Computational D-Peptide Drug Design Skill

A Claude Code Skill for end-to-end computational D-peptide inhibitor design.

Quick Start

# 1. Clone and setup
git clone https://github.com/mingjianzhang20-glitch/computational-drug-design-skill.git
cd computational-drug-design-skill
conda env create -f environment.yml
conda activate drug_design

# 2. Download Boltz2 checkpoint (~500MB)
boltz setup --model boltz2

# 3. Run full pipeline
python examples/run_pipeline.py \
    --receptor YOUR_RECEPTOR_SEQUENCE \
    --peptides peptide_list.txt \
    --output results/

What it does

Step	Script	Description
1. Build SMILES	examples/smiles_builder.py	D-peptide sequence → SMILES
2. Predict IC50	examples/boltz2_predict.py	Boltz2 affinity prediction
3. MPO scoring	examples/mpo_analysis.py	LogP + TPSA + IC50 filter
4. MD simulation	examples/openmm_simulation.py	OpenMM 100-500ns MD
5. Full pipeline	examples/run_pipeline.py	End-to-end automation

Individual script usage

# Build D-peptide SMILES
python examples/smiles_builder.py --seq lgrmg
python examples/smiles_builder.py --seq ffflggqpyw --acetylated

# Predict IC50 with Boltz2
python examples/boltz2_predict.py \
    --receptor APTLFRL \
    --smiles "N[C@@H](CC(C)C)C(=O)..." \
    --name d-lgrmg --output results/

# MPO analysis
python examples/mpo_analysis.py \
    --input candidates.csv \
    --lead_tpsa 263.3 \
    --output mpo_results.csv

# MD simulation
python examples/openmm_simulation.py \
    --pdb complex.pdb \
    --output md_results/ \
    --ns 100

MPO Criteria

Criterion	Threshold	Rationale
C1: LogP	1 – 3	Membrane permeability
C2: TPSA	< lead peptide TPSA	More compact than reference
C3: IC50	< 1000 nM	Nanomolar potency

IC50 formula (Boltz2)

IC50_nM = 10**(-affinity_pred_value) * 1000

Tested targets

Target	Lead D-peptide	IC50
NDUFA9	d-LGRMG	45.2 nM
NOTCH1	d-SSQCF	574.2 nM
2VSM	d-GITLGGGS	44.8 nM

Requirements

• Python 3.10+
• Boltz2 checkpoint: ~/.boltz/boltz2_aff.ckpt
• GPU recommended for Boltz2
• See environment.yml for full dependencies