skills.rest
ClawBio
ClawBioOFFICIAL
Explore all skills by this author →
WebsiteGitHub

drug-repurposing-screen

Analyzes pooled viability compound screens to rank context-selective drug repurposing candidates.

1.1k|257|Updated Feb 25, 2026
Search Tags:#pharmacogenomics#drug repurposing#biomarker#selectivity#PRISM#viability screen#hit calling
One-click install
npx skills add https://github.com/ClawBio/ClawBio --skill drug-repurposing-screen
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: drug-repurposing-screen
Source: https://github.com/ClawBio/ClawBio/tree/main/skills/drug-repurposing-screen
Command: npx skills add https://github.com/ClawBio/ClawBio --skill drug-repurposing-screen

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires numpy, pandas, scipy, pyyaml, pyarrow.

What problem does it solve?

Analyzing a pooled compound-by-sample viability screen (QC, normalization, hit calling, selectivity classification, biomarker association, and candidate prioritization) is normally a multi-week manual project with no shared format or audit trail. This Skill turns a screen bundle plus an explicit objective YAML into a ranked, biomarker-supported repurposing shortlist in one reproducible run.

Core Features & Use Cases

  • Schema-driven ingest and QC: Accepts any bundle layout defined in schema.yaml, computes robust SSMD per sample-plate pair, and anchors viability to per-plate DMSO controls.
  • Hit calling and selectivity: Calls hits via viability and robust-z thresholds, then classifies compounds as inactive, context-selective, or broadly active using kill rates and the SAS bimodality coefficient against objective-defined target and off-target contexts.
  • Biomarker sweep and priority scoring: Runs Spearman associations across feature matrices (expression, methylation, etc.) with per-feature-type BH-FDR, then combines selectivity, biomarker strength, clinical phase, mechanism novelty, and phenocopy support into a weighted priority score.
  • Use Case: Given a PRISM-style screen of IBD organoids versus fibroblast references, produce a top-20 table of approved compounds that selectively kill the target context, with supporting biomarker evidence and a full reproducibility bundle.

Quick Start

Run the bundled offline demo by asking the agent to run the drug repurposing screen demo and write the report, tables, and parquet caches to an output directory.

Frequently Asked Questions about drug-repurposing-screen

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a drug repurposing screen analysis on my own data?▼

Prepare a bundle directory with readouts, treatment_info, and sample_info CSVs, plus a schema.yaml describing the layout and an objective.yaml defining target and off-target sample queries. Then run the script with --bundle, --schema, --objective, and --output flags, or use --demo for the bundled toy dataset.

What input format does a PRISM-style viability screen need?▼

The bundle needs a wide primary readout CSV (samples by wells), a treatment_info CSV mapping wells to compounds and control types, and a sample_info CSV with context columns. File names and column names are flexible because schema.yaml declares all paths and mappings.

Can I run the screen analysis without an objective.yaml file?▼

No. The skill refuses to run without an explicit objective defining target_context and off_target_context sample_info queries. This prevents silently defaulting to a cancer-centric framing when the screen may involve organoids, fibrosis lines, or other contexts.

How are context-selective compounds classified in viability screens?▼

Compounds are classified using kill rate, median viability, and the SAS bimodality coefficient. A compound is context_selective when its kill rate is between 0.15 and 0.7 with a bimodality coefficient of at least 0.55, indicating differential response across sample contexts.

Why is biomarker FDR computed separately per feature type?▼

BH-FDR is applied within each feature matrix because feature counts differ by orders of magnitude (e.g., ~450k methylation CpGs versus ~25k copy-number features). A single global correction would let large matrices dominate and mis-rank candidates.

What are the limitations of this viability screen pipeline?▼

Version 0.1 handles single-dose primary readouts only and does not fit dose-response curves at scale. It also does not score drug-target interactions independently of the screen or perform literature searches; those require separate skills.

Related Skills

Explore More Education & Research Skills→
stellaromicsstellaromics

bio-data-visualization-heatmaps-clustering

Generate clustered heatmaps from gene expression or omics matrices with annotations.

Official
Intermediate
patsnappatsnap
33

biomarker-investigation

Investigate biomarkers for a target disease across patents, literature, and clinical trials.

Official
Advanced
mims-harvardmims-harvard
1.6k

drug-repurposing

Identify drug repurposing candidates across disease targets, drugs, and safety data.

Official
Advanced
olaTechieolaTechie

dose-response

Perform dose-response meta-analyses with GLS trend estimation and restricted cubic splines.

Community
Advanced