DeepCDpred: Inter-residue distance and contact prediction for improved prediction of protein structure

Abstract

Rapid, accurate prediction of protein structure from amino acid sequence would accelerate fields as diverse as drug discovery, synthetic biology and disease diagnosis. Massively improved prediction of protein structures has been driven by improving the prediction of the amino acid residues that contact in their 3D structure. For an average globular protein, around 92% of all residue pairs are non-contacting, therefore accurate prediction of only a small percentage of inter-amino acid distances could increase the number of constraints to guide structure determination. We have trained deep neural networks to predict inter-residue contacts and distances. Distances are predicted with an accuracy better than most contact prediction techniques. Addition of distance constraints improved de novo structure predictions for test sets of 158 protein structures, as compared to using the best contact prediction methods alone. Importantly, usage of distance predictions allows the selection of better models from the structure pool without a need for an external model assessment tool. The results also indicate how the accuracy of distance prediction methods might be improved further.

Structure of SPH (self-incompatibility protein homologue) proteins: a widespread family of small, highly stable, secreted proteins

“…DeepCDPred [28] was used to generate structures for SPH15, PrsS1, PrsS3 and PrsS8. DeepCDPred uses deep learning to predict contacts and distances between residues, based on many inputs including amino acid profile, secondary structure prediction from SPIDER2 [16] and amino acid co-evolution couplings.…”

“…In addition, the programme uses the number of amino acids in the target protein, the number of homologous sequences in the MSA built by HHblits [14] and an estimate of the number of non-redundant sequences in the alignment. DeepCDPred [28] also has a β-sheet prediction algorithm that provides hydrogen-bonding restraints between strands. These restraints, together with restraints to enforce the secondary structure prediction from SPIDER2 [16], are fed into the protein structure modelling program, AbinitioRelax, which is from the Rosetta suite [34].…”

“…While SPH15 and the PrsS proteins clearly belong to the same protein family, they are in different classes within the family and there is only 15-18% sequence identity between them, rather small for reliable homology modelling (Figure 1a,b). However, the relatively large number of proteins in this family enabled us to test a newly developed de novo structure prediction method, DeepCDPred [28], based on co-evolution/correlated mutation to model SPH15. We then used this method to predict structures of the three poppy proteins.…”

“…For each protein, left histogram (red): HHblits [14] alignment of protein sequence with SPH15 followed by MODELLER. Centre histograms: structural alignment of the NMR template structure with the structural predictions from DeepCDPred [28], using TM-align [35], either without (blue) or with (white) the predicted contact restraints from DeepCDPred, followed by MODELLER. Right histogram (yellow outline): DOPE scores of 100 models from the de novo DeepCDPred calculation.…”

Introduction of a bounded penalty function in contact‐assisted simulations of protein structures to omit false restraints

“…Contact‐distance restraints are very useful in predicting the structures of proteins and other biological macromolecules . The sets of contacts can be predicted, especially based on correlated mutations or evolutionary coupling, or determined experimentally by nuclear magnetic resonance (NMR), chemical cross‐link mass‐spectroscopy (XLMS) or fluorescence resonance energy transfer (FRET) measurements; there are also approaches that use combinations of both experimental and predicted contacts . These sets usually do not include all contact information necessary to determine the target structure and, moreover, contain false or ambiguous contacts.…”