ESPEYB19 15. Editors’ Choice Basic Science and Genetics (5 abstracts)
15.15. Accurate prediction of protein structures and interactions using a three-track neural network
Baek M , DiMaio F , Anishchenko I , Dauparas J , Ovchinnikov S , Lee GR , Wang J , Cong Q , Kinch LN , Schaeffer RD , Millan C , Park H , Adams C , Glassman CR , DeGiovanni A , Pereira JH , Rodrigues AV , van Dijk AA , Ebrecht AC , Opperman DJ , Sagmeister T , Buhlheller C , Pavkov-Keller T , Rathinaswamy MK , Dalwadi U , Yip CK , Burke JE , Garcia KC , Grishin NV , Adams PD , Read RJ & Baker D
Science. 2021;373(6557):871-6. doi: 10.1126/science.abj8754 PubMed ID: 34282049
Brief summary: This study reveals a Deep Learning method, ‘RoseTTA fold’, based on DeepMind’s Alphafold2 framework, to predict 3-dimensional protein structures from 1-dimensional sequence information and generate models of protein–protein complexes with high accuracy.
Previous Yearbook commentaries have described the application of machine learning (‘artificial intelligence’, AI) approaches to science and medicine questions, such as growth modelling and adult height prediction. Prediction of 3-dimensional protein structures is a challenge that is several orders of magnitude tougher, due to the endless possible amino-acid sequence compositions and interactions. It also offers important rewards. The authors used 150K known structures as a training set and illustrate the potential advances in understanding by application to MC4R ligand binding. These academic authors, from the University of Washington, Seattle, acknowledge that RoseTTA fold is not quite as accurate as Alphafold2, which had been announced in 2020 but kept private by its developer, DeepMind, an AI company owned by Google. A valuable achievement of the current paper was that it prompted DeepMind to publish and make publically available Alphafold2 on the same day (1). RoseTTA also has the advantage of predicting protein interactions and protein complexes.
These computer approaches now can solve in a few hours an important challenge that could typically take 1 year to achieve using traditional laborious laboratory techniques, such as X-ray crystallography. Knowledge of such structures can help to identify the key functional parts of a protein, can be used to model the structural and functional consequences of genetic mutations, and valuably predict the nature of protein–protein interactions, such as natural or synthetic ligand–receptor interactions, and also epitope–antibody interactions. Thus, such protein characterisation is fundamental for designing targets in the development of new drugs.
References: 1. Tunyasuvunakool K, et al. Highly accurate protein structure prediction for the human proteome. Nature. 2021 Aug;596(7873):590–596. doi: 10.1038/s41586-021-03828-1. PMID: 34293799.
Article tools
My recent searches
My recently viewed abstracts
Authors
ESPE Yearbook of Paediatric Endocrinology
ISSN 1662-4009 (Online)
© Bioscientifica 2024 |
Privacy policy |
Cookie settings
BiosciAbstracts
Bioscientifica Abstracts is the gateway to a series of products that provide a permanent, citable record of abstracts for biomedical and life science conferences.
Find out more