Aligning Protein Generative Models to Experimental Fitness with ProteinDPO
Widatalla, T., Borah, A.A., King, S.H., Driscoll, C.L., Rafailov, R., Hie, B.L. · Paper ↗ · Code ↗
PhD candidate in Biophysics at Stanford and the Arc Institute, advised by Brian Hie. I build and apply machine learning systems for molecular design, spanning graph neural networks, language models, and diffusion models across various biological applications. I've also worked on protein generative models at Generate Biomedicines and ML for drug discovery at Merck.
I'm especially interested in imbuing biophysical and experimental information into biological generative models for drug discovery and optimization, with a growing interest in agentic lab-in-the-loop design.
Always up for a chat. If you are local I love trying new cafés in SF, or we can hop on a call online, just send me an email!
Widatalla, T., Borah, A.A., King, S.H., Driscoll, C.L., Rafailov, R., Hie, B.L. · Paper ↗ · Code ↗
Mille-Fragoso, L.*, Driscoll, C.*, Wang, J.*, Dai, H.*, Widatalla, T.*, Zhang, J.*, Zhang, X., Rao, B., Feng, L., Hie, B., Gao, X. · Paper ↗ · Code ↗
Widatalla, T.*, Shuai, R.*, Huang, P-S, Hie, B. · Paper ↗ · Code ↗
Rollins, Z., Widatalla, T., Cheng, A., Metwally, E. · Paper ↗ · Code ↗
Rollins, Z., Widatalla, T., Waight, A., Cheng, A., Metwally, E. · Paper ↗
Waight, A., et al. (incl. Widatalla, T.) · Paper ↗
Reinforcement learning for structure-conditioned generative models to experimental fitness and applied to vaccine optimization (ProteinDPO) , Diffusion models for full-atom protein sequence design (FAMPNN), Reinforcement learning for structure-conditioned generative models to experimental fitness and applied to vaccine optimization, first open-source framework for single-shot and low-N de novo antibody design (Germinal) , and other cool projects yet to be released!
Worked on training of diffusion-based protein generative model for improved enzyme and binder design.
Built production ML systems combining GNNs, language models and molecular dynamics simulations for antibody property prediction, and a high-throughput small molecule virtual-screening pipeline using active learning and structural deep learning.
Developed software for mining structural properties from PDB structures to guide structure-based ligand design.