Improved inter-protein contact prediction using dimensional hybrid residual networks and protein language models

Abstract

The knowledge of contacting residue pairs between interacting proteins is very useful for the structural characterization of protein–protein interactions (PPIs). However, accurately identifying the tens of contacting ones from hundreds of thousands of inter-protein residue pairs is extremely challenging, and performances of the state-of-the-art inter-protein contact prediction methods are still quite limited. In this study, we developed a deep learning method for inter-protein contact prediction, which is referred to as DRN-1D2D_Inter. Specifically, we employed pretrained protein language models to generate structural information-enriched input features to residual networks formed by dimensional hybrid residual blocks to perform inter-protein contact prediction. Extensively bechmarking DRN-1D2D_Inter on multiple datasets, including both heteromeric PPIs and homomeric PPIs, we show DRN-1D2D_Inter consistently and significantly outperformed two state-of-the-art inter-protein contact prediction methods, including GLINTER and DeepHomo, although both the latter two methods leveraged the native structures of interacting proteins in the prediction, and DRN-1D2D_Inter made the prediction purely from sequences. We further show that applying the predicted contacts as constraints for protein–protein docking can significantly improve its performance for protein complex structure prediction.

Protein language model embedded geometric graphs power inter-protein contact prediction

“…Besides, coming from the same group, Lin et al further developed DeepHomo2(Lin et al, 2023) for inter-protein contact prediction for homomeric PPIs by including the MSA (multiple sequence alignment) embeddings and attentions from an MSA-based protein language model (MSA transformer)(R. M. Rao et al, 2021b) in their prediction model, which further improved the prediction performance. In almost the same time with DeepHomo2, we proved that embeddings from protein language models(R. Rao et al, 2021; Rives et al, 2021) (PLMs) are very effective features to predict inter-protein contacts for both homomeric and heteromeric PPIs, and we further show the sequence embeddings (ESM-1b(Rives et al, 2021)), MSA embeddings (ESM-MSA-1b(R. M. Rao et al, 2021b) & Position-Specific Scoring Matrix (PSSM)) and the inter-protein coevolutionary information complement each other in the prediction, with which we developed DRN-1D2D_Inter(Si & Yan, 2023). Extensive benchmark results show that DRN-1D2D_Inter significantly outperforms DeepHomo and GLINTER in inter-protein contact prediction, although DRN-1D2D_Inter makes the prediction purely from sequences.…”

“…In almost the same time with DeepHomo2, we proved that embeddings from protein language models(R. Rao et al, 2021;Rives et al, 2021) (PLMs) are very effective features to predict inter-protein contacts for both homomeric and heteromeric PPIs, and we further show the sequence embeddings (ESM-1b (Rives et al, 2021)), MSA embeddings (ESM-MSA-1b(R. M. Rao et al, 2021b) & Position-Specific Scoring Matrix (PSSM)) and the inter-protein coevolutionary information complement each other in the prediction, with which we developed DRN-1D2D_Inter (Si & Yan, 2023). Extensive benchmark results show that DRN-1D2D_Inter significantly outperforms DeepHomo and GLINTER in inter-protein contact prediction, although DRN-1D2D_Inter makes the prediction purely from sequences.…”

“…We used the training set and test sets prepared in our previous work DRN-1D2D_Inter(Si & Yan, 2023) to train and evaluate PLMGraph-Inter. More details for the dataset generation can be found in the previous work.…”

Protein language model-embedded geometric graphs power inter-protein contact prediction

“…Besides, coming from the same group, Lin et al, 2023 further developed DeepHomo2 for inter-protein contact prediction for homomeric PPIs by including the multiple sequence alignment (MSA) embeddings and attentions from an MSA-based protein language model (PLM) (MSA transformer) ( Rao et al, 2021b ) in their prediction model, which further improved the prediction performance. At almost the same time as DeepHomo2, we proved that embeddings from PLMs ( Rao et al, 2021a ; Rives et al, 2021 ) are very effective features in predicting inter-protein contacts for both homomeric and heteromeric PPIs, and we further show the sequence embeddings (ESM-1b [ Rives et al, 2021 ]), MSA embeddings (ESM-MSA-1b [ Rao et al, 2021b ] and Position-Specific Scoring Matrix [PSSM]), and the inter-protein coevolutionary information complement each other in the prediction, with which we developed DRN-1D2D_Inter ( Si and Yan, 2023 ). Extensive benchmark results show that DRN-1D2D_Inter significantly outperforms DeepHomo and GLINTER in inter-protein contact prediction, although DRN-1D2D_Inter makes the prediction purely from sequences.…”

ProAffinity-GNN: A Novel Approach to Structure-based Protein-Protein Binding Affinity Prediction via a Curated Dataset and Graph Neural Networks

“…The model's efficiency and scale make it particularly suitable for contemporary protein research. 38,39 To ensure comprehensive sequence representation, we input the FASTA sequences corresponding to each protein complex into the ESM-2 model, whose outputs are used as node embeddings. This approach allows us to cap-ture the most complete sequence information available.…”