Skip to content

ClawBio Hackathons: Agentic Genomics

ClawBio Hackathon Banner

Hackathon #1: Imperial College London

19 March 2026 | Sir Michael Uren Hub, White City Campus

Our first hackathon brought together 70+ participants from across genomics, AI engineering, proteomics, clinical diagnostics, and epidemiology. In a single afternoon, attendees built and submitted 8 new bioinformatics skills as pull requests to ClawBio.

  • Presentation

    Opening session. Full room for the welcome talks at the Sir Michael Uren Hub.

  • Organisers presenting

    Manuel Corpas and Jay Moore introduce ClawBio and the docs.clawbio.ai hackathon guide.

  • Speaker presenting

    Guided tutorial. Walking through agentic tools and skill building, step by step.

  • Hacking session

    Build time. Laptops open, skills taking shape across all five tracks.

  • Breakout area

    Breakout hacking with sunset views over White City. Discussions on reproducibility, sensitive biodata, and the future of agentic genomics.

  • Jay Moore

    Jay Moore (Imperial) between mentoring rounds.

Group photo

Group photo at the end of the day.

By the numbers

Metric Value
Registrants 99+
Attended 70+
Skill PRs submitted 8
GitHub stars (day of event) 468
Forks 83
Repo views (14 days) 10,972
Unique cloners 645
Tracks 5 (AI engineers, genomics, proteomics, clinical, epi)

Skills submitted on the day

Skill Author Description
PubMed Summariser @emanueleriontino PubMed research briefing from gene or disease query
Protocols.io Bridge @camlloyd Search and retrieve lab protocols
Skill Builder @thepigdestroyer Scaffold new ClawBio skills from spec
Variant Annotation @toby-clark4 Annotate variants with ClinVar and gnomAD
Variant Annotation @HadiKhan-dev Alternative variant annotation approach
Bioconductor Bridge @HDash Doc-enriched Bioconductor package discovery
FHIR PGx @MarceloGal Fetch electronic clinical pharmacogenomics data
Clinical Trial Finder @Duvet05 Find clinical trials with multiple output formats

Institutions represented

Imperial College London, King's College London, The Crick Institute, UCL, QMUL, University of Westminster, Brunel University, PUCP Peru, FinalDose.ai, Canos.ai, Vivid-Dx, Flow.bio, FLock.io, Valink Tx, Inforcer

Watch the recording

Organised by Jay Moore (Imperial), Manuel Corpas (Westminster), Nathan Skene (Imperial), and Josh Beale.

Join the next one

We are planning more hackathons in 2026. To get notified:

Want to host a ClawBio hackathon at your institution? Open an issue or reach out on Discord.

Hackathon Guide

Everything below is the reusable guide for participating in any ClawBio hackathon. It covers setup, skill building, and submission.

The Problem

Modern bioinformatics knowledge is fragmented across papers, scripts, and private pipelines. Reproducing even simple analyses often requires reconstructing hidden decisions from incomplete documentation. Only about 1 in 4 computational biology papers can be reproduced without contacting the authors (Garijo et al., PLOS ONE 2013; Collberg and Proebsting, 2016). Meanwhile, general-purpose LLMs hallucinate gene-drug associations, use outdated clinical guidelines, and produce results with no audit trail.

What is ClawBio?

ClawBio proposes a different unit of knowledge: a skill that packages the code, the scientific assumptions, the test data, and the execution contract in one inspectable artefact.

Each skill includes:

  • A SKILL.md contract that explains the scientific decisions the tool makes (thresholds, databases, safety rules)
  • Demo data that anyone can run without their own files
  • A Python script with --input, --output, and --demo flags
  • A reproducibility bundle: commands.sh, environment.yml, and checksums.sha256

SKILL.md is not just documentation. It is the contract that tells humans and AI agents how the skill should be used, what assumptions it makes, and when it should refuse to run.

Who is this for?

You are... You already know You do not need to know Good first project
AI / software engineer APIs, Python, automation Any biology or genomics PubMed Summariser, Clinical Trial Finder
Genomics researcher VCFs, pipelines, domain expertise Agentic AI or SKILL.md Variant Annotation, QC Report
Proteomics / multi-omics Mass spec, protein databases ClawBio (no skills exist yet) Protein Interaction Mapper, Protein Domain Annotator
Clinical / diagnostics Variant classification, PGx Agent frameworks PGx Interaction Checker
Epidemiology / public health Population data, outbreak analysis Python scripting (helpers available) Vaccine Equity Scorer, GBD Visualiser

Quick Path

01 Beginner 10 min

Setup

Clone repo, install dependencies, run a demo.
02 Beginner 20 min

Your First Skill

Scaffold a skill, write SKILL.md, add demo data.
03 Intermediate 20 min

Add Python

Implement the skill logic with a CLI endpoint.
04 Beginner 10 min

Test and Submit

Validate, test, open a PR.

Most participants submit their first skill in 90 to 120 minutes. First-time Git or GitHub users should allow extra time; helpers will be available throughout.

Starter template: copy templates/SKILL-TEMPLATE.md into your skill directory to get the correct structure immediately. See Your First Skill for the full walkthrough.

Example completed skill: see the NutriGx Advisor PR by @drdaviddelorenzo, the first community contribution.

What counts as a successful submission?

  • [ ] One new skill directory under skills/
  • [ ] A SKILL.md with frontmatter (name, version, inputs, outputs) and three body sections (Domain Decisions, Safety Rules, Agent Boundary)
  • [ ] Synthetic demo data (never real patient data)
  • [ ] A runnable --demo command that produces output
  • [ ] A pull request to github.com/ClawBio/ClawBio

That is it. A focused, well-documented skill with clear domain decisions is better than an ambitious but incomplete one.

Three Themes

  1. Gaps in genomics tools: What analyses are still manual, brittle, or unreproducible? Build a skill that fills one of those gaps.
  2. Trustworthy agentic approaches: How do we make AI agents safe for biology? Encode domain decisions and safety rules into SKILL.md so agents execute with proven logic, not hallucinations.
  3. Accessible complexity: Genomics data is vast and complex. Build skills that present results so clearly that a non-specialist can act on them.

Resources

Two guided tutorials are available to get you started. Both paths converge at the same goal: a working skill you can submit as a PR.

Intermediate 15 min

Agentic Tools Tutorial

How AI coding agents work, setup options (GitHub Copilot, Claude Code, OpenCode), your first AI-assisted analysis, and context management tips.
Beginner 10 min

ClawBio Setup

Clone ClawBio, run a demo skill, then build your own following the step-by-step guide.

Pre-event setup

# Clone the repo and run a demo
git clone https://github.com/ClawBio/ClawBio.git
cd ClawBio
pip3 install -r requirements.txt
python3 skills/pharmgx-reporter/pharmgx_reporter.py --demo

To submit your work you will need the GitHub CLI (gh). Install it if you don't have it:

brew install gh          # macOS
sudo apt install gh      # Debian / Ubuntu
winget install GitHub.cli # Windows

Then authenticate: gh auth login

If you prefer not to install gh, you can submit your PR through the GitHub web interface instead. Both routes are covered in the Submit guide.

Choose Your Track

We have attendees ranging from AI agent engineers with no genomics background to researchers with 40+ years in computational biology. Pick the track that fits you. Each project is labelled as a 90-minute build (achievable in the afternoon) or a stretch build (ambitious, likely a prototype).

Track A: AI Engineers New to Genomics

You build agents, automation, and APIs professionally. You just haven't touched genomic data before. These skills let you apply your engineering strengths to biology using public APIs, no wet-lab knowledge required.

90-minute builds:

Stretch builds:

Track B: Genomics Researchers New to Agentic AI

You work with genomic data daily. Wrap your existing expertise into a skill.

90-minute builds:

Stretch builds:

Track C: Proteomics and Multi-Omics

ClawBio currently has zero proteomics skills. Build the first one.

90-minute builds:

Stretch builds:

Track D: Clinical and Diagnostic Applications

All clinical track outputs are educational prototypes only. They must not be used for clinical decision-making or patient management.

90-minute builds:

Stretch builds:

Track E: Epidemiology and Public Health

90-minute builds:

Stretch builds:

Judging Criteria

Criterion Weight
Domain correctness 40%
Reproducibility 25%
Usefulness 20%
Code quality 15%

Communication