ADMET Prediction Skill

Predict ADMET properties to prioritize compounds for development.

Quick Start

/admet "CC1=CC=C(C=C1)CNC" --full
/pk-prediction --library compounds.sdf --threshold 0.7
/toxicity-screen CHEMBL210 --include hERG,DILI,Ames

What's Included

Property	Prediction	Model
Absorption	Caco-2, HIA, Pgp	ML/QSAR
Distribution	VDss, PPB, BBB	ML/QSAR
Metabolism	CYP inhibition, clearance	ML/QSAR
Excretion	Clearance, half-life	ML/QSAR
Toxicity	hERG, DILI, Ames, mutagenicity	ML/QSAR

Output Structure

# ADMET Profile: CHEMBL210 (Osimertinib)

## Summary
| Property | Value | Status |
|----------|-------|--------|
| Drug-likeness | Pass | ✓ |
| Lipinski Ro5 | 0 violations | ✓ |
| VEBER | Pass | ✓ |
| PAINS | 0 alerts | ✓ |
| Brenk | 0 alerts | ✓ |

## Absorption
| Property | Prediction | Confidence |
|----------|------------|-------------|
| HIA | 98% | High |
| Caco-2 | 15.2 × 10⁻⁶ cm/s | High |
| Pgp substrate | Yes | Medium |
| F30% | 65% | Medium |

## Distribution
| Property | Prediction | Confidence |
|----------|------------|-------------|
| VDss | 5.2 L/kg | Medium |
| PPB | 95% | High |
| BBB | Yes | High |
| CNS MPO | 5.5 | Good |

## Metabolism
| Property | Prediction | Confidence |
|----------|------------|-------------|
| CYP3A4 substrate | Yes | High |
| CYP3A4 inhibitor | Yes | Medium |
| CYP2D6 inhibitor | No | High |
| CYP2C9 inhibitor | No | Medium |
| Clearance | 8.5 mL/min/kg | Low |

## Excretion
| Property | Prediction | Confidence |
|----------|------------|-------------|
| Renal clearance | 10% | Medium |
| Half-life | 48 hours | High |

## Toxicity
| Property | Prediction | Confidence |
|----------|------------|-------------|
| hERG inhibition | No | High |
| DILI | Concern | Medium |
| Ames mutagenicity | Negative | High |
| Carcinogenicity | Negative | Medium |
| Respiratory toxicity | No | Low |

## Recommendations
**Strengths**:
- Good oral bioavailability (65%)
- Brain penetration (BBB permeable)
- Low hERG risk

**Concerns**:
- DILI concern - monitor in preclinical studies
- CYP3A4 inhibition - potential DDIs

**Overall**: Good ADMET profile. Progress to in vivo PK.

Property Ranges

Drug-Likeness

Rule	Pass Criteria
Lipinski Ro5	≤ 1 violation
Veber	RotB ≤ 10, PSA ≤ 140 Ų
Egan	LogP ≤ 5, PSA ≤ 131 Ų
MDDR	MW ≤ 600, LogP ≤ 5

Absorption

Property	Good	Moderate	Poor
HIA	>80%	40-80%	<40%
Caco-2	>10	1-10	<1
F30%	>70%	30-70%	<30%

Distribution

Property	Good	Moderate	Poor
VDss	0.3-5 L/kg	<0.3 or >5	Extreme
PPB	<90%	90-95%	>95%
BBB	LogBB > 0.3	-0.3 to 0.3	< -0.3

Toxicity Alerts

Alert	Action
hERG inhibition	Cardiotoxicity risk
DILI positive	Hepatotoxicity risk
Ames positive	Mutagenicity risk
PAINS	Assay interference
Structural alerts	Investigate further

Running Scripts

# Full ADMET profile
python scripts/admet_predict.py --smiles "CC1=CC=C..." --full

# Batch prediction
python scripts/admet_predict.py --library compounds.sdf --output results.csv

# Specific properties
python scripts/admet_predict.py --smiles "..." --properties hERG,DILI,CYP

# Filter by criteria
python scripts/admet_filter.py --library compounds.sdf --rules lipinski,veber

Requirements

pip install rdkit

# Optional for advanced models
pip install deepchem admet-x

Reference

• reference/admet-properties.md - Detailed property reference
• reference/toxicity-alerts.md - Toxicity alerts reference
• reference/pk-models.md - PK prediction models

Best Practices

1. Use multiple models: Consensus predictions more reliable
2. Check confidence: Low confidence = experimental verification needed
3. Consider chemistry: Novel structures less reliable
4. Iterative design: Use predictions to guide synthesis
5. Validate early: Confirm key predictions experimentally

Common Pitfalls

Pitfall	Solution
Over-reliance on predictions	Experimental validation required
Ignoring confidence	Check model applicability domain
Single model only	Use consensus of multiple models
Ignoring chemistry	Novel scaffolds = uncertain predictions
Late-stage testing	Early ADMET screening saves time

Limitations

• Models are approximate: Errors common
• Novel chemistry: Less reliable for new scaffolds
• In vitro-in vivo gap: Predictions don't always translate
• Species differences: Human predictions based on animal data
• Complex mechanisms: Some toxicity not predicted