EPSILON-CP: using deep learning to combine information from multiple sources for protein contact prediction

Abstract

BackgroundAccurately predicted contacts allow to compute the 3D structure of a protein. Since the solution space of native residue-residue contact pairs is very large, it is necessary to leverage information to identify relevant regions of the solution space, i.e. correct contacts. Every additional source of information can contribute to narrowing down candidate regions. Therefore, recent methods combined evolutionary and sequence-based information as well as evolutionary and physicochemical information. We develop a new contact predictor (EPSILON-CP) that goes beyond current methods by combining evolutionary, physicochemical, and sequence-based information. The problems resulting from the increased dimensionality and complexity of the learning problem are combated with a careful feature analysis, which results in a drastically reduced feature set. The different information sources are combined using deep neural networks.ResultsOn 21 hard CASP11 FM targets, EPSILON-CP achieves a mean precision of 35.7% for top- L/10 predicted long-range contacts, which is 11% better than the CASP11 winning version of MetaPSICOV. The improvement on 1.5L is 17%. Furthermore, in this study we find that the amino acid composition, a commonly used feature, is rendered ineffective in the context of meta approaches. The size of the refined feature set decreased by 75%, enabling a significant increase in training data for machine learning, contributing significantly to the observed improvements.ConclusionsExploiting as much and diverse information as possible is key to accurate contact prediction. Simply merging the information introduces new challenges. Our study suggests that critical feature analysis can improve the performance of contact prediction methods that combine multiple information sources. EPSILON-CP is available as a webservice: http://compbio.robotics.tu-berlin.de/epsilon/ Electronic supplementary materialThe online version of this article (doi:10.1186/s12859-017-1713-x) contains supplementary material, which is available to authorized users.

Assessing the accuracy of contact predictions in CASP13

“…After 2010, when first papers on correlated mutations were published, a number of refinements took place, such as combining a set of nonoverlapping contact predictions in a consensus approach and especially the application of various supervised deep neural network based approaches, 42 further improving the accuracy of contact prediction techniques. [43][44][45][46][47][48][49][50][51][52] In this work, we assess performance of the state-of-the-art of contact prediction methods in 2018 at CASP13. We explore three general questions concerning contact predictions.…”

Peer Review #1 of "Evolutionary couplings detect side-chain interactions (v0.1)"

“…Practically speaking, contact identification methods and evolutionary couplings are increasingly important for a variety of applications. Frequently, evolutionary couplings are combined with a variety of other features and used as input for machine learning and associated neural network-based algorithms to predict structural and functional properties of proteins (Cheng and Baldi, 2007;Tegge et al, 2009;Lena et al, 2012;Jones et al, 2015;Michel et al, 2017;Xiong et al, 2017;Stahl et al, 2017;He et al, 2017;Riesselman et al, 2018;Liu et al, 2018;Wozniak et al, 2018;Jones and Kandathil, 2018;Adhikari et al, 2018;Hanson et al, 2018). Our analysis suggests that there may be biases in the training data-essential to supervised learning techniques-owing to the method used to define true positive contacts.…”

“…Manuscript to be reviewed (Burger and Van Nimwegen, 2008;Hopf et al, 2014;Ovchinnikov et al, 2014), as well as to predict the effect of mutations on protein stability and function (Hopf et al, 2017). Many of these approaches have been further improved through the use of machine learning (Cheng and Baldi, 2007;Jones et al, 2015;Michel et al, 2017), and specifically deep neural networks that leverage evolutionary couplings along-side numerous other protein features (Tegge et al, 2009;Lena et al, 2012;Xiong et al, 2017;Stahl et al, 2017;He et al, 2017;Riesselman et al, 2018;Liu et al, 2018;Wozniak et al, 2018;Jones and Kandathil, 2018;Adhikari et al, 2018;Hanson et al, 2018).…”

Prediction of Structures and Interactions from Genome Information

“…3 Machine learning methods to augment the contact prediction accuracy based on amino acid coevolution All the DCA methods such as mfDCA, plmDCA, GREMLIN, and PSICOV predict significantly nonoverlapping sets of contacts [46,52,108]. Then, increasing prediction accuracy by combining their predictions together with other sequence/structure information have been attempted [94,96,95,51,46,52,105,91]; see Table 3.…”