Tristan Bepler

TARDIS is a deep learning framework that rapidly and accurately segments membranes and filaments in electron microscopy data, enabling large-scale biophysical analysis.

This paper uses a Bayesian language model to design high-affinity scFv antibody libraries, achieving 28.8-fold better binding than directed evolution.

CryoDRGN uses deep neural networks to reconstruct heterogeneous protein complexes and their motion from cryo-EM data, revealing previously unseen structural states.

Topaz-Denoise improves cryoEM and cryoET image quality by increasing SNR, enabling faster analysis and solving previously elusive structures.

The paper identifies intrinsic DNA-binding specificity differences between transcription factor paralogs that partly explain their distinct genomic binding patterns.