Pharma Drug R&D Skill Suite

A public collection of reusable ChatGPT skills for computational and digital drug research and development.

This repository packages workflow-oriented skills, not third-party model weights or upstream codebases. The goal is to provide reusable skill specifications that help ChatGPT support common tasks across target discovery, repurposing, safety review, structure preparation, virtual screening, ADMET modeling, de novo design, clinical study operations, lab knowledge capture, and now CMC / GMP / eCTD readiness.

Why this repository exists

Drug R&D work is fragmented across evidence synthesis, structure-based design, molecular machine learning, clinical operations, quality systems, CMC documentation, and regulatory submission packaging. This repository turns those recurring workflows into reusable ChatGPT skills so teams can:

• standardize reasoning and deliverable formats
• shorten setup time for repeated analysis tasks
• make project handoffs more consistent
• connect open-source drug discovery tooling with clear task-oriented prompts
• keep regulatory-oriented skills grounded in official agency sources

Scope

This repository currently focuses on:

• target discovery and validation triage
• repurposing hypothesis screening
• safety, pharmacogenomics, and labeling review
• protein structure planning
• virtual screening orchestration
• ADMET and property modeling
• target-conditioned de novo molecule design planning
• clinical study operations planning
• laboratory knowledge capture and reproducibility
• CMC dossier planning
• GMP / cGMP readiness assessment
• eCTD publishing readiness for FDA and NMPA

This repository does not currently provide production-ready wet-lab protocols, regulated GxP execution systems, validated submission documents, or bundled copies of upstream project code.

Repository structure

.
├── README.md
├── README_CN.md
├── OPEN_SOURCE_SOURCES.md
├── GITHUB_LANDSCAPE_CMC_GMP_ECTD.md
├── REGULATORY_SOURCES_FDA_NMPA.md
├── LICENSE
├── CONTRIBUTING.md
├── CITATION.cff
├── target-evidence-triage/
├── repurposing-hypothesis-screen/
├── safety-pgx-brief/
├── structure-folding-planner/
├── virtual-screening-orchestrator/
├── admet-property-modeler/
├── de-novo-molecule-generator/
├── clinical-study-ops/
├── lab-knowledge-capture/
├── cmc-dossier-planner/
├── gmp-quality-readiness/
└── ectd-publishing-readiness/

Included skills

Stage	Skill	What it does
Target discovery / validation	target-evidence-triage	ranks targets using disease, mechanism, druggability, and translational evidence
Repurposing	repurposing-hypothesis-screen	screens approved or existing assets for indication expansion
Safety translation	safety-pgx-brief	combines ADR, DDI, PGx, and labeling risks into a decision brief
Structure preparation	structure-folding-planner	plans target structure modeling and downstream docking readiness
Virtual screening	virtual-screening-orchestrator	specifies reproducible docking and screening runbooks
ADMET / QSAR	admet-property-modeler	plans molecular property and toxicity modeling workflows
De novo design	de-novo-molecule-generator	defines constraint-driven generative chemistry campaigns
Clinical operations	clinical-study-ops	turns study concepts into EDC and data-management setup plans
Lab operations	lab-knowledge-capture	standardizes experiment records, traceability, and handoffs
CMC / quality dossier	cmc-dossier-planner	builds phase-appropriate FDA and NMPA CMC gap reviews and module maps
GMP / cGMP readiness	gmp-quality-readiness	assesses quality-system readiness, inspection risk, and remediation priorities
eCTD / publishing	ectd-publishing-readiness	organizes regional eCTD assembly, validation risk, and sequence planning

Referenced open-source projects

The following public projects informed the task boundaries, workflow patterns, and project selection in this repository. Their code is not vendored here unless explicitly stated in the future.

Project	Repository	How it informs this repository
DrugClaw	https://github.com/DrugClaw/DrugClaw	evidence synthesis, target intelligence, literature, compound triage, and docking-oriented AI assistant concepts
OpenFold	https://github.com/aqlaboratory/openfold	protein structure prediction planning and model-readiness workflows
PaddleHelix	https://github.com/PaddlePaddle/PaddleHelix	broader bio-computing coverage across structure, DTI, molecular generation, and ADMET-related tasks
DockM8	https://github.com/DrugBud-Suite/DockM8	consensus docking and virtual screening workflow design
EasyDock	https://github.com/ci-lab-cz/easydock	automated docking pipelines and distributed execution patterns
Chemprop	https://github.com/chemprop/chemprop	message-passing neural network workflows for molecular property prediction
DeepChem	https://github.com/deepchem/deepchem	general-purpose open-source deep learning workflows for drug discovery and chemistry
TorchDrug	https://github.com/DeepGraphLearning/torchdrug	graph learning workflows for drug discovery and molecular ML research
DrugGEN	https://github.com/HUBioDataLab/DrugGEN	target-conditioned de novo molecule generation campaign design
OpenClinica	https://github.com/OpenClinica/OpenClinica	open clinical EDC/CDM workflow concepts for study operations
eLabFTW	https://github.com/elabftw/elabftw	electronic lab notebook, traceability, and experiment knowledge capture patterns
AI4Green	https://github.com/AI4Green/AI4Green	chemistry-oriented ELN and sustainability-aware laboratory workflow ideas
OpenQMS / open-eQMS / beCPG / RConsortium submission examples	public GitHub repositories	adjacent quality-system or submission examples used only as ecosystem references, not as normative regulatory sources

Regulatory grounding for CMC / GMP / eCTD skills

The repository now includes CMC, GMP, and eCTD-oriented skills that are intentionally grounded in official FDA and NMPA/CDE documents first.

• cmc-dossier-planner
• gmp-quality-readiness
• ectd-publishing-readiness

These three skills were added only after checking the public GitHub landscape and finding that mature, reusable SKILL.md-style skills for pharmaceutical CMC, GMP, and eCTD are not readily available. Public GitHub contains adjacent tools and examples, but not a trustworthy ready-made skill layer for these regulated tasks.

See:

• GITHUB_LANDSCAPE_CMC_GMP_ECTD.md
• REGULATORY_SOURCES_FDA_NMPA.md

Attribution and citation

If you use this repository in research or internal enablement work:

1. cite or acknowledge the specific upstream open-source projects that informed your workflow;
2. do not imply endorsement by those upstream maintainers;
3. verify license compatibility before copying code, models, or assets from upstream repositories;
4. treat FDA and NMPA official documents as the primary authority for CMC, GMP, and eCTD work.

See OPEN_SOURCE_SOURCES.md for the open-source provenance summary and CITATION.cff for repository citation metadata.

How to use

1. Browse the repository

Open the skill folder that matches the stage of your program.

2. Read the skill specification

Each skill exposes its own SKILL.md with:

• trigger conditions
• workflow steps
• output structure
• handoff rules to other skills

3. Upload or adapt the skill

Use the skill content as a reusable workflow specification in ChatGPT or adapt it for your internal deployment process.

Suggested skill bundles

• Early discovery: target-evidence-triage + repurposing-hypothesis-screen + safety-pgx-brief
• Structure-based design: structure-folding-planner + virtual-screening-orchestrator + admet-property-modeler
• Generative design: de-novo-molecule-generator + virtual-screening-orchestrator + admet-property-modeler
• Translation and execution: clinical-study-ops + lab-knowledge-capture
• Regulatory quality readiness: cmc-dossier-planner + gmp-quality-readiness + ectd-publishing-readiness

Design principles

• Workflow-first: optimize for repeatable decision workflows rather than one-off prompts.
• Source-aware: keep upstream project provenance visible.
• Agency-first for regulated tasks: use official FDA and NMPA sources before GitHub examples.
• Non-vendoring by default: avoid copying third-party code into this repository unless licensing and maintenance are explicit.
• Composable skills: each skill should stand alone but also hand off cleanly to adjacent stages.
• Transparent boundaries: separate evidence generation, model planning, operational execution, and publishing readiness.

Limitations

This version still does not yet include dedicated skills for:

• enterprise-specific QMS integrations
• manufacturing execution systems (MES)
• validation master plan authoring
• data-integrity forensics and audit-trail reconstruction
• eCTD sequence automation scripts or live publisher integrations

Those areas usually require organization-specific SOPs, validated templates, connected systems, or product-specific process knowledge.

Contributing

Issues and pull requests are welcome for:

• better skill wording
• improved stage coverage
• more precise upstream attribution
• new open-source project mappings
• additional deployment metadata such as agents/openai.yaml
• more explicit official-source mapping for FDA and NMPA skills

Please review CONTRIBUTING.md before opening a pull request.

License

This repository is licensed under the MIT License. See LICENSE for the full text.