Protein prediction takes the prize

Abstract

The three-dimensional structure of a protein — as determined by its primary amino acid sequence — ultimately dictates how it can interact with other molecules and therefore governs its function, which could be, for example, catalysing a chemical transformation, transporting a molecule or providing structure. The capacity to accurately characterize a protein’s 3D structure directly from its primary sequence could help scientists predict its functional output and enable the design of ligands to selectively modulate protein activity. Likewise, being able to design new proteins built with selected functions of interest could potentially afford new biological tools and therapeutics and serve as a starting point for biomaterials.

The Critical Assessment of Protein Structure Prediction (CASP) — a biennial global competition with the goal of finding solutions to predicting protein structure — has featured works from all laureates of this year’s prize. DeepMind’s first iteration of AlphaFold applied a type of AI called ‘deep learning’ to predict the distance between pairs of amino acids within a protein using both genetic and structural data. This first version outperformed others in the 2018 CASP13 for prediction accuracy, but it was the second iteration AlphaFold2 — taking the top spot in the 2020 CASP14 — that became a real game changer.