Dataset Viewer
Duplicate
Search is not available for this dataset
text
string
# BUSCO version is: 5.8.0
# The lineage dataset is: bacillales_odb10 (Creation date: 2021-02-23, number of genomes: 412, number of BUSCOs: 450)
# Summarized benchmarking in BUSCO notation for file /mnt/pulsar/files/staging/12682572/working/input.fa
# BUSCO was run in mode: prok_genome_prod
# Gene predictor used: prodigal
***** Results: *****
C:94.0%[S:93.8%,D:0.2%],F:3.1%,M:2.9%,n:450
423 Complete BUSCOs (C)
422 Complete and single-copy BUSCOs (S)
1 Complete and duplicated BUSCOs (D)
14 Fragmented BUSCOs (F)
13 Missing BUSCOs (M)
450 Total BUSCO groups searched
Assembly Statistics:
21 Number of scaffolds
21 Number of contigs
3992854 Total length
0.000% Percent gaps
517 KB Scaffold N50
517 KB Contigs N50
Dependencies and versions:
hmmsearch: 3.4
bbtools: None
prodigal: 2.6.3
python: sys.version_info(major=3, minor=10, micro=16, releaselevel='final', serial=0)
busco: 5.8.0
# BUSCO version is: 5.8.0
# The lineage dataset is: bacillales_odb10 (Creation date: 2021-02-23, number of genomes: 412, number of BUSCOs: 450)
# Summarized benchmarking in BUSCO notation for file /mnt/pulsar/files/staging/12682567/working/input.fa
# BUSCO was run in mode: prok_genome_prod
# Gene predictor used: prodigal
***** Results: *****
C:36.0%[S:36.0%,D:0.0%],F:38.9%,M:25.1%,n:450
162 Complete BUSCOs (C)
162 Complete and single-copy BUSCOs (S)
0 Complete and duplicated BUSCOs (D)
175 Fragmented BUSCOs (F)
113 Missing BUSCOs (M)
450 Total BUSCO groups searched
Assembly Statistics:
21 Number of scaffolds
21 Number of contigs
3982784 Total length
0.000% Percent gaps
516 KB Scaffold N50
516 KB Contigs N50
Dependencies and versions:
hmmsearch: 3.4
bbtools: None
prodigal: 2.6.3
python: sys.version_info(major=3, minor=10, micro=16, releaselevel='final', serial=0)
busco: 5.8.0
# BUSCO version is: 5.8.0
# The lineage dataset is: bacillales_odb10 (Creation date: 2021-02-23, number of genomes: 412, number of BUSCOs: 450)
# Summarized benchmarking in BUSCO notation for file /mnt/pulsar/files/staging/12682562/working/input.fa
# BUSCO was run in mode: prok_genome_prod
# Gene predictor used: prodigal
***** Results: *****
C:99.1%[S:98.9%,D:0.2%],F:0.9%,M:0.0%,n:450
446 Complete BUSCOs (C)
445 Complete and single-copy BUSCOs (S)
1 Complete and duplicated BUSCOs (D)
4 Fragmented BUSCOs (F)
0 Missing BUSCOs (M)
450 Total BUSCO groups searched
Assembly Statistics:
1 Number of scaffolds
1 Number of contigs
4045677 Total length
0.000% Percent gaps
4 MB Scaffold N50
4 MB Contigs N50
Dependencies and versions:
hmmsearch: 3.4
bbtools: None
prodigal: 2.6.3
python: sys.version_info(major=3, minor=10, micro=16, releaselevel='final', serial=0)
busco: 5.8.0
# BUSCO version is: 5.8.0
# The lineage dataset is: bacillales_odb10 (Creation date: 2021-02-23, number of genomes: 412, number of BUSCOs: 450)
# Summarized benchmarking in BUSCO notation for file /mnt/pulsar/files/staging/12682569/working/input.fa
# BUSCO was run in mode: prok_genome_prod
# Gene predictor used: prodigal
***** Results: *****

Bioinformatics-Workflow-Bench

142 execution-verified bioinformatics workflow tasks across Galaxy, Nextflow, and WDL platforms.

Tasks Verified Platforms Sources

Overview

Bioinformatics-Workflow-Bench is a unified benchmark for evaluating AI agents on real-world bioinformatics workflow execution tasks. It merges 142 execution-verified tasks from four peer-reviewed sources:

Source Platform Tasks Ground Truth Verification Method
IWC Galaxy 75 Planemo test assertions CI-verified on every PR
nf-core Nextflow 45 nf-test MD5 snapshots Docker execution + snapshot comparison
GTN Galaxy 14 Planemo test assertions Execution on usegalaxy.org
BioWDL / WARP / ENCODE WDL 8 pytest-workflow miniwdl + Docker execution

Difficulty distribution: 31 easy, 52 medium, 37 hard, 22 expert.

Every task has been executed end-to-end and its outputs verified against ground truth assertions filed by human domain experts. No task is included based on structural validation alone.

What makes this benchmark different?

  • Execution-verified ground truth: Every task has been run on a real compute platform with outputs checked against expert-filed assertions
  • Three workflow platforms: Galaxy (.ga), Nextflow (nf-core), and WDL -- tests agent adaptability across ecosystems
  • Peer-reviewed workflows: All source workflows passed human code review and automated CI before publication
  • Real experimental data: Input data from Zenodo DOIs, nf-core test profiles, or bundled test datasets -- no simulated data
  • Three prompt tiers: Each task has open (biological question only), guided (tool names given), and assisted (step-by-step scaffold) prompts
  • Task-specific rubrics: Auto-generated evaluation rubrics with expected steps, output assertions, and scoring weights
  • Wide domain coverage: 35 domains from genomics to computational chemistry, single-cell to metagenomics

Tasks

Galaxy / IWC (75 tasks)

# Task ID Name Difficulty Domain Steps Assertions Notes
1 iwc--velocyto Velocyto easy scRNAseq 1 1 IWC Planemo CI verified
2 iwc--avg-bigwig Avg Bigwig easy other 2 2 IWC Planemo CI verified
3 iwc--baredsc Baredsc easy scRNAseq 2 23 IWC Planemo CI verified
4 iwc--short-read-qc Short Read Qc easy read-preprocessing 2 2 IWC Planemo CI verified
5 iwc--cgmlst-bacterial-genome Cgmlst Bacterial Genome easy bacterial_genomics 3 7 IWC Planemo CI verified
6 iwc--assembly-with-flye Assembly With Flye easy genome-assembly 4 10 IWC Planemo CI verified
7 iwc--fastq-to-matrix-10x Fastq To Matrix 10X easy scRNAseq 4 5 IWC Planemo CI verified
8 iwc--hyphy Hyphy easy comparative_genomics 4 8 IWC Planemo CI verified
9 iwc--parallel-accession-download Parallel Accession Download easy data-fetching 4 1 IWC Planemo CI verified
10 iwc--qcxms-sdf Qcxms Sdf easy metabolomics 4 1 IWC Planemo CI verified
11 iwc--bacterial-genome-assembly Bacterial Genome Assembly easy genome-assembly 5 6 IWC Planemo CI verified
12 iwc--brew3r Brew3R easy transcriptomics 5 2 IWC Planemo CI verified
13 iwc--goseq Goseq easy transcriptomics 5 25 IWC Planemo CI verified
14 iwc--lncrna-annotation Lncrna Annotation easy genome_annotation 5 10 IWC Planemo CI verified
15 iwc--vgp0-mitogenome Vgp0 Mitogenome easy VGP-assembly 5 4 IWC Planemo CI verified
16 iwc--annotation-braker3 Annotation Braker3 medium genome_annotation 6 12 IWC Planemo CI verified
17 iwc--amr-gene-detection Amr Gene Detection medium other 7 21 IWC Planemo CI verified
18 iwc--annotation-helixer Annotation Helixer medium genome_annotation 7 13 IWC Planemo CI verified
19 iwc--chipseq-pe Chipseq Pe medium epigenetics 7 18 IWC Planemo CI verified
20 iwc--chipseq-sr Chipseq Sr medium epigenetics 7 18 IWC Planemo CI verified
21 iwc--gcms-metams Gcms Metams medium metabolomics 7 7 IWC Planemo CI verified
22 iwc--nuclei-segmentation Nuclei Segmentation medium imaging 7 3 IWC Planemo CI verified
23 iwc--raw-reads-qc Raw Reads Qc medium other 7 10 IWC Planemo CI verified
24 iwc--annotation-maker Annotation Maker medium genome_annotation 8 19 IWC Planemo CI verified
25 iwc--openms-metaprosip Openms Metaprosip medium proteomics 8 6 IWC Planemo CI verified
26 iwc--polish-long-reads Polish Long Reads medium other 8 1 IWC Planemo CI verified
27 iwc--bacterial-genome-annotation Bacterial Genome Annotation medium bacterial_genomics 9 15 IWC Planemo CI verified
28 iwc--cutandrun Cutandrun medium epigenetics 9 17 IWC Planemo CI verified
29 iwc--mfassignr Mfassignr medium metabolomics 9 15 IWC Planemo CI verified
30 iwc--protein-ligand-param Protein Ligand Param medium other 9 2 IWC Planemo CI verified
31 iwc--sars-cov2-se-wgs Sars Cov2 Se Wgs medium sars-cov-2-variant-calling 9 2 IWC Planemo CI verified
32 iwc--mag-genome-annotation-parallel Mag Genome Annotation Parallel medium microbiome 10 7 IWC Planemo CI verified
33 iwc--fragment-docking Fragment Docking medium computational-chemistry 11 1 IWC Planemo CI verified
34 iwc--rnaseq-sr Rnaseq Sr medium transcriptomics 11 9 IWC Planemo CI verified
35 iwc--sars-cov2-pe-wgs Sars Cov2 Pe Wgs medium sars-cov-2-variant-calling 11 1 IWC Planemo CI verified
36 iwc--vgp2-kmer-trio Vgp2 Kmer Trio medium VGP-assembly 11 7 IWC Planemo CI verified
37 iwc--vgp9-decontamination Vgp9 Decontamination medium VGP-assembly 11 18 IWC Planemo CI verified
38 iwc--generic-vc Generic Vc medium other 12 5 IWC Planemo CI verified
39 iwc--lcms-preprocessing Lcms Preprocessing medium metabolomics 12 2 IWC Planemo CI verified
40 iwc--ploidy-vc Ploidy Vc medium other 12 8 IWC Planemo CI verified
41 iwc--rnaseq-pe Rnaseq Pe medium transcriptomics 12 11 IWC Planemo CI verified
42 iwc--vgp7-scaffolding-bionano Vgp7 Scaffolding Bionano medium VGP-assembly 12 5 IWC Planemo CI verified
43 iwc--bacterial-qc-post-assembly Bacterial Qc Post Assembly medium bacterial_genomics 13 18 IWC Planemo CI verified
44 iwc--haploid-vc Haploid Vc medium other 13 4 IWC Planemo CI verified
45 iwc--pseudobulk-decoupler-edger Pseudobulk Decoupler Edger medium other 13 5 IWC Planemo CI verified
46 iwc--sra-manifest-fastqs Sra Manifest Fastqs medium other 13 2 IWC Planemo CI verified
47 iwc--dada2 Dada2 medium amplicon 14 9 IWC Planemo CI verified
48 iwc--vgp-plot-nx-size Vgp Plot Nx Size medium VGP-assembly 14 4 IWC Planemo CI verified
49 iwc--vgp1-kmer-profiling Vgp1 Kmer Profiling medium VGP-assembly 14 9 IWC Planemo CI verified
50 iwc--gromacs-mmgbsa Gromacs Mmgbsa hard computational-chemistry 17 4 IWC Planemo CI verified
51 iwc--rnaseq-de Rnaseq De hard transcriptomics 17 6 IWC Planemo CI verified
52 iwc--sars-cov2-consensus Sars Cov2 Consensus hard sars-cov-2-variant-calling 17 1 IWC Planemo CI verified
53 iwc--metagenomic-raw-reads-amr Metagenomic Raw Reads Amr hard microbiome 20 22 IWC Planemo CI verified
54 iwc--sars-cov2-ivar Sars Cov2 Ivar hard sars-cov-2-variant-calling 20 5 IWC Planemo CI verified
55 iwc--gromacs-dctmd Gromacs Dctmd hard computational-chemistry 21 6 IWC Planemo CI verified
56 iwc--consensus-peaks Consensus Peaks hard epigenetics 22 7 IWC Planemo CI verified
57 iwc--mags-taxonomy-annotation Mags Taxonomy Annotation hard microbiome 22 14 IWC Planemo CI verified
58 iwc--sars-cov2-ont-artic Sars Cov2 Ont Artic hard sars-cov-2-variant-calling 22 1 IWC Planemo CI verified
59 iwc--viral-generic Viral Generic hard other 23 13 IWC Planemo CI verified
60 iwc--sars-cov2-pe-artic Sars Cov2 Pe Artic hard sars-cov-2-variant-calling 25 2 IWC Planemo CI verified
61 iwc--atacseq Atacseq hard epigenetics 27 19 IWC Planemo CI verified
62 iwc--sars-cov2-reporting Sars Cov2 Reporting hard sars-cov-2-variant-calling 30 2 IWC Planemo CI verified
63 iwc--variation-reporting Variation Reporting hard sars-cov-2-variant-calling 31 4 IWC Planemo CI verified
64 iwc--metagenomic-genes-catalogue Metagenomic Genes Catalogue hard microbiome 33 17 IWC Planemo CI verified
65 iwc--vgp6b-purge-dups-haplotype Vgp6B Purge Dups Haplotype hard VGP-assembly 35 10 IWC Planemo CI verified
66 iwc--mags-building Mags Building expert microbiome 38 5 IWC Planemo CI verified
67 iwc--vgp3-hifi-assembly Vgp3 Hifi Assembly expert VGP-assembly 42 11 IWC Planemo CI verified
68 iwc--vgp5-hifi-trio-assembly Vgp5 Hifi Trio Assembly expert VGP-assembly 44 8 IWC Planemo CI verified
69 iwc--vgp6-purge-dups Vgp6 Purge Dups expert VGP-assembly 45 13 IWC Planemo CI verified
70 iwc--binning-evaluation Binning Evaluation expert microbiome 47 7 IWC Planemo CI verified
71 iwc--influenza Influenza expert virology 54 8 IWC Planemo CI verified
72 iwc--scanpy-clustering Scanpy Clustering expert scRNAseq 55 51 IWC Planemo CI verified
73 iwc--vgp8-scaffolding-hic Vgp8 Scaffolding Hic expert VGP-assembly 55 17 IWC Planemo CI verified
74 iwc--hic-contact-map Hic Contact Map expert VGP-assembly 56 22 IWC Planemo CI verified
75 iwc--vgp4-hifi-hic-assembly Vgp4 Hifi Hic Assembly expert VGP-assembly 57 22 IWC Planemo CI verified

Galaxy / GTN (14 tasks)

# Task ID Name Difficulty Domain Steps Assertions Notes
1 gtn--abc_intro_phylo Abc Intro Phylo easy evolution 3 3 Verified with planemo test on usegalaxy.org. All assertions passed.
2 gtn--bacterial-isolate-species-contamination-checking Bacterial Isolate Species Contamination Checking easy ecology 3 5 Verified on usegalaxy.org
3 gtn--mitochondrion-assembly Mitochondrion Assembly easy assembly 3 4 Verified on usegalaxy.org
4 gtn--quality-contamination-control Quality Contamination Control easy sequence-analysis 5 6 Verified on usegalaxy.org
5 gtn--small_ncrna_clustering Small Ncrna Clustering easy transcriptomics 5 4 Verified on usegalaxy.org
6 gtn--mrsa-illumina Mrsa Illumina medium assembly 6 4 Verified on usegalaxy.org
7 gtn--unicycler-assembly Unicycler Assembly medium assembly 6 3 Verified on usegalaxy.org
8 gtn--med-chem-data Med Chem Data medium comp-chem 7 8 Verified on usegalaxy.org
9 gtn--scrna-raceid Scrna Raceid medium scRNAseq 8 3 Verified interactively on usegalaxy.org. All 23 jobs completed. Trajectory analy
10 gtn--galaxy-intro-101-everyone Galaxy Intro 101 Everyone medium introductory 9 4 Verified on usegalaxy.org
11 gtn--assembly-decontamination Assembly Decontamination medium assembly 10 3 Verified with planemo test on usegalaxy.org. All assertions passed.
12 gtn--hybrid_denovo_assembly Hybrid Denovo Assembly medium assembly 12 11 Verified on usegalaxy.org
13 gtn--gtn-hic-drosophila Gtn Hic Drosophila hard epigenetics 18 2 Verified on usegalaxy.org
14 gtn--tal1-binding-site-identification Tal1 Binding Site Identification expert epigenetics 57 3 Verified on usegalaxy.org

Nextflow / nf-core (45 tasks)

# Task ID Name Difficulty Domain Steps Assertions Notes
1 nfcore--metaboigniter nf-core/metaboigniter easy metagenomics 2 3 Verified with nf-test snapshots
2 nfcore--ribomsqc nf-core/ribomsqc easy other 2 14 Verified with nf-test snapshots
3 nfcore--drugresponseeval nf-core/drugresponseeval easy other 3 10 Verified with nf-test snapshots
4 nfcore--nanostring nf-core/nanostring easy gene-expression 3 13 Verified on Docker workstation. Pipeline completed successfully.
5 nfcore--fetchngs nf-core/fetchngs easy metagenomics 4 55 27/30 strict MD5 match. 2 'missing' are pipeline_info files with timestamps in f
6 nfcore--epitopeprediction nf-core/epitopeprediction easy other 5 139 Verified with nf-test snapshots
7 nfcore--detaxizer nf-core/detaxizer medium other 8 46 Verified with nf-test snapshots
8 nfcore--fastqrepair nf-core/fastqrepair medium other 8 39 15/30 strict MD5 match. All 15 mismatches are MultiQC report files (SVG plots, l
9 nfcore--molkart nf-core/molkart medium other 8 18 Verified on Docker workstation. Pipeline completed successfully.
10 nfcore--airrflow nf-core/airrflow medium other 9 51 Docker-only pipeline. 18/18 snapshot MD5s matched on Docker profile.
11 nfcore--readsimulator nf-core/readsimulator medium simulation 9 8 Verified with nf-test snapshots
12 nfcore--hic nf-core/hic medium 3d-genome 10 18 Verified on Docker workstation. Pipeline completed successfully.
13 nfcore--bamtofastq nf-core/bamtofastq medium other 12 34 Verified with nf-test snapshots
14 nfcore--pairgenomealign nf-core/pairgenomealign medium sequence-analysis 12 20 Verified on Docker workstation. Pipeline completed successfully.
15 nfcore--mhcquant nf-core/mhcquant medium proteomics 13 107 Verified with nf-test snapshots
16 nfcore--isoseq nf-core/isoseq medium transcriptomics 14 148 20/30 strict MD5 match. 15 expected files are BAM/PBI excluded from MD5 computat
17 nfcore--phyloplace nf-core/phyloplace medium other 15 31 Verified with nf-test snapshots
18 nfcore--coproid nf-core/coproid hard other 16 1 Verified on Docker workstation. Pipeline completed successfully.
19 nfcore--pangenome nf-core/pangenome hard pangenomics 16 39 Verified with nf-test snapshots
20 nfcore--demultiplex nf-core/demultiplex hard data-processing 17 301 Verified on Docker workstation. Pipeline completed successfully.
21 nfcore--differentialabundance nf-core/differentialabundance hard other 19 53 Conda GUNZIP path unresolvable. Docker profile works. Verified on Docker worksta
22 nfcore--createtaxdb nf-core/createtaxdb hard metagenomics 20 21 Verified on Docker workstation. Pipeline completed successfully.
23 nfcore--scrnaseq nf-core/scrnaseq hard transcriptomics 20 136 Docker-only pipeline (CELLRANGERARC_COUNT requires Docker). Verified on Docker w
24 nfcore--magmap nf-core/magmap hard metagenomics 21 80 Verified with nf-test snapshots
25 nfcore--pacvar nf-core/pacvar hard variant-calling 21 12 Verified with nf-test snapshots
26 nfcore--genomeassembler nf-core/genomeassembler hard assembly 22 33 7/8 strict MD5 match. 1 mismatch is assembler output (Flye/Hifiasm non-determini
27 nfcore--methylong nf-core/methylong hard epigenetics 23 38 Verified with nf-test snapshots
28 nfcore--metatdenovo nf-core/metatdenovo hard metagenomics 24 5 0/5 strict MD5 match. All 5 expected entries are 'versions.yml' from different m
29 nfcore--scnanoseq nf-core/scnanoseq hard scRNAseq 25 2 Verified with nf-test snapshots
30 nfcore--smrnaseq nf-core/smrnaseq hard transcriptomics 30 156 Verified on Docker workstation. Pipeline completed successfully.
31 nfcore--multiplesequencealign nf-core/multiplesequencealign hard sequence-analysis 31 16 Verified with nf-test snapshots
32 nfcore--atacseq nf-core/atacseq hard epigenetics 34 189 Verified on Docker workstation. Pipeline completed successfully. (Failed on cond
33 nfcore--chipseq nf-core/chipseq hard epigenetics 34 397 Verified with nf-test snapshots
34 nfcore--callingcards nf-core/callingcards hard other 35 370 25/30 strict MD5 match. 5 mismatches are biologically equivalent: genes.bed (ref
35 nfcore--variantbenchmarking nf-core/variantbenchmarking expert variant-calling 36 52 Verified on Docker workstation. Pipeline completed successfully.
36 nfcore--pathogensurveillance nf-core/pathogensurveillance expert other 37 73 Verified with nf-test snapshots
37 nfcore--methylseq nf-core/methylseq expert epigenetics 39 134 Verified with nf-test snapshots
38 nfcore--phaseimpute nf-core/phaseimpute expert other 39 13 0/12 on conda (all MD5 mismatch). Docker profile produces correct outputs. Verif
39 nfcore--riboseq nf-core/riboseq expert other 40 231 Verified on Docker workstation. Pipeline completed successfully.
40 nfcore--rnavar nf-core/rnavar expert transcriptomics 41 27 21/25 strict MD5 match. 2 'missing' are BAM files excluded from MD5 by design (.
41 nfcore--nascent nf-core/nascent expert other 51 67 23/30 strict MD5 match. 7 mismatches are biologically equivalent: 6 samtools sta
42 nfcore--taxprofiler nf-core/taxprofiler expert metagenomics 53 7 7/7 strict MD5 match (100%). Pipeline exit=1 from a non-essential late step. All
43 nfcore--viralrecon nf-core/viralrecon expert variant-calling 57 158 Verified on Docker workstation. Pipeline completed successfully.
44 nfcore--viralmetagenome nf-core/viralmetagenome expert variant-calling 69 258 22/30 strict MD5 match. 8 mismatches are biologically equivalent: all are assemb
45 nfcore--rnaseq nf-core/rnaseq expert transcriptomics 70 657 Verified with nf-test snapshots

WDL (8 tasks)

# Task ID Name Difficulty Domain Steps Assertions Notes
1 wdl--qc Qc easy quality-control 2 91 Verified with miniwdl + Docker. Pipeline completed successfully.
2 wdl--aligning Aligning easy alignment 4 8 Verified with miniwdl + Docker. Pipeline completed successfully.
3 wdl--expression_quantification Expression Quantification easy transcriptomics 4 32 Verified with miniwdl + Docker.
4 wdl--bam_metrics Bam Metrics easy quality-control 5 70 Verified with miniwdl + Docker. Pipeline completed successfully.
5 wdl--gatk_preprocess Gatk Preprocess easy preprocessing 5 9 Verified with miniwdl + Docker. Pipeline completed successfully.
6 wdl--gatk_variantcalling Gatk Variantcalling hard variant-calling 17 51 Verified with miniwdl + Docker. Pipeline completed successfully (exit 0).
7 wdl--structural_variantcalling Structural Variantcalling hard variant-calling 23 20 Verified with miniwdl + Docker. Pipeline completed successfully.
8 wdl--somatic_variantcalling Somatic Variantcalling hard variant-calling 25 308 Verified with miniwdl + Docker. Pipeline completed successfully (exit 0).
# Clone
git clone https://github.com/lingzhi227/Bioinformatics-Workflow-Bench.git
cd Bioinformatics-Workflow-Bench

# List all tasks
python src/dataset.py list-tasks

# Filter by platform or difficulty
python src/dataset.py list-tasks --platform galaxy
python src/dataset.py list-tasks --difficulty hard

# Show task details
python src/dataset.py show-task --task iwc--rnaseq-pe

# Show ground truth assertions
python src/dataset.py show-assertions --task nfcore--rnaseq

# Download input data
python src/dataset.py download --task gtn--mrsa-illumina

Task format

tasks/{source}--{task-id}/
  prompt_open.md              # Biological question only (agent must infer tools and pipeline)
  prompt_guided.md            # Pipeline structure with tool names
  prompt_assisted.md          # Step-by-step scaffold with verification hints
  rubric.json                 # Evaluation rubric (steps, assertions, scoring weights)

  # Platform-specific files:
  # Galaxy (IWC/GTN):
  workflow/
    *.ga                      # Galaxy workflow file
    *-tests.yml               # Planemo test assertions (ground truth)
    test-data/                # Local test data (if any)
  environment.yml             # Conda environment
  Dockerfile                  # Container definition
  run_script.sh               # Execution script (IWC only)

  # Nextflow (nf-core):
  nfcore/
    test.config               # Nextflow test configuration
    nextflow_schema.json       # Parameter schema
    tests/*.nf.test           # nf-test definitions
    tests/*.nf.test.snap      # Snapshot assertions (MD5 ground truth)
  test_config.txt             # Original test config

  # WDL:
  wdl/
    *.wdl                     # WDL workflow files
  tests/
    *.json                    # Input JSONs
    data/                     # Test data

Prompt tiers

  • prompt_open.md: Describes only the biological objective. The agent must determine which tools, pipeline, and parameters to use. Tests biological reasoning + autonomous planning.
  • prompt_guided.md: Provides the pipeline structure with tool names. Tests whether the agent can execute a known workflow correctly.
  • prompt_assisted.md: Adds explicit step ordering and verification targets. Tests reproducibility with reduced planning ambiguity.

The delta across open/guided/assisted scores reveals where agents struggle: biological reasoning, tool selection, or faithful execution.

rubric.json schema

{
  "task_id": "iwc--rnaseq-pe",
  "original_task_id": "rnaseq-pe",
  "platform": "galaxy",
  "source": "IWC",
  "verification": "execution-verified",
  "metadata": {
    "difficulty": "medium",
    "domain": "transcriptomics",
    "biological_steps": 6,
    "assertion_count": 24,
    "primary_tools": ["hisat2", "featurecounts", "deseq2"]
  },
  "scoring_weights": {
    "pipeline_completion": 0.4,
    "result_correctness": 0.4,
    "biological_validity": 0.2
  }
}

Evaluation

# Deterministic rubric-based evaluation
python scripts/eval_agent.py --task iwc--rnaseq-pe --workspace /path/to/agent/output

# Agentic LLM judge (requires OpenAI API key)
python scripts/judge_agent.py --task nfcore--rnaseq --workspace /path/to/output \
  --api-key "$OPENAI_API_KEY"

Scoring formula: 40% pipeline completion + 40% result correctness + 20% biological validity.

Domain coverage

Domain Tasks Platforms
other 27 galaxy, nextflow
VGP-assembly 13 galaxy
transcriptomics 12 galaxy, nextflow, wdl
epigenetics 11 galaxy, nextflow
sars-cov-2-variant-calling 8 galaxy
variant-calling 7 nextflow, wdl
scRNAseq 6 galaxy, nextflow
microbiome 6 galaxy
metagenomics 6 nextflow
assembly 6 galaxy, nextflow
genome_annotation 4 galaxy
metabolomics 4 galaxy
bacterial_genomics 3 galaxy
computational-chemistry 3 galaxy
sequence-analysis 3 galaxy, nextflow
genome-assembly 2 galaxy
proteomics 2 galaxy, nextflow
quality-control 2 wdl
amplicon 1 galaxy
comparative_genomics 1 galaxy

Provenance

This benchmark merges execution-verified tasks from four independently maintained repositories:

  • IWC-bench (75 tasks): Galaxy Intergalactic Workflow Commission workflows with Planemo CI
  • nf-core-bench (45 tasks): nf-core Nextflow pipelines with nf-test snapshot verification
  • GTN-bench (14 tasks): Galaxy Training Network tutorial workflows verified on usegalaxy.org
  • WDL-bench (8 tasks): BioWDL/WARP/ENCODE WDL pipelines verified with miniwdl

Each task directory preserves all original files from its source repository. The rubric.json contains original_task_id and source fields for full traceability.

Citation

@article{hiltemann2023galaxy,
  title={Galaxy Training: A powerful framework for teaching!},
  author={Hiltemann, Saskia and others},
  journal={PLoS Computational Biology},
  volume={19},
  pages={e1010752},
  year={2023}
}

@article{ewels2020nf,
  title={The nf-core framework for community-curated bioinformatics pipelines},
  author={Ewels, Philip A and others},
  journal={Nature Biotechnology},
  volume={38},
  pages={276--278},
  year={2020}
}
Downloads last month
7