admet-prediction

huifer's avatarfrom huifer

ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) predictionfor drug candidates. Use for assessing drug-likeness, PK properties, andsafety risks early in drug discovery.Keywords: ADMET, PK, toxicity, drug-likeness, DILI, hERG, bioavailability

⭐0stars🔀0forks📁View on GitHub🕐Updated Jan 4, 2026
[Academic Research]

When & Why to Use This Skill

The ADMET Prediction skill is a specialized tool for pharmaceutical research designed to evaluate the Absorption, Distribution, Metabolism, Excretion, and Toxicity profiles of drug candidates. By utilizing advanced ML and QSAR models, it enables researchers to assess drug-likeness, pharmacokinetic (PK) properties, and potential safety risks early in the drug discovery pipeline, significantly reducing the cost and time associated with late-stage failures.

Use Cases

  • Lead Optimization: Prioritize chemical compounds by predicting key parameters like oral bioavailability (HIA, Caco-2) and blood-brain barrier (BBB) permeability.
  • Early Safety Screening: Identify high-risk candidates by screening for cardiotoxicity (hERG inhibition), hepatotoxicity (DILI), and mutagenicity (Ames test) before moving to in vivo studies.
  • High-Throughput Library Filtering: Process large SDF chemical libraries to filter out compounds that violate medicinal chemistry rules such as Lipinski’s Rule of Five, Veber, or PAINS alerts.
  • Metabolic Profiling: Predict CYP450 enzyme inhibition and clearance rates to anticipate potential drug-drug interactions and determine half-life estimates.
nameadmet-prediction
description|
KeywordsADMET, PK, toxicity, drug-likeness, DILI, hERG, bioavailability
categoryDMPK
tags[admet, pk, toxicity, drug-likeness, safety]
version1.0.0
authorDrug Discovery Team

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

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