Open repositories & reusable skills
Eight open-source repositories from the hackathon — a disease-agnostic drug-program pipeline, target-discovery and epitope-mapping tools, market and peer-review skills, and a scientific-media engine. Most are packaged as Claude Science skills you can drop in and run.
Therapeutic program tooling
therapeutic-program-architect
The flagship specialist & skill: an end-to-end, human-in-the-loop drug-development pipeline that is archetype-adaptive across common and genetic/rare disease. Give it a disease; it runs 10 stages to a fundable plan, stopping at every consequential gate.
Featured — full walkthrough & install →
patient-centered-market-and-survey
Reusable Claude Science skill: build a patient-centered therapeutic market report and design a patient-community survey. Disease-agnostic, with an EoE worked example.
synthetic-peer-review
Run a synthetic (mock) peer-review panel on a manuscript before submission — distinct expert reviewers plus a handling-editor synthesis.
Target discovery & antigen design
omics-target-mining
A reproducible pipeline for mining public omics datasets (GEO, ArrayExpress, PRIDE, CELLxGENE) to nominate and prioritize therapeutic targets — from “here is a disease” to a ranked, druggability-annotated shortlist.
Ten-stage workflow — details below
antigen-epitope-pipeline
A disease-agnostic pipeline mapping antigen proteins to HLA-restricted epitopes and per-HLA presentation burden — with TG2-deamidation modeling and known-epitope benchmarking.
inflammatory-target-discovery-skills
A disease-agnostic methodology stack for computational target discovery and antigen-specific therapeutic design in inflammatory / immune disorders (EGIDs, celiac, IBD, autoimmune).
Communication & media
claude-science-skills
Reusable Claude Science skills for scientific molecular- and cell-scale video generation (ChimeraX + Blender / Modal) — the engine behind the structural fly-around renders on this site.
podcast-debrief-skill
Turn a project's data artifacts into a fact-grounded, two-voice podcast plus a local text-to-speech render package.
omics-target-mining — ten reproducible stages
From a disease name to a ranked, druggability-annotated target shortlist, using only public data. Each stage
ends by saving artifacts, so a run is fully auditable. Two core statistics (moderated_de,
meta_analysis) are dependency-light and usable stand-alone.
- Dataset inventory Query GEO / ArrayExpress / PRIDE / CELLxGENE; build one master table; audit for keyword-only false positives.
- Tier & prioritize Score datasets on relevance, analytical role, and tractability (Tier 1 bulk case-control → Tier 3 complementary).
- Retrieve & QC bulk Harmonize IDs to gene symbols, normalize, PCA + canonical-marker sanity checks.
- Per-study differential expression Variance-moderated Welch t-test with BH-FDR.
- Cross-study meta-analysis DerSimonian–Laird random-effects pooling with heterogeneity stats.
- Single-cell validation scanpy QC → HVG → Harmony → Leiden → composition shift + per-cell-type DE.
- Pathway / regulator / cell-cell communication Enrichr (GO_BP, Reactome, Hallmark, TF libraries) + LIANA.
- Multi-omics integration Per-gene evidence table across independently-varying layers.
- Druggability Open Targets GraphQL: tractability, clinical candidates, subcellular location, target class.
- Consolidate A Tier-A shortlist with mechanism rationale and a suggested modality per target.
Try the helpers directly
from kernel import moderated_de, meta_analysis
# per-study DE on log2 expression (genes x samples)
de = moderated_de(expr, labels, case="disease", control="control")
# pool per-study effects (log2FC and SE across studies)
meta = meta_analysis(fc_df, se_df, min_studies=3)