domains / science / adme-property-prediction / adme-property-predictor

adme-property-predictor

Skill card

combined

Both copies describe the same ADME prediction skill and share the Overview, Integration, and Absorption sections. However, the LeoYeAI copy contains material steps and constraints the aipoch copy lacks, specifically a detailed 'When to Use' section with explicit do-not-use cases (biologics, prodrugs, clinical dosing), a 'Common Issues and Solutions' troubleshooting section, and a 'Distribution' section with specific property tables and best practices. The aipoch copy contains a generic 'Workflow' and 'Audit-Ready Commands' section that the LeoYeAI copy does not. Because the LeoYeAI copy is a more complete and specific version of the domain logic, but the aipoch copy provides a different structural wrapper (audit commands), they are not identical reproductions. The LeoYeAI copy is the 'authentic' domain procedure, but the aipoch copy's specific audit commands are a distinct procedural element. Given the instruction to 'combine' when copies overlap but each has material steps the others lack, and the aipoch copy's 'Audit-Ready Commands' are not in the LeoYeAI copy, I will mark as combine to ensure the audit steps are preserved in the combined instruction, although the LeoYeAI copy is the primary source of the scientific logic.

Repositories

Other copies

Combined skill

---
name: adme-property-predictor
description: Predict ADME properties for drug candidates using cheminformatics models.
license: MIT
---

# ADME Property Predictor

## When to Use
- Screening compound libraries for drug-like properties.
- Prioritizing lead compounds based on predicted PK.
- Identifying ADME liabilities.

**Do NOT use when:**
- Exact PK parameters needed for dosing (use experimental studies).
- Biologics or prodrugs.
- Clinical dosing decisions (experimental validation required).

## Workflow
1. Confirm user objective and inputs.
2. Validate request scope.
3. Run `python scripts/main.py` with validated inputs.
4. Return structured results separating assumptions and deliverables.

## Audit-Ready Commands
```bash
python -m py_compile scripts/main.py
python scripts/main.py --help
```

## Core Capabilities

### 1. Absorption (A) Prediction
Predict HIA, Caco-2, Solubility, LogS, Lipinski, and Veber rules.

```python
from scripts.adme_predictor import ADMEPredictor
predictor = ADMEPredictor()
absorption = predictor.predict_absorption(smiles="CC(=O)Oc1ccccc1C(=O)O", properties=["all"])
```

**Best Practices:**
- Consider HIA and solubility together.
- Use both rule-based and ML-based predictions for consensus.

### 2. Distribution (D) Prediction
Predict Vd, PPB, BBB, and fu.

```python
distribution = predictor.predict_distribution(smiles="CC(=O)Oc1ccccc1C(=O)O", properties=["vd", "ppb", "bbb"])
```

**Best Practices:**
- High PPB may require higher doses.
- Low Vd indicates plasma retention.

## Integration
- **Upstream**: `chemical-structure-converter`, `lipinski-rule-filter`
- **Downstream**: `drug-candidate-evaluator`, `toxicity-structure-alert`