Deep learning in modeling protein complex structures: From contact prediction to end-to-end approaches.

Protein-protein interactions play crucial roles in many biological processes. Traditionally, protein complex structures are normally built by protein-protein docking. With the rapid development of artificial intelligence and its great success in monomer protein structure prediction, deep learning has widely been applied to modeling protein-protein complex structures through inter-protein contact prediction and end-to-end approaches in the past few years. This article reviews the recent advances of deep-learning-based approaches in modeling protein-protein complex structures as well as their advantages and limitations. Challenges and possible future directions are also briefly discussed in applying deep learning for the prediction of protein complex structures.

Impact of AlphaFold on structure prediction of protein complexes: The CASP15-CAPRI experiment.

We present the results for CAPRI Round 54, the 5th joint CASP-CAPRI protein assembly prediction challenge. The Round offered 37 targets, including 14 homodimers, 3 homo-trimers, 13 heterodimers including 3 antibody-antigen complexes, and 7 large assemblies. On average ~70 CASP and CAPRI predictor groups, including more than 20 automatics servers, submitted models for each target. A total of 21 941 models submitted by these groups and by 15 CAPRI scorer groups were evaluated using the CAPRI model quality measures and the DockQ score consolidating these measures. The prediction performance was quantified by a weighted score based on the number of models of acceptable quality or higher submitted by each group among their five best models. Results show substantial progress achieved across a significant fraction of the 60+ participating groups. High-quality models were produced for about 40% of the targets compared to 8% two years earlier. This remarkable improvement is due to the wide use of the AlphaFold2 and AlphaFold2-Multimer software and the confidence metrics they provide. Notably, expanded sampling of candidate solutions by manipulating these deep learning inference engines, enriching multiple sequence alignments, or integration of advanced modeling tools, enabled top performing groups to exceed the performance of a standard AlphaFold2-Multimer version used as a yard stick. This notwithstanding, performance remained poor for complexes with antibodies and nanobodies, where evolutionary relationships between the binding partners are lacking, and for complexes featuring conformational flexibility, clearly indicating that the prediction of protein complexes remains a challenging problem.

Deep transfer learning for inter-chain contact predictions of transmembrane protein complexes.

Membrane proteins are encoded by approximately a quarter of human genes. Inter-chain residue-residue contact information is important for structure prediction of membrane protein complexes and valuable for understanding their molecular mechanism. Although many deep learning methods have been proposed to predict the intra-protein contacts or helix-helix interactions in membrane proteins, it is still challenging to accurately predict their inter-chain contacts due to the limited number of transmembrane proteins. Addressing the challenge, here we develop a deep transfer learning method for predicting inter-chain contacts of transmembrane protein complexes, named DeepTMP, by taking advantage of the knowledge pre-trained from a large data set of non-transmembrane proteins. DeepTMP utilizes a geometric triangle-aware module to capture the correct inter-chain interaction from the coevolution information generated by protein language models. DeepTMP is extensively evaluated on a test set of 52 self-associated transmembrane protein complexes, and compared with state-of-the-art methods including DeepHomo2.0, CDPred, GLINTER, DeepHomo, and DNCON2_Inter. It is shown that DeepTMP considerably improves the precision of inter-chain contact prediction and outperforms the existing approaches in both accuracy and robustness.

DeepHomo2.0: improved protein-protein contact prediction of homodimers by transformer-enhanced deep learning.

Protein-protein interactions play an important role in many biological processes. However, although structure prediction for monomer proteins has achieved great progress with the advent of advanced deep learning algorithms like AlphaFold, the structure prediction for protein-protein complexes remains an open question. Taking advantage of the Transformer model of ESM-MSA, we have developed a deep learning-based model, named DeepHomo2.0, to predict protein-protein interactions of homodimeric complexes by leveraging the direct-coupling analysis (DCA) and Transformer features of sequences and the structure features of monomers. DeepHomo2.0 was extensively evaluated on diverse test sets and compared with eight state-of-the-art methods including protein language model-based, DCA-based and machine learning-based methods. It was shown that DeepHomo2.0 achieved a high precision of >70% with experimental monomer structures and >60% with predicted monomer structures for the top 10 predicted contacts on the test sets and outperformed the other eight methods. Moreover, even the version without using structure information, named DeepHomoSeq, still achieved a good precision of >55% for the top 10 predicted contacts. Integrating the predicted contacts into protein docking significantly improved the structure prediction of realistic Critical Assessment of Protein Structure Prediction homodimeric complexes. DeepHomo2.0 and DeepHomoSeq are available at http://huanglab.phys.hust.edu.cn/DeepHomo2/.

Docking cyclic peptides formed by a disulfide bond through a hierarchical strategy.

MOTIVATION: Cyclization is a common strategy to enhance the therapeutic potential of peptides. Many cyclic peptide drugs have been approved for clinical use, in which the disulfide-driven cyclic peptide is one of the most prevalent categories. Molecular docking is a powerful computational method to predict the binding modes of molecules. For protein-cyclic peptide docking, a big challenge is considering the flexibility of peptides with conformers constrained by cyclization. RESULTS: Integrating our efficient peptide 3D conformation sampling algorithm MODPEP2.0 and knowledge-based scoring function ITScorePP, we have proposed an extended version of our hierarchical peptide docking algorithm, named HPEPDOCK2.0, to predict the binding modes of the peptide cyclized through a disulfide against a protein. Our HPEPDOCK2.0 approach was extensively evaluated on diverse test sets and compared with the state-of-the-art cyclic peptide docking program AutoDock CrankPep (ADCP). On a benchmark dataset of 18 cyclic peptide-protein complexes, HPEPDOCK2.0 obtained a native contact fraction of above 0.5 for 61% of the cases when the top prediction was considered, compared with 39% for ADCP. On a larger test set of 25 cyclic peptide-protein complexes, HPEPDOCK2.0 yielded a success rate of 44% for the top prediction, compared with 20% for ADCP. In addition, HPEPDOCK2.0 was also validated on two other test sets of 10 and 11 complexes with apo and predicted receptor structures, respectively. HPEPDOCK2.0 is computationally efficient and the average running time for docking a cyclic peptide is about 34 min on a single CPU core, compared with 496 min for ADCP. HPEPDOCK2.0 will facilitate the study of the interaction between cyclic peptides and proteins and the development of therapeutic cyclic peptide drugs. AVAILABILITY AND IMPLEMENTATION: http://huanglab.phys.hust.edu.cn/hpepdock/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Model building of protein complexes from intermediate-resolution cryo-EM maps with deep learning-guided automatic assembly.

Advances in microscopy instruments and image processing algorithms have led to an increasing number of cryo-electron microscopy (cryo-EM) maps. However, building accurate models into intermediate-resolution EM maps remains challenging and labor-intensive. Here, we propose an automatic model building method of multi-chain protein complexes from intermediate-resolution cryo-EM maps, named EMBuild, by integrating AlphaFold structure prediction, FFT-based global fitting, domain-based semi-flexible refinement, and graph-based iterative assembling on the main-chain probability map predicted by a deep convolutional network. EMBuild is extensively evaluated on diverse test sets of 47 single-particle EM maps at 4.0-8.0 Å resolution and 16 subtomogram averaging maps of cryo-ET data at 3.7-9.3 Å resolution, and compared with state-of-the-art approaches. We demonstrate that EMBuild is able to build high-quality complex structures that are comparably accurate to the manually built PDB structures from the cryo-EM maps. These results demonstrate the accuracy and reliability of EMBuild in automatic model building.

Efficient 3D conformer generation of cyclic peptides formed by a disulfide bond.

Cyclic peptides formed by disulfide bonds have been one large group of common drug candidates in drug development. Structural information of a peptide is essential to understand its interaction with its target. However, due to the high flexibility of peptides, it is difficult to sample the near-native conformations of a peptide. Here, we have developed an extended version of our MODPEP approach, named MODPEP2.0, to fast generate the conformations of cyclic peptides formed by a disulfide bond. MODPEP2.0 builds the three-dimensional (3D) structures of a cyclic peptide from scratch by assembling amino acids one by one onto the cyclic fragment based on the constructed rotamer and cyclic backbone libraries. Being tested on a data set of 193 diverse cyclic peptides, MODPEP2.0 obtained a considerable advantage in both accuracy and computational efficiency, compared with other sampling algorithms including PEP-FOLD, ETKDG, and modified ETKDG (mETKDG). MODPEP2.0 achieved a high sampling accuracy with an average C[Formula: see text] RMSD of 2.20 Å and 1.66 Å when 10 and 100 conformations were considered, respectively, compared with 3.41 Å and 2.62 Å for PEP-FOLD, 3.44 Å and 3.16 Å for ETKDG, 3.09 Å and 2.72 Å for mETKDG. MODPEP2.0 also reproduced experimental peptide structures for 81.35% of the test cases when an ensemble of 100 conformations were considered, compared with 54.95%, 37.50% and 50.00% for PEP-FOLD, ETKDG, and mETKDG. MODPEP2.0 is computationally efficient and can generate 100 peptide conformations in one second. MODPEP2.0 will be useful in sampling cyclic peptide structures and modeling related protein-peptide interactions, facilitating the development of cyclic peptide drugs.

TRScore: a 3D RepVGG-based scoring method for ranking protein docking models.

MOTIVATION: Protein-protein interactions (PPI) play important roles in cellular activities. Due to the technical difficulty and high cost of experimental methods, there are considerable interests towards the development of computational approaches, such as protein docking, to decipher PPI patterns. One of the important and difficult aspects in protein docking is recognizing near-native conformations from a set of decoys, but unfortunately, traditional scoring functions still suffer from limited accuracy. Therefore, new scoring methods are pressingly needed in methodological and/or practical implications. RESULTS: We present a new deep learning-based scoring method for ranking protein-protein docking models based on a 3D RepVGG network, named TRScore. To recognize near-native conformations from a set of decoys, TRScore voxelizes the protein-protein interface into a 3D grid labeled by the number of atoms in different physicochemical classes. Benefiting from the deep convolutional RepVGG architecture, TRScore can effectively capture the subtle differences between energetically favorable near-native models and unfavorable non-native decoys without needing extra information. TRScore was extensively evaluated on diverse test sets including protein-protein docking benchmark 5.0 update set, DockGround decoy set, as well as realistic CAPRI decoy set and overall obtained a significant improvement over existing methods in cross-validation and independent evaluations. AVAILABILITY AND IMPLEMENTATION: Codes available at: https://github.com/BioinformaticsCSU/TRScore.

Challenges and opportunities of automated protein-protein docking: HDOCK server vs human predictions in CAPRI Rounds 38-46.

Protein-protein docking plays an important role in the computational prediction of the complex structure between two proteins. For years, a variety of docking algorithms have been developed, as witnessed by the critical assessment of prediction interactions (CAPRI) experiments. However, despite their successes, many docking algorithms often require a series of manual operations like modeling structures from sequences, incorporating biological information, and selecting final models. The difficulties in these manual steps have significantly limited the applications of protein-protein docking, as most of the users in the community are nonexperts in docking. Therefore, automated docking like a web server, which can give a comparable performance to human docking protocol, is pressingly needed. As such, we have participated in the blind CAPRI experiments for Rounds 38-45 and CASP13-CAPRI challenge for Round 46 with both our HDOCK automated docking web server and human docking protocol. It was shown that our HDOCK server achieved an "acceptable" or higher CAPRI-rated model in the top 10 submitted predictions for 65.5% and 59.1% of the targets in the docking experiments of CAPRI and CASP13-CAPRI, respectively, which are comparable to 66.7% and 54.5% for human docking protocol. Similar trends can also be observed in the scoring experiments. These results validated our HDOCK server as an efficient automated docking protocol for nonexpert users. Challenges and opportunities of automated docking are also discussed.