Journal Article

OpenFold: retraining AlphaFold2 yields new insights into its learning mechanisms and capacity for generalization

Zhang, Minjia, Nowaczynski, Arkadiusz, Bonneau, Richard, Lorenzo, Pablo Ribalta, Wang, Bei, Floristean, Christina, Guney, Murat Efe, Ojewole, Adegoke, Ahdritz, Gustaf, Zhang, Bo, Stepniewska-Dziubinska, Marta M., Ban, Yih-En Andrew, He, Yuxiong, Xia, Qinghui, Ra, Stephen, Weitzner, Brian, Kadyan, Sachin, Watkins, Andrew M., Li, Conglong, Berenberg, Daniel, Zhang, Shang, Sorger, Peter K., Zhang, Zhao, AlQuraishi, Mohammed, Gerecke, William, Biderman, Stella, Zanichelli, Niccolò, Nivon, Lucas, Chen, Shiyang, Mostaque, Emad, Bouatta, Nazim, Fisk, Ian, Song, Shuaiwen Leon, O’Donnell, Timothy J.

2024

Overview

AlphaFold2 revolutionized structural biology with the ability to predict protein structures with exceptionally high accuracy. Its implementation, however, lacks the code and data required to train new models. These are necessary to (1) tackle new tasks, like protein–ligand complex structure prediction, (2) investigate the process by which the model learns and (3) assess the model’s capacity to generalize to unseen regions of fold space. Here we report OpenFold, a fast, memory efficient and trainable implementation of AlphaFold2. We train OpenFold from scratch, matching the accuracy of AlphaFold2. Having established parity, we find that OpenFold is remarkably robust at generalizing even when the size and diversity of its training set is deliberately limited, including near-complete elisions of classes of secondary structure elements. By analyzing intermediate structures produced during training, we also gain insights into the hierarchical manner in which OpenFold learns to fold. In sum, our studies demonstrate the power and utility of OpenFold, which we believe will prove to be a crucial resource for the protein modeling community. OpenFold is a trainable open-source implementation of AlphaFold2. It is fast and memory efficient, and the code and training data are available under a permissive license.

• Publisher: Nature Publishing Group US,Nature Publishing Group
• Subject: 631/114/1305 / 631/114/470 / Accuracy / Algorithms / Biochemistry & Molecular Biology / Bioinformatics / Biological Microscopy / Biological Techniques / Biomedical and Life Sciences / Biomedical Engineering/Biotechnology / Computational Biology - methods / Learning / Life Sciences / Models, Molecular / Protein Conformation / Protein Folding / Protein structure / Protein Structure, Secondary / Proteins / Proteins - chemistry / Proteomics / Secondary structure / Software / Source code / Task complexity