OverFiT

OverFiT (internally called beeHIVe) is a full-stack application for predicting how effective a molecule is at inhibiting HIV. Rather than betting on a single model, we built a mixture-of-experts system combining several complementary approaches: Graph Neural Networks, Morgan Fingerprint MLPs, structural feature models, and a Graph Variational Autoencoder for generating novel molecules.

A stacked ensemble meta-classifier synthesizes predictions across all of them. The frontend is Next.js, the backend is FastAPI, and the whole thing supports interactive molecule visualization and prediction through a web interface.