The integration of artificial intelligence and protein science: the protein structure prediction and protein design revolution behind the 2024 Nobel Prize in Chemistry
Lists
Only you see this
Abstract
<sec><p indent="0mm">In recent years, groundbreaking advances in artificial intelligence (AI) have transformed various aspects of human society at an unprecedented pace, driving revolutionary progress in scientific research. The 2024 Nobel Prizes in Physics and Chemistry were both awarded for AI-related research, highlighting the immense impact of AI in reshaping scientific fields. Notably, the Nobel Prize in Chemistry was awarded to Demis Hassabis and John M. Jumper of Google DeepMind for their remarkable contributions to protein structure prediction, while the other half of the award went to Professor David Baker from the University of Washington, recognizing his leadership in the revolutionary progress of protein design. The Rosetta software suite, developed by Baker’s team, not only predicts protein structures but also designs entirely new proteins, including those that do not exist in nature, opening up infinite possibilities for pharmaceutical development, vaccine creation, and sensor technology. </sec><sec> Proteins, as the fundamental building blocks of life, are the direct executors of biological activities. Their structure and function are determined by the amino acid sequence encoded in their primary structure. The peptide chains forming proteins fold into three-dimensional structures as they are synthesized by ribosomes, ultimately assuming specific functional roles. However, predicting the three-dimensional structure of proteins from their amino acid sequences has proven to be an extraordinarily complex challenge due to the vast number of possible folding configurations. Despite decades of efforts, significant breakthroughs only occurred in the past few years with the advent of AI-driven methods, particularly AlphaFold2 (AF2), a deep learning algorithm developed by DeepMind that has revolutionized protein structure prediction with unprecedented accuracy. </sec><sec> This article traces the historical development of protein science, emphasizing key milestones, from X-ray crystallography and NMR spectroscopy to cryo-electron microscopy, all of which have advanced our understanding of protein structure. The pursuit of predicting protein structure from amino acid sequences has been a central goal in biochemistry since the 1970s, yet it remained elusive until AI breakthroughs. The early methods, such as homology modeling and threading, relied on comparing known structures to predict the unknown. In contrast, template-free modeling, which is grounded in first principles, strives to predict the structure by minimizing energy functions, but computational limitations hinder its progress. </sec><sec> The real breakthrough came with the introduction of deep learning techniques in the last decade. AI algorithms, including AlphaFold2, revolutionized the field by directly learning from large-scale datasets of protein sequences and their associated structures. By combining multi-sequence alignments (MSA) and template modeling, AlphaFold2 effectively captures evolutionary information and molecular interactions to predict protein structures with remarkable accuracy. Furthermore, the use of deep neural networks and novel training strategies in AlphaFold2 significantly improved the efficiency and reliability of predictions, making it a game-changer for structural biology. </sec><sec> This article discusses the technical innovations behind AlphaFold2, including its neural network architecture, integration of evolutionary data through MSA, and the novel use of a structure refinement module. These advancements allowed AlphaFold2 to predict protein structures with atomic-level precision, transforming the field and demonstrating its potential in drug discovery and disease research. The success of AlphaFold2 has implications beyond basic biology, offering a new paradigm for understanding protein functions, accelerating the development of therapeutic agents, and enabling the design of synthetic proteins with specific properties. </sec><sec> The article also evaluates AlphaFold2’s performance in predicting structures of G protein-coupled receptors (GPCRs), highlighting its accuracy in predicting single-domain structures and its potential for advancing drug discovery, especially in ligand-receptor interactions. However, challenges remain, particularly in predicting protein side-chain conformations, where improvements are still needed. Despite these challenges, AlphaFold2 represents a monumental step forward in the field, providing an unprecedented tool for structural biologists and molecular biologists alike. </sec><sec> In conclusion, the fusion of AI and protein science has reshaped the landscape of biomedical research, as evidenced by the 2024 Nobel Prize in Chemistry. As AI continues to evolve, its impact on protein science will expand, unlocking new avenues for therapeutic development and revolutionizing the way we approach disease treatment and drug design. </sec>
BibTeX
@inproceedings{liu2025integration,
author = {Shenhui Liu and Zhi‐Jie Liu},
title = {{The integration of artificial intelligence and protein science: the protein structure prediction and protein design revolution behind the 2024 Nobel Prize in Chemistry}},
booktitle = {Chinese},
year = {2025},
doi = {10.1360/tb-2024-1178},
}Loading related content...