Prediction of inter-chain distance maps of protein complexes with 2D attention-based deep neural networks

Abstract

Residue-residue distance information is useful for predicting tertiary structures of protein monomers or quaternary structures of protein complexes. Many deep learning methods have been developed to predict intra-chain residue-residue distances of monomers accurately, but few methods can accurately predict inter-chain residue-residue distances of complexes. We develop a deep learning method CDPred (i.e., Complex Distance Prediction) based on the 2D attention-powered residual network to address the gap. Tested on two homodimer datasets, CDPred achieves the precision of 60.94% and 42.93% for top L/5 inter-chain contact predictions (L: length of the monomer in homodimer), respectively, substantially higher than DeepHomo’s 37.40% and 23.08% and GLINTER’s 48.09% and 36.74%. Tested on the two heterodimer datasets, the top Ls/5 inter-chain contact prediction precision (Ls: length of the shorter monomer in heterodimer) of CDPred is 47.59% and 22.87% respectively, surpassing GLINTER’s 23.24% and 13.49%. Moreover, the prediction of CDPred is complementary with that of AlphaFold2-multimer.

Reconciling membrane protein simulations with experimental DEER spectroscopy data

“…We also compared closest heavy carbon distances, which are majorly used for contact prediction between protein residues, with the probe distances. 58 The median value of the closest heavy atom based distance distribution is ∼0.9 nm lower than the ON–ON atom distance median.…”

Combining pairwise structural similarity and deep learning interface contact prediction to estimate protein complex model accuracy in CASP15

“…The interaction interfaces between two chains that interact in at least 20% of multimers are selected for the ICPS calculation. The inter-chain residue-residue contact probability map for every two chains forming an identified interaction interface are then predicted by an inter-chain residue-residue contact predictor - CDPred(Guo et al, 2022). CDPred is a deep learning-based tool that can predict the inter-chain distance map for homodimers and heterodimers.…”

“…As the tertiary structure prediction problem has been largely solved, the field started to focus more on predicting the quaternary structures of protein complexes and assemblies, which also had a long history of development but did not progress as fast as the tertiary structure prediction (Lensink et al, 2016, 2018, 2021). However, the situation started to change as more and more deep learning methods were developed to predict inter-protein contacts and quaternary structures(Quadir et al, 2021; Xie & Xu, 2021; Yan & Huang, 2021; Evans et al, 2022; Guo et al, 2022; Roy et al, 2022). Particularly, adapting AlphaFold2 for protein quaternary structure prediction (i.e., AlphaFold-multimer (Evans et al, 2022)) drastically improved the accuracy of protein complex/assembly structure prediction.…”

“…To address the weakness, we developed a hybrid method (MULTICOM_qa) combining the pairwise similarity score between a multimer model and other models (PSS) with a single-model-based interface contact probability score (ICPS) that does not depend on the similarity between models to estimate the accuracy of the model. ICPS is calculated from the inter-chain contact map predicted by inter-chain residue-residue contact/distance predictors (Guo et al, 2022; Roy et al, 2022; Xie & Xu, 2022) such as CDPred (Guo et al, 2022). MULTICOM_qa uses a weighted average of PSS and ICPS to predict the global accuracy (quality) of a quaternary structural model.…”

A Gated Graph Transformer for Protein Complex Structure Quality Assessment and its Performance in CASP15

“…Recently, the problem of protein tertiary structure prediction has largely been solved by deep learning methods (e.g., Jumper et al (2021)). Furthermore, new deep learning methods (e.g., Evans et al (2021); Bryant et al (2022); Guo et al (2022)) have begun making advancements in protein complex (quaternary) structure prediction.…”