Pranjal Kalekar

Research Engineer @Harvard Medical School and Massachusetts General Hospital

Boston, MA, US

Developed a custom graph-based imputation algorithm to address sparsity in protein expression data by leveraging protein-protein interaction networks, achieving a Root Mean Squared Error (RMSE) of 0.07, significantly outperforming baseline methods. • Trained and benchmarked XGBoost regression and classification models to capture complex nonlinear feature interactions, resulting in a 40% performance improvement over finely tuned linear models in predicting key biological outcomes. • Diagnosed and resolved instability issues in Graph Neural Network (GNN) regression models exhibiting exploding gradients by implementing architectural changes, He initialization, and systematically tuning loss functions and learning rates, ultimately restoring performance beyond benchmark levels.