Applied AI • Molecular Discovery • Model Interpretability

AI Drug Analysis & Toxicity Pipeline

A high-throughput machine learning pipeline designed to predict molecular properties, including drug toxicity, with high accuracy and granular SHAP interpretability.

Key Engineering Highlights

SHAP Explainability

Interpret model predictions with TreeSHAP visualisations for direct insights into molecular substructure contributions.

Partitioned Data Processing

Handle large chemical compound datasets efficiently with memory-optimised Parquet / PyArrow partitioning.

Precision Benchmark

Achieved RMSE 0.1577 and R² 0.99 on validation benchmarks.

Tech Stack

PythonXGBoostTreeSHAPPyArrowPandasScikit-Learn

Model Performance Metrics

RMSE
0.1577
R² Score
0.99

Feature attribution analysis isolates top molecular descriptors responsible for toxic chemical pathways.

View Repository on GitLab →