A new AI approach to protein design

During my master's thesis at the Laboratory for Biomolecular Modeling at EPFL, I had the chance to work with the new deep learning model designed to predict protein sequences from backbone scaffolds, incorporating complex molecular environments. In this illustration, I wanted to illustrate how this model called CARBonAra (Context-aware Amino acid Recovery from Backbone Atoms and heteroatoms), is working.

I was also very honored and happy to see that my illustration was used by the former EPFL President, Martin Vetterli, during his last speech at the Magistrale 2024. 

"AI clearly holds huge potential for research and innovation. And probably for society too. But without an adequate governance system, it’ll become our Achille’s heel. AI should remain nothing more than a tool used by humans."  

-Martin Vetterli

Learn more about CARBonAra: https://actu.epfl.ch/news/a-new-ai-approach-to-protein-design/

© EPFL / Adrien Buttier

The end of a long-standing debate

In the illustration is shown the single molecule nanopore system (Aerolysin) that was used to investigate the mechanism of a Heat Shock Protein (HSP70s).

HSP70s is a protein chaperone involved in protein disaggregation and translocation, but its physical mechanism of action remains debated, with three models proposed to date (Power Stroke, Brownian Ratchet, or Entropic Pulling).
With this powerful nanopore system, they found that HSP70s extract polypeptide trapped in the nanopore by generating strong forces and thus providing evidence for the Entropic Pulling mechanism of action, solving a long-standing debate. 

Learn more: https://lnkd.in/dppExPkp

SCIENTIFIC FIGURES