Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,51 @@
+# Extended-BioAgentBench
+
+Extension of [BioAgentBench](https://github.com/bioagent-bench/bioagent-bench) with 3 additional bioinformatics benchmark tasks covering new domains: epigenomics (ChIP-seq), bacterial genome assembly, and mobile genetic element characterization.
+
+## Tasks
+
+| # | Task ID | Domain | CLI Tools | DAG Depth | Difficulty |
+|---|---------|--------|-----------|-----------|------------|
+| 11 | chipseq-peak-calling | Epigenomics | trimmomatic, bwa, samtools, macs2, bedtools, deeptools | 2 | medium |
+| 12 | bacterial-assembly | Genome assembly + annotation | fastp, shovill, quast, busco, prokka, mlst, abricate | 2 | medium |
+| 13 | mobile-elements | Mobile genetic elements | fastp, shovill, prokka, isescan, integron_finder, amrfinder, abricate, mob_recon, plasmidfinder, quast, busco | 5 | hard |
+
+## Usage
+
+```bash
+# Install dependencies
+pip install requests click
+
+# List tasks
+python src/dataset.py list-tasks
+
+# Download a specific task
+python src/dataset.py download --task chipseq-peak-calling --reference --results
+
+# Download all tasks
+python src/dataset.py download --all --reference --results
+```
+
+## Task Format
+
+Each task follows the BioAgentBench format:
+
+```
+tasks/{task-id}/
+  run_script.sh        # Reference pipeline (ground truth generator)
+  environment.yml      # Conda environment specification
+  data/                # Input data (FASTQ, BED, etc.)
+  reference/           # Reference genomes (if needed)
+  results/             # Ground truth output
+```
+
+## Evaluation
+
+Tasks are evaluated using GPT-5.1 as an LLM-as-Judge, scoring:
+- `steps_completed` / `steps_to_completion`
+- `completion_rate` (0-1)
+- `results_match` (full_match / partial_match / no_match)
+
+## Citation
+
+Based on BioAgentBench: https://arxiv.org/abs/2601.21800