Bio Admet Prediction

Version Compatibility

Reference examples tested with: RDKit 2024.09+, requests 2.31+, DeepChem 2.8+, chemprop 2.0+ (note major API change from 1.x), admet-ai 1.3+, pandas 2.2+.

Before using code patterns, verify installed versions match. If versions differ:

• Python: pip show <package> then help(module.function) to check signatures

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

ADMET Prediction

Predict absorption, distribution, metabolism, excretion, and toxicity properties of drug candidates. ADMET prediction underpins lead selection and de-risking; calibrated, applicability-domain-aware predictions distinguish a working filter from a costly false-confidence rejection. Modern best practice combines online services (ADMETlab 3.0 with uncertainty estimates), open-source models (chemprop D-MPNN), and rule-based filters (Lipinski / Veber / BBB heuristics) -- each with known failure modes.

For PAINS / Brenk / structural alerts, see chemoinformatics/substructure-search. For QSAR model building from in-house data, see chemoinformatics/qsar-modeling.

ADMET Model Taxonomy

Decision: For batch screening of <10k compounds with no in-house data, **ADMETlab 3.0** (free API, 119 endpoints, calibrated uncertainty) is the modern standard. For in-house QSAR on a specific endpoint with >500 measurements, train a chemprop D-MPNN + Morgan + MOE descriptors model (Liu et al. 2024 achieved AUC 0.956 on hERG with this combo).

Decision Tree by Scenario

OECD QSAR Principles (5 Pillars)

For regulatory-grade ADMET QSAR (REACH, ECHA, FDA submissions), models must satisfy:

1. Defined endpoint -- specific bioassay, units, conditions
2. Unambiguous algorithm -- reproducible model + code
3. Defined applicability domain (AD) -- where the model is valid
4. Appropriate statistical validation -- external test set, cross-validation
5. Mechanistic interpretation -- biological / chemical rationale

For non-regulatory work, AD assessment is still critical. The OECD's applicability domain is the workhorse: predictions outside the AD are unreliable, but operational AD measures (leverage, kNN, conformal prediction) often disagree.

Applicability Domain Methods

For deep-learning ADMET, conformal prediction (Bostrom et al. 2024) is becoming the standard.

ADMETlab 3.0 API

The current standard for free ADMET prediction. 119 endpoints across 6 categories; per-prediction uncertainty.

Goal: Predict 119 ADMET endpoints with uncertainty for a batch of SMILES using a hosted API.

Approach: POST batches of <=500 SMILES to ADMETlab 3.0 REST endpoint and parse the returned JSON into a per-compound endpoint DataFrame.

import requests
import pandas as pd

# ADMETlab 3.0 API base: https://admetlab3.scbdd.com
# Endpoints: /api/admet (full 119-endpoint batch), /api/wash (standardization),
# /api/single/admet (single SMILES), /api/render (visualization)
# See https://admetlab3.scbdd.com/apis/ for current API spec.
def admetlab_predict(smiles_list, endpoint='admet'):
    url = f'https://admetlab3.scbdd.com/api/{endpoint}'
    payload = {'smiles': smiles_list}
    response = requests.post(url, json=payload, timeout=120)
    response.raise_for_status()
    return pd.DataFrame(response.json())

smiles = ['CCO', 'c1ccc(C(=O)O)cc1', 'CC(=O)Oc1ccccc1C(=O)O']
results = admetlab_predict(smiles)  # POST batches of <=500 SMILES at a time

ADMETlab 3.0 endpoints (sample):

• Absorption: Caco-2 permeability (logPapp), HIA (%), Pgp inhibitor/substrate, MDCK
• Distribution: BBB+, PPB (%), VDss (L/kg), Fu (fraction unbound)
• Metabolism: CYP1A2/2C9/2C19/2D6/3A4 inhibitor / substrate
• Excretion: CL (mL/min/kg), T1/2 (h)
• Toxicity: hERG, AMES, hepatotoxicity (DILI), carcinogenicity, immunotoxicity, mutagenicity, respiratory, skin, eye, cardiotoxicity, mitochondrial, NR-AR, NR-ER, SR-MMP
• Drug-likeness: Lipinski, Veber, Ghose, Egan, Muegge, QED, SAscore

chemprop D-MPNN for Custom Endpoints

When in-house data is available, train a target-specific model. chemprop's D-MPNN architecture + atom/bond features + optional Morgan / MOE descriptors is the modern open-source SOTA.

Goal: Train a target-specific ADMET classifier or regressor on in-house bioassay data.

Approach: Use chemprop 2.x CLI with rdkit_2d_normalized descriptor features, 5-fold scaffold-balanced split, and 5-model ensemble for uncertainty estimation.

# chemprop 2.x CLI (current; 'chemprop train' with space + dashed args)
# chemprop train --data-path data.csv --task-type classification \
#                --save-dir model_dir --molecule-featurizers rdkit_2d_normalized \
#                --num-folds 5 --ensemble-size 5

# Or chemprop 2.x programmatic API (full programmatic API documented at chemprop.readthedocs.io)
# See chemoinformatics/qsar-modeling for the full chemprop 2.x training pipeline.

Key: 5-fold ensemble for calibrated uncertainty; D-MPNN + rdkit_2d_normalized typically outperforms either alone.

hERG Cardiotoxicity (Gold Standard Endpoint)

hERG (KCNH2) blockade causes QT prolongation, Torsades de Pointes, and is the #1 reason for late-stage drug attrition. ICH S7B and FDA require non-clinical assessment.

Postdoc-grade interpretation: A single-model probability > 0.5 is NOT a kill signal. Triangulate ADMETlab + ProTox + literature search for cardiotox liability. Consider the active concentration vs predicted hERG IC50 (10 uM threshold for "concern"); a hERG-positive compound at 100 nM active concentration may still be safe with sufficient hERG selectivity.

CYP Inhibition (DDI Risk)

5 CYP isoforms cover most clinically relevant DDIs:

PAINS, BRENK, REOS Filters

ADMET prediction is separate from structural alerts; combine. See chemoinformatics/substructure-search for PAINS/BRENK/REOS pattern catalogs.

from rdkit.Chem.FilterCatalog import FilterCatalog, FilterCatalogParams

def alerts(mol, catalogs=('PAINS_A', 'BRENK', 'ZINC')):
    params = FilterCatalogParams()
    for cat in catalogs:
        params.AddCatalog(getattr(FilterCatalogParams.FilterCatalogs, cat))
    catalog = FilterCatalog(params)
    hits = catalog.GetMatches(mol)
    return [h.GetDescription() for h in hits]

Lipinski / Veber / Drug-Likeness

See chemoinformatics/molecular-descriptors for full physchem table. Quick filter:

from rdkit.Chem import Descriptors, Lipinski, QED

def druglike_score(mol):
    mw = Descriptors.MolWt(mol)
    logp = Descriptors.MolLogP(mol)
    hbd = Lipinski.NumHDonors(mol)
    hba = Lipinski.NumHAcceptors(mol)
    tpsa = Descriptors.TPSA(mol)
    rotbonds = Lipinski.NumRotatableBonds(mol)
    qed = QED.qed(mol)

    lipinski_violations = sum([mw > 500, logp > 5, hbd > 5, hba > 10])
    veber_pass = rotbonds <= 10 and tpsa <= 140
    bbb_pfizer = tpsa <= 90 and mw <= 500 and hbd <= 3

    return {'MW': mw, 'LogP': logp, 'HBD': hbd, 'HBA': hba,
            'TPSA': tpsa, 'RotBonds': rotbonds, 'QED': qed,
            'Lipinski_violations': lipinski_violations,
            'Veber_pass': veber_pass, 'BBB_pfizer': bbb_pfizer}

Per-Tool Failure Modes

ADMETlab 3.0 -- out-of-distribution prediction

Trigger: Macrocycle, metal-coordinated, oligomer (peptide >5 AA), or PROTAC submitted.

Mechanism: ADMETlab training set is drug-like organic molecules. Predictions on PROTACs, macrocycles, peptides extrapolate.

Symptom: Uncertainty estimates near max; ambiguous probabilities (0.4-0.6).

Fix: Check uncertainty band; if interval is broad, do not trust point estimate. For PROTACs / macrocycles, prefer literature-derived experimental data.

hERG D-MPNN -- training data bias

Trigger: Compound is novel chemotype not in training set (drug-like but in unexplored region).

Mechanism: D-MPNN learns local chemical features; for genuinely new scaffolds, extrapolation is unreliable.

Symptom: Model predicts hERG- (false negative) for compound that experimentally inhibits.

Fix: Use ensemble + applicability-domain assessment (kNN distance, ensemble variance). If kNN distance to training set > P95, treat prediction as low-confidence.

CYP3A4 inhibitor + substrate ambiguity

Trigger: Model trained on either inhibitor OR substrate; predictions confused.

Mechanism: CYP3A4 inhibitors and substrates have similar SAR; many compounds are both.

Symptom: Both classes report > 0.5.

Fix: Two separate models (inhibitor model, substrate model); compounds that score high in both are flagged for in vitro confirmation.

SwissADME -- no API

Trigger: Wanting to batch programmatically.

Mechanism: SwissADME explicitly forbids automated access (TOS).

Symptom: No programmatic access; manual web upload only.

Fix: Use ADMETlab 3.0 API for programmatic batch (free, no TOS restrictions).

PAINS as a kill filter

Trigger: Treating PAINS_A match as a categorical exclusion.

Mechanism: PAINS is calibrated against HTS assay-interference; matches do NOT predict failed drug development.

Symptom: Library purged of valid leads (curcumin analogs, polyphenol natural products).

Fix: Flag PAINS for orthogonal-assay confirmation; do not exclude pre-emptively. See substructure-search for details.

Class-imbalanced AMES dataset

Trigger: Training/predicting AMES mutagenicity.

Mechanism: AMES public datasets have ~3:1 negative:positive imbalance; baseline F1 misleading.

Symptom: Model reports high accuracy but predicts negative for all.

Fix: Use AUC-ROC or balanced accuracy. Apply SMOTE / class-weighted loss in chemprop.

Reconciliation Across Models

When ADMETlab, ProTox, and chemprop disagree on hERG:

• All three predict hERG+ → high confidence, deprioritize
• Two predict hERG+, one hERG- → moderate confidence, plan in vitro patch-clamp
• Single positive (other two negative) → likely false positive of disagreeing model; verify chemotype is in distribution
• All three hERG- → low confidence (especially novel chemotype); still consider in vitro screen for clinical candidates

Common Errors

References

• Fu et al., Nucleic Acids Res. 52:W422 (2024) -- ADMETlab 3.0.
• Shan W, Jiang Y, Chen J, Qin H, Qin H, Cheng F. 2022. RSC Adv 12(7):3935-3942 -- hERG ML benchmarks (DOI 10.1039/D1RA07956E).
• Heid et al., J. Chem. Inf. Model. 64:9 (2024) -- chemprop 2.0 redesign.
• OECD, "OECD Principles for the Validation of QSAR Models" (2007).
• OECD, "(Q)SAR Assessment Framework" (2023).
• Capuzzi et al., J. Chem. Inf. Model. 57:417 (2017) -- PAINS reality check.
• Wager et al., ACS Chem. Neurosci. 1:435 (2010) -- Pfizer CNS MPO.
• Bickerton et al., Nat. Chem. 4:90 (2012) -- QED.

Related Skills

• chemoinformatics/molecular-descriptors - Compute drug-likeness physchem
• chemoinformatics/substructure-search - PAINS / BRENK / REOS filter
• chemoinformatics/qsar-modeling - Build custom QSAR for in-house data
• chemoinformatics/molecular-standardization - Canonicalize before prediction
• machine-learning/biomarker-discovery - Adjacent ML approaches
• clinical-databases/pharmacogenomics - Patient genotype overlay