On the Encoding of Proteins for Disordered Regions Prediction

Abstract

Disordered regions, i.e., regions of proteins that do not adopt a stable three-dimensional structure, have been shown to play various and critical roles in many biological processes. Predicting and understanding their formation is therefore a key sub-problem of protein structure and function inference. A wide range of machine learning approaches have been developed to automatically predict disordered regions of proteins. One key factor of the success of these methods is the way in which protein information is encoded into features. Recently, we have proposed a systematic methodology to study the relevance of various feature encodings in the context of disulfide connectivity pattern prediction. In the present paper, we adapt this methodology to the problem of predicting disordered regions and assess it on proteins from the 10th CASP competition, as well as on a very large subset of proteins extracted from PDB. Our results, obtained with ensembles of extremely randomized trees, highlight a novel feature function encoding the proximity of residues according to their accessibility to the solvent, which is playing the second most important role in the prediction of disordered regions, just after evolutionary information. Furthermore, even though our approach treats each residue independently, our results are very competitive in terms of accuracy with respect to the state-of-the-art. A web-application is available at http://m24.giga.ulg.ac.be:81/x3Disorder.

DeepCNF-D: Predicting Protein Order/Disorder Regions by Weighted Deep Convolutional Neural Fields

“…Table 4 illustrates the AUC value of 10 cross validation batch datasets of Disorder723 with settings of different combinations of feature classes. Not surprisingly, the most contributing feature class is evolutionary information, which confirms that it is the conservation profile that makes the order/disorder regions different [ 18 ]. The second important feature class is the structural information class which is based on predicted secondary structure and solvent accessibility.…”

“…In short, it has been demonstrated that charged and hydrophilic residues, such as P, E, S, Q and K are more likely to be in disordered states, whereas neutral and hydrophobic residues, such as C, W, I, Y, F, L, M and H, are order-promoting residues [ 2 ]. It has also been suggested that evolutionary conservation largely determines order/disorder states [ 18 ]. Besides these amino acid and evolution related features, we need to find additional relevant and/or complementary features that could contribute to the order/disorder prediction accuracy.…”

“…(a) Weighted DeepCNF combines the advantages of both conditional neural fields (CNF) [ 19 ] and deep convolutional neural networks [ 20 ]. It not only captures medium- and long-range sequence information, but also describes interdependency of order/disorder labels among adjacent residues; (b) To solve the label imbalance issue, the reciprocal of the occurrence frequencies of order/disorder labels were used for weighting during training and prediction; (c) Besides the traditional amino acid and evolution related features that are relevant to order/disorder discrimination, a strong tendency has been shown for disordered regions to exist in the coiled state [ 18 ] and exposed to solvent [ 2 ]. Inspired by this phenomenon, additional structural properties, such as predicted secondary structure [ 21 ] and predicted solvent accessibility [ 22 ], were incorporated into our weighted DeepCNF model.…”

The Order-Disorder Continuum: Linking Predictions of Protein Structure and Disorder through Molecular Simulation

“…1672 proteins in the test set were divided into five bins of size 214, 452, 508, 233, and 265 proteins, in increasing helix/beta content. In general, increased coil content tends to align with increased disorder and associated flexibility in a protein 50,101 , so our test set separation corresponds to an increasing degree of disorder. Figure 2 captures the increasing helix and beta content and corresponding decreasing coil content across the five sampled proteins, one from each test bin.…”

DisoMCS: Accurately Predicting Protein Intrinsically Disordered Regions Using a Multi-Class Conservative Score Approach

“…These excellent predictors performed better in CASP (Critical Assessment of techniques for protein Structure Prediction) comparing with previous methods. Becker et al also presented a predictor [ 40 ], which was competitive in terms of accuracy with respect to the state-of-the-art.…”