A deep dilated convolutional residual network for predicting interchain contacts of protein homodimers

Abstract

Motivation: Deep learning has revolutionized protein tertiary structure prediction recently. The cutting-edge deep learning methods such as AlphaFold can predict high-accuracy tertiary structures for most individual protein chains. However, the accuracy of predicting quaternary structures of protein complexes consisting of multiple chains is still relatively low due to lack of advanced deep learning methods in the field. Because interchain residue-residue contacts can be used as distance restraints to guide quaternary structure modeling, here we develop a deep dilated convolutional residual network method (DRCon) to predict interchain residue-residue contacts in homodimers from residue-residue co-evolutionary signals derived from multiple sequence alignments of monomers, intrachain residue-residue contacts of monomers extracted from true/predicted tertiary structures or predicted by deep learning, and other sequence and structural features.

Results: Tested on three homodimer test datasets (Homo_std dataset, DeepHomo dataset and CASP-CAPRI dataset), the precision of DRCon for top L/5 interchain contact predictions (L: length of monomer in a homodimer) is 43.46%, 47.10% and 33.50% respectively at 6 Å contact threshold, which is substantially better than DeepHomo and DNCON2_inter and similar to Glinter. Moreover, our experiments demonstrate that using predicted tertiary structure or intrachain contacts of monomers in the unbound state as input, DRCon still performs well, even though its accuracy is lower than using true tertiary structures in the bound state are used as input. Finally, our case study shows that good interchain contact predictions can be used to build high-accuracy quaternary structure models of homodimers.

Availability and implementation: The source code of DRCon is available at https://github.com/jianlin-cheng/DRCon. The datasets are available at https://zenodo.org/record/5998532#.YgF70vXMKsB.

Supplementary information: Supplementary data are available at Bioinformatics online.

Fig. 1.

The deep learning architecture of DRCon for interchain contact prediction in homodimers. For a homodimer in which the length of the monomer sequence is L, the input is a L×L×592 tensor. The number of input features for each pair of residues is 592. For convenience, L is set to a fixed number—600. 0 padding is applied if L is less than 600. It is worth noting that in the prediction phase, no zero padding is used in generating the input tensor if L is greater than 600. The input is transformed to a 600×600×48 tensor using a 2D-convolutional layer which has a kernel size of 1 and uses Exponential Linear Unit (elu). The output of the convolution layer is passed through 36 residual blocks with kernel size of 3x3. Each residual block uses a 2D-convolution layer with a kernel size of 3, instance normalization and dropout of 15% probability of a neuron being ignored, followed by a dilated convolution layer without dropout. The step of the dilation in the dilated convolution layers in these blocks changes from 1, 2, 4, 8, 16 periodically. The sigmoid activation function is applied to the output of the last residual block to calculate the contact probability of each interchain residue–residue pair. The probabilities for residue pair (i, j) and residue pair (j, i) are averaged to a symmetric final contact map

Fig. 2.

Illustrating the effect of contact density on interchain contact prediction precision on the Homo_std test dataset

Fig. 3.

(A) The predicted and true contact maps of target 1DR0. The top L/5 predicted contacts (red dots) and true contacts (blue dots) are plotted. Most predicted contacts overlap with the true contacts, indicating a high contact prediction precision. (B) The superimposition of the true quaternary structure (chain A in red and chain B in green) and the predicted quaternary structure (chain A in blue and chain B in orange). The two quaternary structures are quite similar

Fig. 4.

Impact of different groups of features on the average top L/5 precision on the Homo_std test dataset