Chiranjib Sarkar

The paper introduces LGNM, a physics-constrained graph neural network that improves protein flexibility prediction by learning per-residue flexibility coefficients, achieving higher accuracy than mENM on test proteins.

Deep-Interact Studio allows users to build, train, and compare custom deep learning models for predicting biomolecular interactions, with interpretability and support for multiple interaction types.

MoE-Bind is a sparse Mixture-of-Experts model for protein binder generation that outperforms dense baselines while reducing compute requirements.

ARACoFusion predicts protein-protein interactions in Arabidopsis thaliana using a deep learning model that improves accuracy and reliability through uncertainty-aware techniques and benchmarking against existing predictors.