Cheminformatics & Drug Discovery
What was the Problem?
Traditional drug discovery is a slow, expensive process relying on physical testing of millions of chemical compounds.
How I Solved It
Built a machine learning pipeline using molecular fingerprints and docking simulations to rapidly predict compound bioactivity before physical synthesis.
Effort Reduced
Reduced the time needed to screen potential drug candidates by a factor of 10x, enabling researchers to focus only on the most promising compounds.
Profit / Impact Achieved
Cut computational and research costs significantly by avoiding fruitless physical synthesis trials.
Difficulties Faced
Ensuring the machine learning models were reproducible and didn't just memorize the training data (overfitting).
How I Overcame It
Applied rigorous cross-validation and Optuna hyperparameter search to ensure robust, reproducible research that generalizes to new chemical spaces.