In research published in Nature, André Faure and Professor Ben Lehner – co-founders of ALLOX – have discovered that the impact of mutations on protein stability is more predictable than previously thought. By experimentally sampling from protein sequence spaces larger than 10 billion, they show that fully interpretable energy models allow accurate genetic prediction in high-dimensional sequence spaces.

“Our approach, which showcases and further validates ALLOX’s foundational technology, stands in stark contrast to the extremely complicated deep neural networks which are increasingly being used to navigate protein sequences. We demonstrate that the genetic architecture of at least some proteins is remarkably simple.” says André Faure, first author of the study.

For years, there has been an underlying assumption that two mutations might interact with each other in unexpected ways, enhancing or suppressing each other’s effects. The study found that while mutations do interact, it is a relatively rare occurrence, and the vast majority affect a protein independently of each other.

“Our discovery turns an old understanding on its head, showing that the endless possibilities of protein mutations boil down to straightforward rules. We don’t need supercomputers to predict a protein’s behaviour – just good measurements and simple maths will do.” says Professor Ben Lehner, Chief Scientific Advisor at ALLOX and dual affiliate between the CRG and the Wellcome Sanger Institute in the UK.

André Faure, CDO at ALLOX, has been awarded a three-year Torres Quevedo industrial research grant (PTQ2023-013021 awarded by MCIU/AEI/10.13039/501100011033) towards the prediction of allosteric sites in silico.

Read the Wellcome Sanger Institute News article

Read the research article

A three-dimensional image showing the SH3 domain of the human protein GRB2 (grey) indicating the 34 combinatorially mutated residues in this study (orange). Credit: Martí-Aranda/Wellcome Sanger Institute