Biomanus/experiments/supplementary_info/graph_retrieval_case_studies.csv

Task,Run Variant,Validation,Best LLM Score,Query,Extracted Semantics,Subtasks,Workflow Type,Operation Path,Semantic Type Trace,Selected MCP Servers,Server Descriptions,Retrieved Subgraph Nodes,Retrieved Subgraph Edges,Top Candidate Tools,Output Summary
deseq,aligned,passed,0.95,Perform differential expression analysis using DESeq2 on RNA-Seq data from Candida parapsilosis WT strains under planktonic vs biofilm conditions.,RNA-Seq; differential expression; DESeq2; Candida parapsilosis; planktonic; biofilm; count matrix; up-regulated genes; rewrite; bioagent-bench,"Load RNA-Seq count matrix and sample metadata (condition: planktonic vs biofilm). | Construct DESeqDataSet from count matrix and design formula (~ condition). | Run DESeq2 pipeline: estimate size factors, estimate dispersions, fit negative binomial model, perform Wald test. | Extract results table with log2FoldChange, pvalue, padj. | Filter for significant up-regulated genes (padj < 0.05, log2FoldChange > 0). | Write filtered results to up_regulated_genes.csv with columns: gene_id, log2FoldChange, pvalue, padj.",linear_typed_transformation,agent_reasoning -> sequence_feature_extraction -> count_normalization -> differential_expression -> gene_filtering -> csv_export,fastq -> report_table -> annotation_gff -> count_matrix -> count_matrix -> count_matrix -> count_matrix,"gffread, bioconductor-deseq2",gffread: transcript annotation and sequence feature extraction utility | bioconductor-deseq2: Differential gene expression analysis based on the negative binomial distribution | Estimate variance-mean dependence in count data from high-throughput sequencing assays and test for differential expression based on a model using the negative binomial distribution.,80,202,gffread.gffread_tool (64.1); crispresso2.crispresso_batch (58.05); crispresso2.crispresso (58.05); gffutils.gffutils_update (57.375); gffutils.gffutils_create (57.375),"rows=2232 columns=gene_id,log2FoldChange,pvalue,padj"
cystic-fibrosis,aligned,passed,0.95,Identify the causal recessive CFTR variant in a CEPH/Utah family trio with three affected siblings using WGS data.,Cystic Fibrosis; CFTR; Mendelian recessive; family WGS; CEPH 1463; affected siblings; homozygous alternate; ClinVar annotation; recessive inheritance filter; joint-called VCF,"1. Load joint-called VCF for CEPH_1463 family (4 grandparents, 2 parents, 11 siblings). | 2. Filter to biallelic SNVs/indels with GQ>=20 and DP>=10. | 3. Apply recessive inheritance filter: affected siblings (NA12885, NA12886, NA12879) must be homozygous alternate; all other family members must be either homozygous reference or heterozygous (no homozygous alternate in unaffected). | 4. Restrict to coding, splice-site, or regulatory regions of CFTR gene (chr7:117,120,000-117,308,000). | 5. Annotate remaining variants with VEP or similar for gene name, transcript, HGVS, impact, and ClinVar clinical significance. | 6. Select the highest-confidence pathogenic/likely-pathogenic variant; if multiple, prioritize highest impact (e.g., stop-gained > missense). | 7. Format output as CSV with columns: chromosome,position,variant_id,reference,alternate,gene_name,gene_id,annotation,impact,transcript_id,hgvs_c,hgvs_p,clinical_significance,diseases,review_status,rs_id. | 8. Ensure variant_id is the ClinVar VCF ID and rs_id is the numeric dbSNP ID without 'rs' prefix.",linear_typed_transformation,variant_annotation -> variant_filtering -> variant_annotation -> database_lookup -> variant_summary -> csv_export,bam -> bam -> fastq -> genome_assembly -> genome_assembly -> bam -> report_table,"bcftools, ensembl-vep","bcftools: VCF/BCF variant filtering and manipulation utility | ensembl-vep: Ensembl Variant Effect Predictor | The VEP determines the effect of your variants (SNPs, insertions, deletions, CNVs or structural variants) on genes, transcripts, and protein sequence, as well as regulatory regions.",76,206,ensembl-vep.haplo (45.3); gffread.gffread_tool (45.2); crispresso2.crispresso_wgs (44.3); crispresso2.crispresso_pooled (44.3); crispresso2.crispresso_batch (44.3),"rows=1 columns=chromosome,position,variant_id,reference,alternate,gene_name"
metagenomics,aligned,passed,0.913,Taxonomic profiling and relative abundance comparison of bacterial communities from control and nutrient-enriched metagenomic samples.,metagenomics; taxonomic profiling; relative abundance; phylum; Cuatro Ciénegas; JC1A; JP4D; paired-end reads; FASTQ; Kraken2,"Quality control and trimming of raw FASTQ sequencing reads. | Taxonomic classification of reads against bacterial reference databases (e.g., Kraken2/Bracken, MetaPhlAn, or Kaiju). | Estimation of relative abundances for bacterial phyla from classified reads. | Formatting output as CSV with schema: OTU, Kingdom, Phylum, JP4D, JC1A. | Optionally, compute alpha/beta diversity metrics if needed for validation.",linear_typed_transformation,taxonomic_classification -> csv_export -> option_comparison,fastq -> taxonomic_profile -> count_matrix -> report_table,"kraken2, metaphlan2",kraken2: k-mer based metagenomic taxonomic classification utility | metaphlan2: Metagenomic Phylogenetic Analysis | MetaPhlAn is a computational tool for profiling the composition of microbial,70,122,gffread.gffread_tool (52.8); qcatch.qcatch (49.8); crispresso2.crispresso_batch (48.8); crispresso2.crispresso (48.8); crispresso2.crispresso_wgs (46.8),"rows=43 columns=OTU,Kingdom,Phylum,JP4D,JC1A"
transcript-quant,guarded,passed,1.0,"Quantify transcript expression from simulated paired-end RNA-Seq reads using a reference transcriptome, producing an exact count TSV file.",transcript quantification; RNA-Seq; paired-end reads; FASTQ.GZ; transcriptome FASTA; pseudo-alignment; Salmon; kallisto; TSV; no header,"Input: paired-end FASTQ.GZ files (reads_1.fq.gz, reads_2.fq.gz) and reference transcriptome FASTA (transcriptome.fa). | Step 1: Align or pseudo-align reads to transcriptome using a lightweight quantification tool (e.g., Salmon, kallisto). | Step 2: Run quantification to generate per-transcript read counts (exact, deterministic for simulated data). | Step 3: Extract transcript IDs and count values, format as two-column TSV with no header, tab-separated. | Output: single file truth.tsv (no header, no extra lines).",linear_typed_transformation,alignment -> quantification -> csv_export,fastq -> annotation_gff -> annotation_gff -> report_table,gffread,gffread: transcript annotation and sequence feature extraction utility,76,155,gffread.gffread_tool (60.1); crispresso2.crispresso_batch (54.05); crispresso2.crispresso (54.05); crispresso2.crispresso_wgs (52.05); crispresso2.crispresso_pooled (52.05),rows=278 columns=