_variants,0\nti_tv_ratio,1.98\nhet_count,24\nhom_count,392\nhet_hom_ratio,0.06\nclinvar_annotated,0\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/germline-wes-gatk/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/germline-wes-gatk/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/germline-wes-gatk/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "gwas-association", |
| "name": "GWAS Population Association Testing", |
| "description": "Perform genome-wide association testing on genotype array data including quality control, population stratification correction via principal component analysis, association testing with multiple statistical methods, and result annotation. GWAS identifies genetic variants associated with a trait by testing each variant for statistical association with the phenotype while controlling for confounders. The input includes genotype data in standard binary format (BED/BIM/FAM), a binary phenotype file, and covariates.", |
| "task_prompt": "Perform a genome-wide association study on genotype data. The input genotypes (binary BED/BIM/FAM format), phenotype file (phenotype_bin.txt with Y1/Y2 binary traits, coded 0=control 1=case), and covariates (covariates.txt) are in data/. Perform genotype QC (missingness, MAF, HWE filtering), LD pruning, PCA for population stratification, heterozygosity-based outlier removal, and run at least two association methods (logistic regression with PCA covariates and a whole-genome regression approach). Compute genomic inflation factor (lambda GC), identify significant and suggestive loci, and perform LD clumping. The output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>\nmetric,value\nvariants_after_qc,1000\nsamples_after_qc,499\npca_computed_samples,500\nld_pruned_snps,1000\nmean_het_f,-0.002\nmethod1_tested_variants,1000\nmethod1_genome_wide_sig,0\nmethod1_suggestive_sig,0\nmethod1_min_pvalue,7.96e-04\ngenomic_inflation_lambda,1.1222\nmethod2_tested_variants,1000\nmethod2_suggestive_sig,0\nclumped_loci,0\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/gwas-association/data.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/gwas-association/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "dia-proteomics", |
| "name": "Label-Free Proteomics: BSA Standard Identification", |
| "description": "Identify and quantify proteins from label-free LC-MS/MS data. Three replicate BSA (bovine serum albumin) runs with ~4800 spectra, searched against 6052 reviewed bovine proteins.", |
| "task_prompt": "Identify and quantify proteins from label-free mass spectrometry data. The data/ directory contains three mzML files (BSA1.mzML, BSA2.mzML, BSA3.mzML) from replicate LC-MS/MS runs of a bovine serum albumin standard. The reference/ directory contains a protein sequence database (proteins.fasta) with reviewed bovine proteins. Generate a target-decoy database, search spectra against it using at least one search engine, score peptide-spectrum matches, perform protein inference, and compute identification statistics including sequence coverage of the primary analyte (BSA). The output should be a CSV file with columns: 'metric','value'.\n<example>metric,value\nsamples,3\ntotal_spectra,4812\ntotal_psms,2627\nunique_peptides,932\nidentified_proteins,927\nsequence_coverage_pct,18.6\navg_peptide_length,9.4\npsms_BSA1,954\npsms_BSA2,963\npsms_BSA3,710\npsms_BSA1_msgf,954\npsms_BSA2_msgf,963\npsms_BSA3_msgf,710</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/dia-proteomics/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/dia-proteomics/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/dia-proteomics/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "structural-variant-multi", |
| "name": "Structural Variant Detection: Multi-caller Human SV Analysis", |
| "description": "Structural variant (SV) detection from human whole-exome sequencing data using multiple orthogonal calling approaches. SVs include deletions, duplications, inversions, insertions, and translocations larger than 50 base pairs. This task requires aligning reads, marking duplicates, then running three independent SV detection methods: paired-end/split-read analysis, depth-based anomaly detection with large indel extraction, and discordant read pair clustering. Results are merged across callers, annotated for gene overlap and clinical significance, and summarized by SV type and size.", |
| "task_prompt": "Detect structural variants from whole-exome sequencing data using multiple independent calling approaches. The data/ directory contains paired-end FASTQ files (sample_R1.fastq.gz, sample_R2.fastq.gz). The reference/ directory contains the genome reference (genome.fa), gene annotation (genes.gtf), gene regions (gene_regions.bed), and a clinical SV database (clinvar_sv.vcf.gz). Align reads with proper read groups, sort, and mark duplicates. Run at least three independent SV detection approaches (e.g., paired-end signal analysis, depth-based detection, discordant read clustering). Merge SV calls across methods, compute SV type distribution (deletions, duplications, inversions, insertions, breakends), annotate against gene regions and the clinical database, and generate a summary report. The output should be a CSV file with columns: 'metric','value'.\n<example>\nmetric,value\nraw_reads,500000\nclean_reads,486458\nq30_rate,94.5\nmapped_reads,83358\nmapping_rate,17.06\ncaller1_svs,0\ncaller2_svs,27131\ncaller3_svs,2876\nmerged_svs,30007\ndeletion_count,30007\nduplication_count,0\ninversion_count,0\ninsertion_count,0\nbreakend_count,0\ngene_overlapping_svs,0\nclinvar_sv_annotated,0\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/structural-variant-multi/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/structural-variant-multi/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/structural-variant-multi/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "dda-lfq-proteomics", |
| "name": "DDA Label-Free Quantitative Proteomics", |
| "description": "Label-free quantitative proteomics by DDA mass spectrometry enables unbiased protein quantification across experimental conditions. This task processes four mzML files from a Q-Exactive instrument (2 conditions x 2 replicates) through a dual search engine pipeline. The workflow includes target-decoy database generation, spectral peak picking, parallel database searching with two independent algorithms, peptide indexing, PSM feature extraction, FDR control, identification filtering, label-free quantification, and cross-engine/cross-condition comparison.", |
| "task_prompt": "Quantify proteins from data-dependent acquisition (DDA) label-free mass spectrometry data using dual search engines with FDR control. Four mzML files (two conditions, two replicates each) are in data/ along with an experimental design file. A protein sequence database is in reference/. Generate a target-decoy database, perform centroiding (peak picking), search spectra against the database using two independent search engines, index peptides, extract PSM features, control false discovery rate, filter identifications, quantify features, and compare identifications between conditions and between search engines.\nThe output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>\nmetric,value\nspectra_T2_A1,8028\nspectra_T2_B1,7939\nspectra_T7A_1,2811\nspectra_T7B_1,5762\ntotal_spectra,24540\npsms_T2_A1_comet,7765\npsms_T2_B1_comet,7730\npsms_T7A_1_comet,2689\npsms_T7B_1_comet,4316\ntotal_psms_comet,22500\npsms_T2_A1_msgf,7866\npsms_T2_B1_msgf,7805\npsms_T7A_1_msgf,2736\npsms_T7B_1_msgf,4440\ntotal_psms_msgf,22847\ncomet_unique_peptides,9074\ncomet_unique_proteins,16373\nmsgf_unique_peptides,9218\nmsgf_unique_proteins,16375\npeptides_T2_A1,5177\npeptides_T2_B1,5546\npeptides_T7A_1,2389\npeptides_T7B_1,3610\ntotal_unique_peptides,9873\ntotal_unique_proteins,18236\ndatabase_protein_count,104908\nprotein_identification_rate_pct,17.38\nengine_shared_peptides,8419\nengine_overlap_pct,85.27\ncondition1_peptides,7586\ncondition2_peptides,4506\nshared_condition_peptides,2219\ncondition_overlap_pct,22.48\ntarget_sequences,104908\ndecoy_sequences,104908\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/dda-lfq-proteomics/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/dda-lfq-proteomics/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/dda-lfq-proteomics/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "clinical-wgs-interpretation", |
| "name": "Clinical WGS Interpretation: Full Clinical Genome Analysis", |
| "description": "Clinical whole-genome sequencing interpretation requires a comprehensive pipeline that integrates multiple analysis tracks: single nucleotide variant and indel detection, structural variant discovery, coverage assessment, and multi-source clinical annotation. This task processes paired-end sequencing data from a human sample (chromosome 7 subset) through the full clinical interpretation pipeline, including quality control, alignment, duplicate marking, base quality recalibration, parallel variant calling tracks, variant filtration, clinical database annotation, gene panel filtering, and generation of a comprehensive clinical report with coverage adequacy assessment.", |
| "task_prompt": "Perform clinical whole-genome sequencing interpretation from paired-end sequencing data. The data/ directory contains FASTQ files (sample_R1.fastq.gz, sample_R2.fastq.gz). The reference/ directory contains the genome (genome.fa), gene annotation (genes.gtf), known variant sites (dbsnp.vcf.gz), a clinical significance database (clinvar.vcf.gz), gene regions (gene_regions.bed), and a clinical gene panel (gene_panel.bed). Quality-filter reads, align with proper read groups, sort, mark duplicates, and recalibrate base qualities. Then run three parallel analysis tracks: (1) call small variants (SNPs and indels), normalize, and independently filter SNPs and indels with quality thresholds, (2) detect structural variants, and (3) compute genome-wide coverage statistics. Merge all variant calls, annotate against the clinical database and known variant sites, compute variant summary statistics (transition/transversion ratio, heterozygous/homozygous counts), filter to a clinical gene panel, and assess coverage adequacy by counting low-coverage regions. The output should be a CSV file with columns: 'metric','value'.\n<example>\nmetric,value\nraw_reads,1000000\nclean_reads,983862\nq30_rate,94.39\nmapped_reads,171808\nmapping_rate,17.38\nduplication_rate,0.28\nmean_coverage,0.23\nsnp_count,311\nindel_count,103\ntotal_snv,414\nsv_count,9\nti_tv_ratio,2.02\nhet_count,24\nhom_count,390\nhet_hom_ratio,0.06\nclinvar_annotated,0\ngene_panel_variants,1\nlow_coverage_regions,162832\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/clinical-wgs-interpretation/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/clinical-wgs-interpretation/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/clinical-wgs-interpretation/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "repeat-element-annotation", |
| "name": "Repeat Element Annotation (Transposable Element Analysis)", |
| "description": "Transposable element (TE) annotation identifies and classifies repetitive DNA elements in genome assemblies. This task requires building de novo repeat consensus libraries, combining them with known databases, running comprehensive genome masking, analyzing the repeat divergence landscape, computing TE-gene proximity, estimating TE ages via sequence divergence, and identifying solo-LTR elements. Key challenges include long runtime for de novo library construction, correct merging and deduplication of libraries, distinguishing TEs from simple repeats, and parsing hierarchical TE classification schemes.", |
| "task_prompt": "Annotate and classify transposable elements (TEs) in a genome assembly. The genome sequence is in data/genome.fasta and a gene annotation file is in reference/genes.gtf (with a BED version at reference/genes.bed). Build a de novo repeat library from the genome, combine it with known repeat databases, then perform comprehensive repeat annotation. Analyze the repeat landscape (divergence distribution), compute TE-gene proximity statistics, estimate TE age distributions, classify TEs into major families (LTR, LINE, DNA transposons, etc.), identify solo-LTR elements vs intact LTR retrotransposons, and analyze gene disruption by TE insertions. The output should be a CSV file at results/report.csv with columns: metric,value.\n<example>\nmetric,value\ntotal_length,24890402\nnum_sequences,2\ngc_content,0.4208\nn_content,0.0033\nseq_chr4_length,1348131\nseq_chrX_length,23542271\ntotal_repeat_elements,33148\ntotal_repeat_bp,3971801\nrepeat_fraction,0.1596\nnum_major_te_classes,7\ndenovo_te_families,128\nmerged_library_size,120\nte_ltr_bp,1174008\nte_ltr_count,1384\nte_unknown_bp,845559\nte_unknown_count,2671\nte_line_bp,701957\nte_line_count,590\nte_dna_bp,41761\nte_dna_count,105\nte_mean_divergence,11.75\nyoungest_te_class,rRNA\nte_gene_intronic,0\nte_gene_exonic,1868\nte_gene_upstream_1kb,2499\nte_gene_upstream_5kb,3422\nte_gene_intergenic,3836\nte_gene_genes_with_te_insertions,1868\ntotal_ltr_elements,1384\nsolo_ltr_candidates,1105\nintact_ltr_candidates,279\nsolo_to_intact_ratio,3.96\nlandscape_generated,yes\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/repeat-element-annotation/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/repeat-element-annotation/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/repeat-element-annotation/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "msi-detection", |
| "name": "Microsatellite Instability Detection: Multi-Caller Consensus Analysis", |
| "description": "This task analyzes a tumor-normal BAM pair to detect microsatellite instability (MSI), a biomarker for deficient DNA mismatch repair. Three independent MSI callers are run on the same data, their results are parsed and compared, and a consensus MSI status is determined by majority vote. The pipeline also calls variants at MSI loci, computes coverage depth, and estimates tumor mutation burden (TMB). The data is from the msisensor-pro project demo dataset (synthetic tumor-normal pair on a 1.5 Mb chr1 reference).", |
| "task_prompt": "Detect microsatellite instability (MSI) in a tumor-normal BAM pair using three independent callers, then compute a consensus MSI status by majority vote. Also call variants at microsatellite loci, compute coverage depth at those loci, and estimate tumor mutation burden (TMB). Tumor and normal BAMs with a reference genome are provided in data/ and reference/ directories. A pre-cloned third-party MSI caller repository is available under outputs/. The output should be a CSV file with the following columns: 'metric','value'.\n<example>metric,value\ntumor_total_reads,84535\ntumor_mapped_pct,99.96\nnormal_total_reads,167071\nnormal_mapped_pct,99.96\ncaller1_msi_score,0.0\ncaller1_total_sites,12\ncaller1_unstable_sites,0\ncaller2_msi_score,14.29\ncaller2_total_sites,7\ncaller2_unstable_sites,1\ncaller3_msi_score,9.78\ncaller3_step_wise_diff,0.0978\ncaller3_loci_analyzed,1\nconsensus_msi_status,MSS\nconsensus_msi_score,8.0233\nmsi_callers_agree,1/3\ntumor_mean_coverage,1.98\nnormal_mean_coverage,3.6\nvariants_in_msi_loci,156\ntmb_per_mb,104.0\ncallable_region_mb,1.5\ntmb_status,TMB-High</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/msi-detection/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/msi-detection/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/msi-detection/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "scatac-seq", |
| "name": "Single-Cell ATAC-seq Chromatin Accessibility Analysis", |
| "description": "Single-cell ATAC-seq (scATAC-seq) measures genome-wide chromatin accessibility at single-cell resolution, enabling identification of cell types by their regulatory landscape. This task processes a 10x Genomics PBMC scATAC-seq fragment file through quality control (TSS enrichment, fragment counts), cell filtering, tile matrix construction, dimensionality reduction, doublet detection, cell clustering, per-cluster peak calling, gene activity scoring, and marker gene identification to characterize immune cell populations by their epigenomic signatures.", |
| "task_prompt": "Analyze single-cell ATAC-seq data from human PBMCs to profile chromatin accessibility at single-cell resolution. A fragment file (fragments.tsv.gz) containing 1500 cells is in data/, along with per-barcode quality metrics. A gene annotation file (GENCODE v32 GTF) is in reference/. Import the fragments, compute quality metrics (TSS enrichment, fragment counts), filter low-quality cells, create a genomic tile matrix, perform dimensionality reduction, detect doublets, cluster cells, call accessible chromatin peaks per cluster, compute gene activity scores, identify marker genes per cluster, and generate a comprehensive report.\nThe output should be a CSV file at results/report.csv with columns: 'metric','value'.\n<example>\nmetric,value\ntotal_cells_imported,1500\ncells_after_qc,1499\ncells_after_doublet_removal,1499\nmedian_tsse,26.26\nmean_tsse,26.35\nmedian_fragments_per_cell,15537\nmean_fragments_per_cell,16862\nnum_clusters,5\ntotal_peaks,248468\nn_genes_in_activity_matrix,60606\nn_significant_markers,18038\ncluster_0_cells,588\ncluster_1_cells,291\ncluster_2_cells,256\ncluster_3_cells,234\ncluster_4_cells,130\ndoublets_detected,0\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/scatac-seq/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/scatac-seq/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/scatac-seq/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "multiomics-rna-atac", |
| "name": "Multi-omics Integration (RNA-seq + ATAC-seq)", |
| "description": "Multi-omics integration of matched RNA-seq and ATAC-seq data identifies regulatory circuits by linking chromatin accessibility to gene expression. This task requires parallel processing of two data modalities (transcriptomics and epigenomics), followed by cross-modality integration including peak-to-gene assignment, motif enrichment analysis, accessibility-expression correlation, and regulatory network construction. Key challenges include different normalization requirements across modalities, correct peak calling parameters for open chromatin data, and meaningful integration of distinct signal types.", |
| "task_prompt": "Integrate matched RNA-seq and ATAC-seq data from the same cell line to identify regulatory circuits linking chromatin accessibility to gene expression. RNA-seq paired-end reads are in data/rna_R1.fastq.gz and data/rna_R2.fastq.gz. ATAC-seq paired-end reads are in data/atac_R1.fastq.gz and data/atac_R2.fastq.gz. The reference genome (chr22) is in reference/genome.fasta and gene annotation in reference/genes.gtf. For the RNA-seq track: perform quality control, align reads, quantify transcript expression, and assemble transcripts. For the ATAC-seq track: perform quality control, align reads, remove duplicates, call accessible chromatin peaks, and generate normalized signal tracks. Then integrate both modalities: assign peaks to nearest genes, perform motif enrichment in accessible regions, correlate peak accessibility scores with gene expression, and identify regulatory circuits. The output should be a CSV file at results/report.csv with columns: metric,value.\n<example>\nmetric,value\nrna_total_reads,4000000\nrna_filtered_reads,3920910\nrna_q30_rate,97.7\natac_total_reads,8000000\natac_filtered_reads,7585464\natac_q30_rate,93.3\nrna_mapped_reads,257844\nrna_mapping_rate,4.99\natac_mapping_rate,5.36\natac_duplication_rate,8.4\ntotal_peaks,1705\nmedian_peak_width,374\ntotal_peak_bp,873953\ntotal_genes_annotated,1445\nexpressed_genes,617\nhighly_expressed_genes,103\nassembled_transcripts,1445\npeak_gene_links,1786\ncorr_expressed_genes,617\ncorr_peak_gene_links,1786\ncorr_median_peak_gene_distance,7861\ncorr_peaks_within_10kb,1002\nsignal_track_generated,yes\nmotif_total_motifs_tested,1049\nmotif_significant_motifs,84\nmotif_top_motif,Fli1\ntotal_regulatory_links,1689\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/multiomics-rna-atac/data.tar.gz" |
| } |
| ], |
| "reference_data": [ |
| { |
| "filename": "reference.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/multiomics-rna-atac/reference.tar.gz" |
| } |
| ], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/multiomics-rna-atac/results.tar.gz" |
| } |
| ] |
| } |
| }, |
| { |
| "task_id": "methylation-array-epic", |
| "name": "Methylation Array Analysis (Illumina EPIC)", |
| "description": "Illumina EPIC methylation array analysis processes raw IDAT files through normalization, quality control, differential methylation detection, and pathway enrichment. Key challenges include handling two probe types (Type I and II) with different dynamic ranges, filtering failed and cross-reactive probes, correcting for batch effects without removing biological signal, detecting differentially methylated regions from CpG-level statistics, and correcting for the gene-length bias inherent in array-based GO enrichment analysis.", |
| "task_prompt": "Analyze Illumina EPIC methylation array data for differential methylation analysis. Raw IDAT files (Red and Green channels) are in data/idat/ with subdirectories per sample, and a sample sheet is at data/idat/SampleSheet.csv. Read the raw intensity data, compute detection p-values to identify failed probes, perform probe-level quality control, normalize the data using background correction methods, check sample quality (including sex prediction from X/Y probe intensities), perform principal component analysis to assess batch effects, run differential methylation analysis at the CpG level (differentially methylated positions), detect differentially methylated regions, perform gene ontology and pathway enrichment analysis (correcting for the probe number bias inherent in array-based methylation studies), and generate visualization plots. The output should be a CSV file at results/report.csv with columns: metric,value.\n<example>\nmetric,value\nnum_samples,3\narray_type,EPIC\ntotal_probes,1052641\nprobes_passing_qc,864782\nprobes_failing_qc,2054\nprobe_pass_rate,99.76\nnormalized_probes,864782\nmean_beta_sample1,0.4549\nmean_beta_sample2,0.4614\nmean_beta_sample3,0.4594\npca_pc1_variance,53\npca_pc2_variance,47\ntotal_dmps_tested,864782\nsignificant_dmps,6\ntop_dmp_cpg,cg18899802\ntop_dmp_pvalue,3.04e-10\ntop_dmp_logfc,-6.1065\nnum_dmrs,0\nsignificant_go_terms,0\nsignificant_kegg_pathways,0\npredicted_sex,F;F;F\n</example>", |
| "download_urls": { |
| "data": [ |
| { |
| "filename": "data.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/methylation-array-epic/data.tar.gz" |
| } |
| ], |
| "reference_data": [], |
| "results": [ |
| { |
| "filename": "results.tar.gz", |
| "url": "https://huggingface.co/datasets/lingzhi227/Extended-BioAgentBench/resolve/main/tasks/methylation-array-epic/results.tar.gz" |
| } |
| ] |
| } |
| } |
| ] |
