Séminaire
|
Date : 16 janvier 2025 14:00 - Salle :Amphi 2 - Pôle commun
AlphaFold, the Artificial Intelligence approach (Nobel Prize 2024): a real (r)evolution or not?Alexandre G. de Brevern - INSERM |
The three-dimensional (3D) structure of proteins supports the majority of essential biological functions and also associated pathologies. Access to this information not only enables us to understand biological function at an atomistic level, but is also essential for the design of new drugs. However, obtaining these 3D structures experimentally is complex, time-consuming and expensive, and in many cases impossible. For over 35 years, computational approaches have been used to propose 3D structural models from sequence. Techniques have evolved from simple comparative modelling approaches (copy/paste) to increasingly complex approaches. In 2018, the deep learning method AlphaFold (from the company DeepMind) was proposed in the CASP competition which allows assessing the quality of such approaches, with very fair results, but comparable to the best approaches. DeepMind engineers modified the architecture of the project, and in 2020 AlphaFold2 achieved remarkable results. The media hype was impressive, leading to memorable headlines such as: 'an AI algorithm that solved the 50-year challenge of predicting protein structure'. Declared method of the year by Science, Nature, Life... AlphaFold will be awarded the Nobel Prize in Chemistry in 2024. The aim of this talk is to put the scientific question of the proposal for 3D structural models, real applications and potential limitations of the approach back into perspective with concrete cases, and to open the discussion on AlphaFold as an evolution or revolution in structural bioinformatics and, above all, beyond.
