ResPRE: high-accuracy protein contact prediction by coupling precision matrix with deep residual neural networks

Abstract

Motivation Contact-map of a protein sequence dictates the global topology of structural fold. Accurate prediction of the contact-map is thus essential to protein 3D structure prediction, which is particularly useful for the protein sequences that do not have close homology templates in the Protein Data Bank. Results We developed a new method, ResPRE, to predict residue-level protein contacts using inverse covariance matrix (or precision matrix) of multiple sequence alignments (MSAs) through deep residual convolutional neural network training. The approach was tested on a set of 158 non-homologous proteins collected from the CASP experiments and achieved an average accuracy of 50.6% in the top-L long-range contact prediction with L being the sequence length, which is 11.7% higher than the best of other state-of-the-art approaches ranging from coevolution coupling analysis to deep neural network training. Detailed data analyses show that the major advantage of ResPRE lies at the utilization of precision matrix that helps rule out transitional noises of contact-maps compared with the previously used covariance matrix. Meanwhile, the residual network with parallel shortcut layer connections increases the learning ability of deep neural network training. It was also found that appropriate collection of MSAs can further improve the accuracy of final contact-map predictions. The standalone package and online server of ResPRE are made freely available, which should bring important impact on protein structure and function modeling studies in particular for the distant- and non-homology protein targets. Availability and implementation https://zhanglab.ccmb.med.umich.edu/ResPRE and https://github.com/leeyang/ResPRE. Supplementary information Supplementary data are available at Bioinformatics online.

Assessing the accuracy of contact predictions in CASP13

“…In the same article, the authors used an extreme case simulation when the neural network was fed with a single sequence. Despite the assumption that the input coupling matrix is random, 11 out of 158 cases still achieved a very high F ‐score > 0.5, among a number of other reasonable predictions. Similar observations were made in another recent article by Jones et al While more sequences have a positive impact on the accuracy of contact prediction in general, the deep residual neural networks have the ability to learn the underlying contact patterns from limited coevolutionary information; the latter is important for structural modeling hard protein targets lacking homologous sequences or having very shallow alignments .…”

Deep‐learning contact‐map guided protein structure prediction in CASP13

“…Among the component predictors, ResPRE (Figure C) is a newly developed deep‐learning‐based method to predict the contact‐map (with C β ‐C β distance <8 å) of a query sequence by coupling evolutionary precision matrices with deep residual neural networks . In brief, given the MSA obtained for a query sequence, ResPRE first calculates the covariance between every pair of residue types at every pair of positions.…”

Ensembling multiple raw coevolutionary features with deep residual neural networks for contact‐map prediction in CASP13

“…Since the covariance matrix in Equation (2) encodes the marginal correlations between variables, we calculate the second feature, the precision matrix (PRE), by minimizing the objective function: where the first two terms can be interpreted as the negative log‐likelihood of the inverse covariance matrix, that is, the precision matrix Θ, under the assumption that the data are under a multivariate Gaussian distribution. Here, tr ( S Θ) is the trace of the matrix S Θ and log |Θ| is the log determinant of Θ.…”

“…The inverse of the covariance matrix provides direct couplings between pairs of sites conditional on other positions. Thus, the precision matrix has better performance in the prediction of contact‐maps than the covariance matrix …”