Turn any AI coding assistant into a computational biologist's research partner
📖 Full documentation ——
quick start, skill catalogue, methodology references, and troubleshooting
What is this
A curated collection of Agent Skills for computational biology. Every skill is a standardized
SKILL.md package that gives Claude Code, Cursor, Codex, and any other host supporting the
open Agent Skills standard domain expertise, reproducible workflows, and
validation checks — instead of another generalist that merely knows how to call an API.
Coverage spans transcriptomics, single-cell analysis, protein structure, molecular dynamics, protein
docking, inter-chain contact analysis with PyMOL visualization, antimicrobial peptide design, ML
peptide generation, literature mining, NCBI BioProject metadata, and publication-grade figures.
Refactored from a private archive of 217 real research scripts into a discoverable,
triggerable, reproducible skill library that follows an open standard. The per-script mapping —
what was ported, what was rewritten, and why — is documented in the
migration map.
Why
Most AI assistants can write scanpy code. They do not know:
- how much mitochondrial gene percentage to drop from your single-cell data before the filter starts
eating real cells
- why DESeq2's
padj suddenly came back all NA — almost always a mis-specified design
- which flexible loop the first spike in an RMSF curve corresponds to, and what experiment to run on it
- which of 3,000 retrieved papers are genuinely target discovery rather than functional validation
- where to truncate the axes of a reviewer-requested ROC curve so it does not invite criticism
Those are the things this project carries. Each skill ships a references/methodology.md
recording the scientific reasoning behind parameter choices, a references/troubleshooting.md
recording root causes for common failures, and a Validation section in SKILL.md describing how to
check whether your results are right.
Quick start
Option 1 — install as Claude Code skills
git clone https://github.com/ssinder/bioinformatics-agent-skills.git
cp -r bioinformatics-agent-skills/skills/* ~/.claude/skills/
Restart Claude Code and the skills are discovered automatically. Ask it to "run a bulk RNA-seq
differential expression analysis" and it will load the bulk-rnaseq skill on its own.
Option 2 — use as a Python package
pip install -e .
biopipe --list # list every skill
biopipe bulk-rnaseq --help # usage for one skill
Option 3 — call the scripts directly
export PYTHONPATH="$PWD/packages"
# peptide physicochemical properties
python skills/peptide-design/scripts/analyze_peptides.py \
--input peptides.fasta --output results/peptide_stats.csv
# ROC curve
python skills/scientific-figures/scripts/make_figures.py roc \
--input metrics.csv --output figures/roc.png
# literature search
python skills/literature-search/scripts/search_literature.py \
--topic "antimicrobial peptide delivery" --source pubmed --max-results 50
Available skills
🧬 Transcriptomics
| Skill | Description | Key capabilities |
|---|
bulk-rnaseq | RNA-seq quantification and differential expression | FastQC QC · Salmon quantification · DESeq2 differential analysis · GO/KEGG enrichment |
ena-download | ENA raw data download | Portal API FASTQ retrieval · resumable transfers · MD5 verification · metadata export |
single-cell-rnaseq | Standard single-cell RNA-seq workflow | QC filtering · CP10K normalization · HVG · PCA · Leiden clustering · UMAP · cell annotation |
🔬 Structural biology
| Skill | Description | Key capabilities |
|---|
protein-structure | PDB structure parsing and secondary structure | Built-in PDB parser · DSSP secondary structure · geometry computation · UniProt metadata |
protein-interactions | Protein-ligand interactions | PLIP XML parsing · residue-level extraction · interaction networks · PyMOL annotations |
molecular-dynamics | Molecular dynamics simulation and trajectory analysis | RMSD/RMSF/PCA · hydrogen bonds · Ramachandran · trajectory clustering · built-in DCD parser |
alphafold-db | AlphaFold DB structure download | API link resolution · batch download · pLDDT confidence filtering · metadata export |
protein-docking-hdock | HDOCK protein docking | HTTP job submission · Docking Score · confidence formula · model download · queue-status distinction |
protein-docking-lightdock | LightDock local protein docking | GSO swarm optimization · DFIRE scoring parsing · best-complex assembly · resumable batch runs |
pymol-structure-visualization | Inter-chain residue contacts and PyMOL visualization | Grid spatial partitioning · hydrogen bond/salt bridge/hydrophobic/π-π/cation-π classification · hotspot residues · Mermaid network diagrams · PyMOL script generation |
🧪 Protein engineering
| Skill | Description | Key capabilities |
|---|
peptide-design | Antimicrobial peptide design and screening | Physicochemical properties · hydrophobicity/pI · stability · MMseqs2 clustering for redundancy removal |
ml-peptide-generation | ML peptide generation and screening | PepMLM masked language model · illegal residue filtering · pseudo-perplexity ranking · reproducible seeds |
🧬 Cheminformatics
| Skill | Description | Key capabilities |
|---|
bio-database-fetch | PubChem compound and taxonomy lookup | PUG-REST batch properties · CAS number extraction · full taxonomy lineage · no browser dependency |
📚 Literature mining
| Skill | Description | Key capabilities |
|---|
literature-search | Multi-source scholarly literature search | PubMed · OpenAlex · Semantic Scholar · DOI deduplication · RIS/BibTeX export |
bioproject-search | NCBI BioProject metadata retrieval | E-utilities API · title/description/species · keyword filtering · CSV/JSON export |
paper-fulltext-extract | Structured full-text extraction from PDFs | GROBID to TEI · section tree/figure/reference parsing · JSON/MD/HTML output · corpus search · lead sentences |
📊 Scientific visualization
| Skill | Description | Key capabilities |
|---|
scientific-figures | Publication-grade figure generation | ROC · confusion matrices · heatmaps · volcano plots · SHAP summaries · unified publication styling |
Design principles
Self-contained — core algorithms ship as pure Python implementations. No scanpy installed? The
single-cell workflow still runs. No OpenMM? DCD trajectories still parse, and RMSD/PCA still compute.
Dependencies are optional; capability is not.
Reproducible — fixed random seeds, recorded parameters, pinned versions, and run metadata written
alongside the results.
Verifiable — every skill defines its own quality-control stage, so there is an explicit way to
check whether the output is correct.
No hardcoded paths — all paths are parameterized and all output follows a standard directory
convention. Not a single /Users/xxx/... is baked into the code.
Zero setup — pip install -e . is enough. No cloud services, API keys, or databases to configure.
Layout
bioinformatics-agent-skills/
├── skills/ # 17 independently installable skills
│ ├── bulk-rnaseq/
│ │ ├── SKILL.md # trigger description + workflow + validation
│ │ ├── scripts/ # executable scripts
│ │ └── references/ # methodology + troubleshooting guide
│ └── ...
├── packages/
│ ├── biopipe_core/ # shared layer: config/logging/HTTP retry/CLI/IO
│ └── biopipe/ # unified CLI entry point and skill registry
├── scripts/
│ ├── validate_skills.py # skill spec validator (CI core)
│ └── init_skill.py # skill scaffolding generator
├── tests/ # unit tests and validator self-tests
├── docs/ # usage guides and skill catalogue
└── .github/workflows/ # 7 CI pipelines
Development
git clone https://github.com/ssinder/bioinformatics-agent-skills.git
cd bioinformatics-agent-skills
pip install -e ".[dev]"
python -m pytest tests -v # run the test suite
python scripts/validate_skills.py # validate every skill
python scripts/validate_skills.py --strict # treat warnings as failures
See CONTRIBUTING.md before adding a new skill.
CI pipelines
| Workflow | Purpose |
|---|
skill-spec-validation | Validates that every skill conforms to the Agent Skills spec |
script-syntax | Multi-version Python syntax check + Ruff linting |
tests | Unit tests + --help smoke test for every skill |
security-scan | Dependency vulnerability audit + hardcoded credential scan |
pr-skill-scan | Targeted validation and reporting for skills changed in a PR |
pages | Builds and deploys the documentation site |
release | Automated release packaging |
Citation
If you use this project in your research, please cite:
@software{bioinfoagent2026,
title = {BioinfoAgent Skills: Agent Skills for Computational Biology},
author = {sinder},
year = {2026},
url = {https://github.com/ssinder/bioinformatics-agent-skills}
}
See CITATION.cff for details.
Contributing
Contributions of new skills, bug fixes, and documentation improvements are welcome. See
CONTRIBUTING.md. Read docs/SKILL_SPEC.md before submitting a
new skill.
Participation in this project implies agreement with the Code of Conduct.
License
MIT © sinder. See LICENSE.
Disclaimer
This project provides analysis and computation tooling. Its outputs require independent verification
by the researcher and do not constitute medical, clinical, or regulatory advice.
Star this project ★ — help more
researchers discover these reproducible research skills.
