Authors: Andrew Kirjner, Jason Yim, Raman Samusevich, Shahar Bracha, Tommi Jaakkola, Regina Barzilay, Ila Fiete

Published on: July 02, 2023

Impact Score: 8.3

Arxiv code: Arxiv:2307.00494

Summary

• What is new: Introduces a new method for protein optimization using graph-based smoothing of the fitness landscape, achieving state-of-the-art results.
• Why this is important: Current protein engineering methods are limited by a small mutational radius constraining optimization, which drastically limits the design space.
• What the research proposes: A novel approach involving smoothing the protein fitness landscape with Tikunov regularization and optimizing using discrete energy-based models and MCMC.
• Results: Achieved a 2.5 fold fitness improvement over the training set in the GFP and AAV benchmarks, showing significant potential in protein optimization in a limited data regime.

Technical Details

Technological frameworks used: Tikunov regularization for smoothing, Gibbs sampling with Graph-based Smoothing (GGS) for optimization

Models used: Discrete energy-based models

Data used: GFP and AAV benchmarks

Potential Impact

Biotechnology and medicine sectors, particularly companies involved in protein engineering and therapeutic development

Want to implement this idea in a business?

We have generated a startup concept here: FitProEngine.