A Holy Grail — The Prediction of Protein Structure

This editorial reviews the development and scientific impact of AlphaFold, the deep learning system created by Demis Hassabis and John Jumper that predicts protein 3D structures from amino acid sequences. Recognized by the Lasker Basic Medical Research Award, AlphaFold addresses a 60-year challenge, accurately predicting how a one-dimensional protein sequence folds into a functional three-dimensional shape. Traditional structural methods (e.g., x-ray crystallography, cryo-EM) are costly and slow, whereas AlphaFold leverages deep neural networks trained on protein structures and multiple sequence alignments to infer residue proximities and atomic coordinates. Its architecture includes modules like the Evoformer and Structure Module, combining attention mechanisms and iterative refinement. AlphaFold’s predictions reached experimental accuracy in the 2020 CASP challenge. Applications include drug design, protein–protein interaction modeling, and variant impact assessment. The work exemplifies how AI–ML can solve scientific problems through end-to-end differentiable models and large-scale data integration, marking a transformational advance in biomedical research

This editorial reviews the development and scientific impact of AlphaFold, the deep learning system created by Demis Hassabis and John Jumper that predicts protein 3D structures from amino acid sequences. Recognized by the Lasker Basic Medical Research Award, AlphaFold addresses a 60-year challenge, accurately predicting how a one-dimensional protein sequence folds into a functional three-dimensional shape. Traditional structural methods (e.g., x-ray crystallography, cryo-EM) are costly and slow, whereas AlphaFold leverages deep neural networks trained on protein structures and multiple sequence alignments to infer residue proximities and atomic coordinates. Its architecture includes modules like the Evoformer and Structure Module, combining attention mechanisms and iterative refinement. AlphaFold’s predictions reached experimental accuracy in the 2020 CASP challenge. Applications include drug design, protein–protein interaction modeling, and variant impact assessment. The work exemplifies how AI–ML can solve scientific problems through end-to-end differentiable models and large-scale data integration, marking a transformational advance in biomedical research