Doctoral Thesis: Generative Diffusion Models towards De Novo Protein Design

By: Jason Yim

Supervisors: Tommi Jaakkola, Regina Barzilay

Details

• Date: Monday, March 31
• Time: 1:30 pm - 3:00 pm
• Location: Kiva room 32-G449

Additional Location Details:

Computational modeling of proteins is challenging due to the complexity of the protein’s structure and sequence space. The structure requires modeling continuous 3D coordinates of atoms with rigid biochemical constraints of the polymer chain while the sequence is a series of discrete amino acids that should fold into a plausible protein structure. The structure-sequence relationship is complex, but deep learning approaches have proven promising to learn the relationship from decades of experimentally studied proteins. This proposal aims to develop state-of-the-art generative models that learn to sample protein structure and sequence that can be guided towards desired functions. While the proposed contributions are to develop novel machine learning techniques, the translation of these models to real-world de novo protein design applications will be emphasized and pursued alongside the technical development.

Host

• Jason Yim
• Email: jyim@mit.edu