Active learning framework for high-throughput virtual screening — GIN surrogate model with MC-Dropout uncertainty, Thompson Sampling acquisition, and plug-in docking oracles (QED mock · AutoDock Vina · Glide). Recovers >95% of top-1% hits while docking only ~6% of the library. Based on Graff, Shakhnovich & Coley, Chem. Sci. 2021.

A Machine Learning model that learns the relationship between Chromophores and the Absorption max in the DB for Chromophore dataset

Multi-scale molecular toxicity prediction using hierarchical graph neural networks with adaptive curriculum learning that prioritizes structurally complex molecules during training. Introduces a novel dual-granularity message passing mechanism (atom-level and functional-group-level) combined with difficulty-aware sample weighting based on molecular

A PyTorch/Jupyter project that fine-tunes the MolFormer chemical language model for masked language modeling and downstream regression, producing saved checkpoints and prediction outputs.

Multi-task ADMET property prediction via fine-tuned ChemBERTa molecular transformers

Stability-Aware GNN Framework for molecular toxicity prediction using Graph Attention Networks. Built on Tox21 dataset with ERI reliability scoring. Live demo on HuggingFace Spaces.