Drug Discovery & QSAR

Industry blog and notebook series covering QSAR best practices, dataset curation, ADMET modelling, and ML model validation in drug discovery.

Step-by-step notebooks covering molecular fingerprinting, QSAR modelling, protein-ligand interaction, and virtual screening with DeepChem.

TDC's overview of 22 ADMET prediction tasks with dataset descriptions, baselines, and evaluation protocols.

Reference benchmarking code for MoleculeNet datasets (ESOL, BACE, BBBP, Tox21) with random and scaffold splits.

Notebooks demonstrating VS enrichment metrics (EF, BEDROC, AUROC) and scaffold-aware evaluation of QSAR models.

Comprehensive review of QSAR modelling: representations, algorithms, validation, and applicability domain.

Introduces Chemprop (D-MPNN) and benchmarks it against fingerprint-based methods on MoleculeNet.

Critical assessment of generative models for drug design — validity, novelty, and clinical relevance.

Directed MPNN for molecular property prediction. State-of-the-art on MoleculeNet benchmarks. Minimal code required to fine-tune on new datasets.

Unified benchmark platform: 66+ tasks covering ADMET, drug-target interaction, and molecular generation.

EBI's curated database of 2.4M bioactive molecules with assay data. The primary source for bioactivity training data.

Free web tool predicting ADME properties (logP, solubility, BBB penetration, Lipinski rules) from SMILES.

The standard collection of drug discovery benchmarks: ESOL, BBBP, BACE, Tox21, SIDER, and more — with scaffold splits and evaluation protocols.

66+ ADMET, drug-target interaction, and molecular generation benchmarks in a unified Python API. The most comprehensive drug discovery dataset collection.

2.4M bioactive molecules with curated assay data across 15k targets. The primary raw source for QSAR model building in drug discovery.