메뉴
Expert in pharma-agents project - autonomous multi-agent system for molecular ML optimization. Use when working on experiments, understanding the crew architecture, or debugging agent runs.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
AI-driven patient-to-trial matching for precision medicine and oncology. Given a patient profile (disease, molecular alterations, stage, prior treatments), discovers and ranks clinical trials from ClinicalTrials.gov using multi-dimensional matching across molecular eligibility, clinical criteria, drug-biomarker alignment, evidence strength, and geographic feasibility. Produces a quantitative Trial Match Score (0-100) per trial with tiered recommendations and a comprehensive markdown report. Use when oncologists, molecular tumor boards, or patients ask about clinical trial options for specific cancer types, biomarker profiles, or post-progression scenarios.
Comprehensive computational validation of drug targets for early-stage drug discovery. Evaluates targets across 10 dimensions (disambiguation, disease association, druggability, chemical matter, clinical precedent, safety, pathway context, validation evidence, structural insights, validation roadmap) using 60+ ToolUniverse tools. Produces a quantitative Target Validation Score (0-100) with GO/NO-GO recommendation. Use when users ask about target validation, druggability assessment, target prioritization, or "is X a good drug target for Y?"
Construct and analyze compound-target-disease networks for drug repurposing, polypharmacology discovery, and systems pharmacology. Builds multi-layer networks from ChEMBL, OpenTargets, STRING, DrugBank, Reactome, FAERS, and 60+ other ToolUniverse tools. Calculates Network Pharmacology Scores (0-100), identifies repurposing candidates, predicts mechanisms, and analyzes polypharmacology. Use when users ask about drug repurposing via network analysis, multi-target drug effects, compound-target-disease networks, systems pharmacology, or polypharmacology.
Detect and analyze adverse drug event signals using FDA FAERS data, drug labels, disproportionality analysis (PRR, ROR, IC), and biomedical evidence. Generates quantitative safety signal scores (0-100) with evidence grading. Use for post-market surveillance, pharmacovigilance, drug safety assessment, adverse event investigation, and regulatory decision support.
Discover novel small molecule binders for protein targets using structure-based and ligand-based approaches. Creates actionable reports with candidate compounds, ADMET profiles, and synthesis feasibility. Use when users ask to find small molecules for a target, identify novel binders, perform virtual screening, or need hit-to-lead compound identification.
Retrieves chemical compound information from PubChem and ChEMBL with disambiguation, cross-referencing, and quality assessment. Creates comprehensive compound profiles with identifiers, properties, bioactivity, and drug information. Use when users need chemical data, drug information, or mention PubChem CID, ChEMBL ID, SMILES, InChI, or compound names.
| name | catalyst-pharma |
| description | Expert in pharma-agents project - autonomous multi-agent system for molecular ML optimization. Use when working on experiments, understanding the crew architecture, or debugging agent runs. |
You are an expert in the pharma-agents project - an autonomous multi-agent system for molecular property prediction optimization.
This system uses CrewAI to run autonomous ML optimization loops:
experiments/
├── bbbp/ # Blood-Brain Barrier Penetration (classification)
│ ├── baseline.json # Score, metric, direction config
│ ├── baseline_train.py # Reference baseline (never modified)
│ ├── train.py # Working copy (agents modify this)
│ └── memory.json # Experiment learnings
├── solubility/ # ESOL solubility (regression)
│ └── ...
{
"score": 0.8951,
"metric": "ROC_AUC",
"direction": "higher_is_better" // or "lower_is_better" for RMSE
}
.worktrees/<experiment>/run_XXX/run/<experiment>/<number> (e.g., run/bbbp/001)| File | Purpose |
|---|---|
src/pharma_agents/main.py | Entry point, iteration loop |
src/pharma_agents/crew.py | Agent definitions |
src/pharma_agents/memory.py | Persistent learning, metric helpers |
src/pharma_agents/tools/custom_tools.py | WriteTrainPy, RunTrainPy, CodeCheck |
experiments/<name>/baseline.json | Experiment config |
/run <experiment> <iterations> - Start an optimization run/discard <run> -e <experiment> - Remove a failed/stuck run/promote <run> -e <experiment> - Promote successful run as new baselinefrom pharma_agents.memory import is_better, compute_improvement_pct
if is_better(new_score, old_score): # Handles both higher/lower is better
print(f"Improved by {compute_improvement_pct(old_score, new_score):.1f}%")
PHARMA_EXPERIMENT - Which experiment to run (bbbp, solubility)PHARMA_EXPERIMENTS_DIR - Override experiments path (for worktrees)MAX_ITERATIONS - Number of optimization iterationsget_metric_name(), is_better() not raw comparisonsgit branch -a | grep run/<exp>/