QSAR Toxicity Prediction (GWO-ANN)

OVERVIEW

Built an end-to-end ML pipeline to predict chemical toxicity against the estrogen receptor (NR-ER-LBD endpoint) using the Tox21 dataset, a severely imbalanced binary classification task (18.6:1 non-toxic to toxic). SMILES structures were converted into Morgan Fingerprints (ECFP4) and RDKit molecular descriptors, reduced to 500 final features via Mutual Information Selection. Class imbalance was handled with SMOTETomek (oversampling + undersampling). Three manually tuned ANN baselines were then compared against models whose hyperparameters were optimized automatically by the Grey Wolf Optimizer (GWO), a swarm-based metaheuristic, run across 5 random seeds to test stability. The best GWO model (seed 3103) reached an F1-Score of 0.5946, beating the best manual baseline (0.5778) while also shrinking the generalization gap.

WHAT I DID

• —Chemical feature extraction from SMILES via Morgan Fingerprint (ECFP4) and RDKit molecular descriptors, reduced to 500 features by Mutual Information Selection.
• —Class imbalance handled with SMOTETomek (combined over- and under-sampling) on an 18.6:1 imbalanced Tox21 endpoint.
• —Three manually tuned ANN baselines compared against models auto-tuned by the Grey Wolf Optimizer across 5 random seeds for stability.
• —Best GWO model (seed 3103) reached F1-Score 0.5946 versus 0.5778 for the best manual baseline, with a smaller generalization gap.