# Molecule Design

> ---name: molecule-evolution-agent

- Skill: `majiayu000/molecule-design` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add majiayu000/molecule-design`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majiayu000/molecule-design/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: majiayu000 (https://skillmd.com/u/majiayu000)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/majiayu000/molecule-design

---


---name: molecule-evolution-agent
description: Evolve Molecules
keywords:
  - drug-discovery
  - de-novo-design
  - smiles
  - medicinal-chemistry
  - evolution
measurable_outcome: Generates a molecule candidate with an improved docking score after 3 generations of evolution.
license: MIT
metadata:
  author: Artificial Intelligence Group
  version: "1.0.0"
compatibility:
  - system: Python 3.10+
allowed-tools:
  - run_shell_command
---"

# Molecule Evolution Agent

The **Molecule Evolution Agent** acts as an autonomous medicinal chemist. It takes a starting molecule (or uses a default like Aspirin) and iteratively modifies its structure to optimize binding for a specific protein target.

## When to Use This Skill

*   **Lead Optimization**: When you have a hit molecule and want to improve its potency.
*   **De Novo Design**: To explore chemical space around a target protein.
*   **Idea Generation**: To get creative structural modifications suggested by an LLM.

## Core Capabilities

1.  **SMILES Manipulation**: Reads and writes chemical structures in SMILES format.
2.  **LLM Chemist**: Uses an LLM to suggest chemically valid modifications (e.g., "Add a fluorine group to the ring").
3.  **Mock Scoring**: (Currently) Uses a mock scoring function to simulate docking affinity.

## Workflow

1.  **Input**: Target Protein Name (e.g., "GPRC5D").
2.  **Process**:
    *   Start with a seed molecule.
    *   Loop for *N* generations.
    *   Ask LLM for a modification.
    *   Score the new molecule.
    *   Keep the best candidate.
3.  **Output**: Top candidate SMILES and the evolution history.

## Example Usage

**User**: "Design a better binder for GPRC5D."

**Agent Action**:
```bash
python3 Skills/Drug_Discovery/Molecule_Design/evolution_agent.py
# (Note: The script currently defaults to GPRC5D, but can be extended for arguments)
```