Protein structure prediction using multiple deep neural networks in the 13th Critical Assessment of Protein Structure Prediction (CASP13)

By Andrew W. Senior, Richard Evans, John M. Jumper, James Kirkpatrick, Laurent Sifre, Tim Green, Chongli Qin, Augustin Žídek, A. Nelson, Alex Bridgland and 9 more. 2019

268 citations recorded by the source corpus.

We describe AlphaFold, the protein structure prediction system that was entered by the group A7D in CASP13. Submissions were made by three free‐modeling (FM) methods which combine the predictions of three neural networks. All three systems were guided by predictions of distances between pairs of residues produced by a neural network. Two systems assembled fragments produced by a generative neural network, one using scores from a network trained to regress GDT_TS. The third system shows that simple gradient descent on a properly constructed potential is able to perform on par with more expensive traditional search techniques and without requiring domain segmentation. In the CASP13 FM assessors' ranking by summed z‐scores, this system scored highest with 68.3 vs 48.2 for the next closest group (an average GDT_TS of 61.4). The system produced high‐accuracy structures (with GDT_TS scores of 70 or higher) for 11 out of 43 FM domains. Despite not explicitly using template information, the results in the template category were comparable to the best performing template‐based methods.

Subfields: Materials Chemistry, Molecular Biology, Model Architectures, Computational Biology & Bioinformatics

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Materials Chemistry

In Molecular Biology

In Model Architectures

In Computational Biology & Bioinformatics

Citations in the corpus

MSA Transformer (2021, methodology): “Supervised Structure Prediction Supervised structure prediction using deep neural networks has driven ground-breaking progress on the protein structure prediction problem (Senior et al., 2019; Jumper et al., 2020).”
High‐accuracy protein structure prediction in CASP14 (2021, background): “1,11 Now, in CASP14 we saw a further leap in model accuracy, 36 with the best model for targets at any difficulty level reaching a GDT_TS above 90%, a range at the level of experimental accuracy.”
Critical Assessment of Methods of Protein Structure Prediction (CASP) – Round XIV (2021, methodology): “A number of CASP participants also began using these methods to predict a continuous probability function for inter-residue distances rather than just a binary yes/no for contacts.(36) Together, these developments did result in the CASP13 major jump in 3D accu-”
Transformer protein language models are unsupervised structure learners (2020, methodology): “Senior et al. (2019), use features from a Potts model ﬁt with pseudolikelihood maximization to predict pairwise distances with a deep residual network and optimize the ﬁnal structure using Rosetta.”
Applying and improving AlphaFold at CASP14 (2021, methodology): “The AlphaFold system used was an updated version of the one deployed in CASP-Covid 1 and differs substantially to the version of AlphaFold deployed in CASP13.”
Yuel: Compound-Protein Interaction Prediction with High Generalizability (2021, background): “…heralding a trend to supersede traditional methods due to its unparalleled performance, e.g., AlphaFold2 8 has stunned the protein structure communication by sweeping the biennial protein 3D structure prediction competition, CASP (the critical assessment of protein structure prediction) 9 .”
Leveraging Data Science to Combat COVID-19: A Comprehensive Review (2020, methodology): “AlphaFold [175] is a DL model based on a dilated ResNet architecture [139], which predicts the distance and the distribution of angles between amino acid residing on protein structure.”
Mapping the Landscape of Artificial Intelligence Applications against COVID-19 (2020, methodology): “These predictions are used to construct a “potential of mean force” which can be used to characterize the protein’s shape (Senior et al., 2019).”
SE(3)-Equivariant Energy-based Models for End-to-End Protein Folding (2021, background): “There has been a remarkable progress towards more accurate protein structure prediction, largely owing to more precise inter-residue contact/distance and orientation predictions by deep networks [1, 2, 46, 48, 29, 30, 17].”