BoltzGen to ESM2 Evolver (frAncIs)

My entire design pipeline runs locally on an RTX5060 Ti using a custom multi-tab Streamlit application. It can generate around one thousand BoltzGen candidates per day and then expand those into roughly five hundred thousand ESM2-evolved variants.

For this project I selected the best de novo designs from a set of two thousand BoltzGen proteins and grew that set to one million ESM2-evolved sequences.

Each design was scored with pLDDT, PAE, and iPTM, along with additional biophysical filters. Affinity was estimated using my LoRa-fine-tuned ESM2 KD predictor, trained on about two hundred thousand sequence–KD pairs.

This submission serves as a proof of concept for the broader protein foundry platform I am building.