Distance-based protein folding powered by deep learning

By Jinbo Xu. 2018

233 citations recorded by the source corpus.

Significance Accurate description of protein structure and function is a fundamental step toward understanding biological life and highly relevant in the development of therapeutics. Although greatly improved, experimental protein structure determination is still low-throughput and costly, especially for membrane proteins. As such, computational structure prediction is often resorted. Predicting the structure of a protein without similar structures in the Protein Data Bank is very challenging and usually needs a large amount of computing power. This paper shows that by using a powerful deep learning technique, even with only a personal computer we can predict new folds much more accurately than ever before. This method also works well on membrane protein folding. Direct coupling analysis (DCA) for protein folding has made very good progress, but it is not effective for proteins that lack many sequence homologs, even coupled with time-consuming conformation sampling with fragments. We show that we can accurately predict interresidue distance distribution of a protein by deep learning, even for proteins with ∼60 sequence homologs. Using only the geometric constraints given by the resulting distance matrix we may construct 3D models without involving extensive conformation sampling. Our method successfully folded 21 of the 37 CASP12 hard targets with a median family size of 58 effective sequence homologs within 4 h on a Linux computer of 20 central processing units. In contrast, DCA-predicted contacts cannot be used to fold any of these hard targets in the absence of extensive conformation sampling, and the best CASP12 group folded only 11 of them by integrating DCA-predicted contacts into fragment-based conformation sampling. Rigorous experimental validation in CASP13 shows that our distance-based folding server successfully folded 17 of 32 hard targets (with a median family size of 36 sequence homologs) and obtained 70% precision on the top L/5 long-range predicted contacts. The latest experimental validation in CAMEO shows that our server predicted correct folds for 2 membrane proteins while all of the other servers failed. These results demonstrate that it is now feasible to predict correct fold for many more proteins lack of similar structures in the Protein Data Bank even on a personal computer.

Subfields: Molecular Biology, 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 Molecular Biology

In Computational Biology & Bioinformatics

Citations in the corpus

Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences (2019, methodology): “For inputs, these methods use raw frequency and covariance statistics, or parameters of maximum-entropy sequence models ﬁtted on MSAs (Jones & Kandathil, 2018; Xu, 2018; Senior et al., 2018).”
De novo protein design by deep network hallucination (2020, background): “(1) is responsible for keeping the amino acid composition of the hallucinated designs within the viable region of the protein sequence space.”
Improved protein structure prediction using predicted interresidue orientations (2019, neutral): “Previous work (4) implicitly captured some orientation information by predicting multiple interresidue distances (Cβ–Cβ, Cα–Cα, Cα–Cg, Cg–Cg, and N–O), but in contrast to our multitask-learning approach, a separate network was used for each of the objectives.”
DEEPCON: Protein Contact Prediction using Dilated Convolutional Neural Networks with Dropout (2019, methodology): “…raw distance prediction is too difficult, some groups have developed methods that can predict contacts at various sequence separation thresholds [17] and demonstrate that such binning of distance ranges actually improves contact prediction at a single standard distance threshold, such as the…”
Energy-based models for atomic-resolution protein conformations (2020, background): “Xu (2018) developed a deep residual network that predicts the pairwise distances between residues in the protein structure from evolutionary covariation information.”
Recent developments in deep learning applied to protein structure prediction (2019, methodology): “Considering the NN models used in A7D (DeepMind's AlphaFold), 10 RaptorX, 8 or our own DMPfold, 9 it is important to understand the limitations of what these models are capable of doing.”
Deep learning techniques have significantly impacted protein structure prediction and protein design. (2021, background): “A similar residual neural network was later extended to predict the probability that the distance between two residues falls within a given distance range instead of predicting a binary contact map [32].”
Analysis of distance-based protein structure prediction by deep learning in CASP13 (2019, background): “3B) since they mainly depend on pairwise distance information predicted by deep ResNet.”
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].”
Folding non-homologous proteins by coupling deep-learning contact maps with I-TASSER assembly simulations (2021, methodology): “State-of-the-art approaches (Greener et al., 2019; Lamb et al., 2019; Marks et al., 2011; Xu, 2019) often utilize traditional distance geometry-based structure reconstruction tools such as the Crystallography & NMR System (CNS) (Brunger et al.”
Toward the solution of the protein structure prediction problem (2021, methodology): “Distance map prediction jumped to the forefront of the ﬁ eld during the CASP13 experiment in 2018, when three predictors (RaptorX-Contact (43), DMPfold (172), and AlphaFold (173)), extended the use of deep ResNets for contact prediction to distance prediction.”
Fast end-to-end learning on protein surfaces (2020, background): “inferring the 3D structure from the aminoacid sequence) [3, 46, 37, 47].”
Protein storytelling through physics (2020, result): “The superior ability of DLmethods (131, 132) in harnessing data relative to earlier statistical methods led to the best results in CASP’s free modeling category (90).”