agentd-drug-discovery

<!--

COPYRIGHT NOTICE

This file is part of the "Universal Biomedical Skills" project.

Copyright (c) 2026 MD BABU MIA, PhD <md.babu.mia@mssm.edu>

All Rights Reserved.

#

This code is proprietary and confidential.

Unauthorized copying of this file, via any medium is strictly prohibited.

#

Provenance: Authenticated by MD BABU MIA

-->

---

name: agentd-drug-discovery

description: Use the AgentD workflow to mine evidence, design molecules, and rank candidates with SAR plus ADMET annotations for early drug discovery tasks.

allowed-tools:

• read_file
• run_shell_command

---

At-a-Glance

• description (10-20 chars): Hypothesis foundry
• keywords: ligand-design, SAR, ADMET, docking, ranking
• measurable_outcome: Generate ≥10 candidate molecules (or requested count) with SMILES, key properties, and rationales per run, all delivered within 15 minutes.

Inputs

• target_protein, optional reference_compound, disease indication.
• constraints dict (LogP, MW, TPSA, etc.) and num_candidates.

Outputs

1. Ranked candidate list with SMILES + property scores + novelty metrics.
2. ADMET/toxicity alerts and SAR rationale per molecule.
3. Reproducibility manifest (data source versions, model checkpoints).

Workflow

1. Evidence retrieval: Mine literature + databases for known ligands and liabilities.
2. Generate candidates: Run AgentD generative step (scaffold hopping/fragment growth) aligned to constraints.
3. Score & filter: Apply Lipinski/QED/ADMET heuristics; include docking setup when requested.
4. Rank & explain: Combine efficacy, developability, novelty; summarize SAR learnings.
5. Deliver outputs: Emit JSON/CSV plus narrative recommendations; mark as in silico.

Guardrails

• Clearly state outputs are hypothetical and need wet-lab validation.
• Flag PAINS/reactive motifs automatically.
• Record data/model versions for audit trails.

References

• Detailed parameter tables and dependencies listed in README.md.

<!-- AUTHOR_SIGNATURE: 9a7f3c2e-MD-BABU-MIA-2026-MSSM-SECURE -->