Structural analogue-based protein structure domain assembly assisted by deep learning

M-DeepAssembly: enhanced DeepAssembly based on multi-objective multi-domain protein conformation sampling

BACKGROUND

Association and cooperation among structural domains play an important role in protein function and drug design. Despite remarkable advancements in highly accurate single-domain protein structure prediction through the collaborative efforts of the community using deep learning, challenges still exist in predicting multi-domain protein structures when the evolutionary signal for a given domain pair is weak or the protein structure is large.

RESULTS

To alleviate the above challenges, we proposed M-DeepAssembly, a protocol based on multi-objective protein conformation sampling algorithm for multi-domain protein structure prediction. Firstly, the inter-domain interactions and full-length sequence distance features are extracted through DeepAssembly and AlphaFold2, respectively. Secondly, subject to these features, we constructed a multi-objective energy model and designed a sampling algorithm for exploring and exploiting conformational space to generate ensembles. Finally, the output protein structure was selected from the ensembles using our in-house developed model quality assessment algorithm. On the test set of 164 multi-domain proteins, the results show that the average TM-score of M-DeepAssembly is 15.4% and 2.0% higher than AlphaFold2 and DeepAssembly, respectively. It is worth noting that there are models with higher accuracy in ensembles, achieving an improvement of 20.3% and 6.4% relative to the two baseline methods, although these models were not selected. Furthermore, when compared to the prediction results of AlphaFold2 for CASP15 multi-domain targets, M-DeepAssembly demonstrates certain performance advantages.

CONCLUSIONS

M-DeepAssembly provides a distinctive multi-domain protein assembly algorithm, which can alleviate the current challenges of weak evolutionary signals and large structures to some extent by forming diverse ensembles using multi-objective protein conformation sampling algorithm. The proposed method contributes to exploring the functions of multi-domain proteins, especially providing new insights into targets with multiple conformational states.

DeepAssembly2: A Web Server for Protein Complex Structure Assembly Based on Domain-Domain Interactions

Proteins often perform biological functions by forming complexes, thereby accurately predicting the structure of protein complexes is crucial to understanding and mastering their functions, as well as facilitating drug discovery. Protein monomeric structure prediction has made a breakthrough in recent years, but the accurate prediction of complex structure remains a challenge. In this work, we present DeepAssembly2, a web server for automatically assembling protein complex structure based on domain-domain interactions. First, the features are constructed according to the input complex sequence and monomeric structures, then these features are used to predict the inter-chain residue distance through a deep learning model, and finally, the complex structure is assembled under the guidance of inter-chain residue distances. Compared with the previously developed version, DeepAssembly2 is trained on a newly constructed inter-chain domain-domain interaction dataset. Meanwhile, several important features have been added, such as Interface Residue Propensity and Ultrafast Shape Recognition. In addition, we introduced the inter-chain residue distance from the AlphaFold-Multimer model to further improve the accuracy. Finally, we also integrate our recently developed model quality assessment method to select the output models. The performance of DeepAssembly2 is significantly improved compared with the previous version, and it is expected to provide new insights and an effective tool for drug development, vaccine design, etc. The web server of DeepAssembly2 is freely available at https://zhanglab-bioinf.com/DeepAssembly/.

BINDTI: A Bi-Directional Intention Network for Drug-Target Interaction Identification Based on Attention Mechanisms

The identification of drug-target interactions (DTIs) is an essential step in drug discovery. In vitro experimental methods are expensive, laborious, and time-consuming. Deep learning has witnessed promising progress in DTI prediction. However, how to precisely represent drug and protein features is a major challenge for DTI prediction. Here, we developed an end-to-end DTI identification framework called BINDTI based on bi-directional Intention network. First, drug features are encoded with graph convolutional networks based on its 2D molecular graph obtained by its SMILES string. Next, protein features are encoded based on its amino acid sequence through a mixed model called ACmix, which integrates self-attention mechanism and convolution. Third, drug and target features are fused through bi-directional Intention network, which combines Intention and multi-head attention. Finally, unknown drug-target (DT) pairs are classified through multilayer perceptron based on the fused DT features. The results demonstrate that BINDTI greatly outperformed four baseline methods (i.e., CPI-GNN, TransfomerCPI, MolTrans, and IIFDTI) on the BindingDB, BioSNAP, DrugBank, and Human datasets. More importantly, it was more appropriate to predict new DTIs than the four baseline methods on imbalanced datasets. Ablation experimental results elucidated that both bi-directional Intention and ACmix could greatly advance DTI prediction. The fused feature visualization and case studies manifested that the predicted results by BINDTI were basically consistent with the true ones. We anticipate that the proposed BINDTI framework can find new low-cost drug candidates, improve drugs' virtual screening, and further facilitate drug repositioning as well as drug discovery.

Computational tools to predict context-specific protein complexes

Interactions between thousands of proteins define cells' protein-protein interaction (PPI) network. Some of these interactions lead to the formation of protein complexes. It is challenging to identify a protein complex in a haystack of protein-protein interactions, and it is even more difficult to predict all protein complexes of the complexome. Simulations and machine learning approaches try to crack these problems by looking at the PPI network or predicted protein structures. Clustering of PPI networks led to the first protein complex predictions, while most recently, atomistic models of protein complexes and deep-learning-based structure prediction methods have also emerged. The simulation of PPI level interactions even enables the quantitative prediction of protein complexes. These methods, the required data sources, and their potential future developments are discussed in this review.

Recent Progress of Protein Tertiary Structure Prediction

The prediction of three-dimensional (3D) protein structure from amino acid sequences has stood as a significant challenge in computational and structural bioinformatics for decades. Recently, the widespread integration of artificial intelligence (AI) algorithms has substantially expedited advancements in protein structure prediction, yielding numerous significant milestones. In particular, the end-to-end deep learning method AlphaFold2 has facilitated the rise of structure prediction performance to new heights, regularly competitive with experimental structures in the 14th Critical Assessment of Protein Structure Prediction (CASP14). To provide a comprehensive understanding and guide future research in the field of protein structure prediction for researchers, this review describes various methodologies, assessments, and databases in protein structure prediction, including traditionally used protein structure prediction methods, such as template-based modeling (TBM) and template-free modeling (FM) approaches; recently developed deep learning-based methods, such as contact/distance-guided methods, end-to-end folding methods, and protein language model (PLM)-based methods; multi-domain protein structure prediction methods; the CASP experiments and related assessments; and the recently released AlphaFold Protein Structure Database (AlphaFold DB). We discuss their advantages, disadvantages, and application scopes, aiming to provide researchers with insights through which to understand the limitations, contexts, and effective selections of protein structure prediction methods in protein-related fields.

Recent Advances and Challenges in Protein Structure Prediction

Artificial intelligence has made significant advances in the field of protein structure prediction in recent years. In particular, DeepMind's end-to-end model, AlphaFold2, has demonstrated the capability to predict three-dimensional structures of numerous unknown proteins with accuracy levels comparable to those of experimental methods. This breakthrough has opened up new possibilities for understanding protein structure and function as well as accelerating drug discovery and other applications in the field of biology and medicine. Despite the remarkable achievements of artificial intelligence in the field, there are still some challenges and limitations. In this Review, we discuss the recent progress and some of the challenges in protein structure prediction. These challenges include predicting multidomain protein structures, protein complex structures, multiple conformational states of proteins, and protein folding pathways. Furthermore, we highlight directions in which further improvements can be conducted.

Estimating protein complex model accuracy based on ultrafast shape recognition and deep learning in CASP15

This article reports and analyzes the results of protein complex model accuracy estimation by our methods (DeepUMQA3 and GraphGPSM) in the 15th Critical Assessment of techniques for protein Structure Prediction (CASP15). The new deep learning-based multimeric complex model accuracy estimation methods are proposed based on the ensemble of three-level features coupling with deep residual/graph neural networks. For the input multimeric complex model, we describe it from three levels: overall complex features, intra-monomer features, and inter-monomer features. We designed an overall ultrafast shape recognition (USR) to characterize the relationship between local residues and the overall complex topology, and an inter-monomer USR to characterize the relationship between the residues of one monomer and the topology of other monomers. DeepUMQA3 (Group name: GuijunLab-RocketX) ranked first in the interface residue accuracy estimation of CASP15. The Pearson correlation between the interface residue Local Distance Difference Test (lDDT) predicted by DeepUMQA3 and the real lDDT is 0.570, the only method that exceeds 0.5. Among the top 5 methods, DeepUMQA3 achieved the highest Pearson correlation of lDDT on 25 out of 39 targets. GraphGPSM (Group name: GuijunLab-PAthreader) has TM-score Pearson correlations greater than 0.9 on 14 targets, showing a good ability to estimate the overall fold accuracy. The DeepUMQA3 server is available at http://zhanglab-bioinf.com/DeepUMQA/ and the GraphGPSM server is available at http://zhanglab-bioinf.com/GraphGPSM/.

Multi-domain and complex protein structure prediction using inter-domain interactions from deep learning

Accurately capturing domain-domain interactions is key to understanding protein function and designing structure-based drugs. Although AlphaFold2 has made a breakthrough on single domain, it should be noted that the structure modeling for multi-domain protein and complex remains a challenge. In this study, we developed a multi-domain and complex structure assembly protocol, named DeepAssembly, based on domain segmentation and single domain modeling algorithms. Firstly, DeepAssembly uses a population-based evolutionary algorithm to assemble multi-domain proteins by inter-domain interactions inferred from a developed deep learning network. Secondly, protein complexes are assembled by means of domains rather than chains using DeepAssembly. Experimental results show that on 219 multi-domain proteins, the average inter-domain distance precision by DeepAssembly is 22.7% higher than that of AlphaFold2. Moreover, DeepAssembly improves accuracy by 13.1% for 164 multi-domain structures with low confidence deposited in AlphaFold database. We apply DeepAssembly for the prediction of 247 heterodimers. We find that DeepAssembly successfully predicts the interface (DockQ ≥ 0.23) for 32.4% of the dimers, suggesting a lighter way to assemble complex structures by treating domains as assembly units and using inter-domain interactions learned from monomer structures.

E2EDA: Protein Domain Assembly Based on End-to-End Deep Learning

With the development of deep learning, almost all single-domain proteins can be predicted at experimental resolution. However, the structure prediction of multi-domain proteins remains a challenge. Achieving end-to-end protein domain assembly and further improving the accuracy of the full-chain modeling by accurately predicting inter-domain orientation while improving the assembly efficiency will provide significant insights into structure-based drug discovery. In this work, we propose an End-to-End Domain Assembly method based on deep learning, named E2EDA. We first develop RMNet, an EfficientNetV2-based deep learning model that fuses multiple features using an attention mechanism to predict inter-domain rigid motion. Then, the predicted rigid motions are transformed into inter-domain spatial transformations to directly assemble the full-chain model. Finally, the scoring strategy RMscore is designed to select the best model from multiple assembled models. The experimental results show that the average TM-score of the model assembled by E2EDA on the benchmark set (282) is 0.827, which is better than those of other domain assembly methods SADA (0.792) and DEMO (0.730). Meanwhile, on our constructed multi-domain data set from AlphaFold DB, the model reassembled by E2EDA is 7.0% higher in TM-score compared to the full-chain model predicted by AlphaFold2, indicating that E2EDA can capture more accurate inter-domain orientations to improve the quality of the model predicted by AlphaFold2. Furthermore, compared to SADA and AlphaFold2, E2EDA reduced the average runtime on the benchmark by 64.7% and 19.2%, respectively, indicating that E2EDA can significantly improve assembly efficiency through an end-to-end approach. The online server is available at http://zhanglab-bioinf.com/E2EDA.

Predicting potential microbe-disease associations with graph attention autoencoder, positive-unlabeled learning, and deep neural network

Background

Microbes have dense linkages with human diseases. Balanced microorganisms protect human body against physiological disorders while unbalanced ones may cause diseases. Thus, identification of potential associations between microbes and diseases can contribute to the diagnosis and therapy of various complex diseases. Biological experiments for microbe-disease association (MDA) prediction are expensive, time-consuming, and labor-intensive.

Methods

We developed a computational MDA prediction method called GPUDMDA by combining graph attention autoencoder, positive-unlabeled learning, and deep neural network. First, GPUDMDA computes disease similarity and microbe similarity matrices by integrating their functional similarity and Gaussian association profile kernel similarity, respectively. Next, it learns the feature representation of each microbe-disease pair using graph attention autoencoder based on the obtained disease similarity and microbe similarity matrices. Third, it selects a few reliable negative MDAs based on positive-unlabeled learning. Finally, it takes the learned MDA features and the selected negative MDAs as inputs and designed a deep neural network to predict potential MDAs.

Results

GPUDMDA was compared with four state-of-the-art MDA identification models (i.e., MNNMDA, GATMDA, LRLSHMDA, and NTSHMDA) on the HMDAD and Disbiome databases under five-fold cross validations on microbes, diseases, and microbe-disease pairs. Under the three five-fold cross validations, GPUDMDA computed the best AUCs of 0.7121, 0.9454, and 0.9501 on the HMDAD database and 0.8372, 0.8908, and 0.8948 on the Disbiome database, respectively, outperforming the other four MDA prediction methods. Asthma is the most common chronic respiratory condition and affects ~339 million people worldwide. Inflammatory bowel disease is a class of globally chronic intestinal disease widely existed in the gut and gastrointestinal tract and extraintestinal organs of patients. Particularly, inflammatory bowel disease severely affects the growth and development of children. We used the proposed GPUDMDA method and found that Enterobacter hormaechei had potential associations with both asthma and inflammatory bowel disease and need further biological experimental validation.

Conclusion

The proposed GPUDMDA demonstrated the powerful MDA prediction ability. We anticipate that GPUDMDA helps screen the therapeutic clues for microbe-related diseases.

Identifying potential drug-target interactions based on ensemble deep learning

Introduction

Drug-target interaction prediction is one important step in drug research and development. Experimental methods are time consuming and laborious.

Methods

In this study, we developed a novel DTI prediction method called EnGDD by combining initial feature acquisition, dimensional reduction, and DTI classification based on Gradient boosting neural network, Deep neural network, and Deep Forest.

Results

EnGDD was compared with seven stat-of-the-art DTI prediction methods (BLM-NII, NRLMF, WNNGIP, NEDTP, DTi2Vec, RoFDT, and MolTrans) on the nuclear receptor, GPCR, ion channel, and enzyme datasets under cross validations on drugs, targets, and drug-target pairs, respectively. EnGDD computed the best recall, accuracy, F1-score, AUC, and AUPR under the majority of conditions, demonstrating its powerful DTI identification performance. EnGDD predicted that D00182 and hsa2099, D07871 and hsa1813, DB00599 and hsa2562, D00002 and hsa10935 have a higher interaction probabilities among unknown drug-target pairs and may be potential DTIs on the four datasets, respectively. In particular, D00002 (Nadide) was identified to interact with hsa10935 (Mitochondrial peroxiredoxin3) whose up-regulation might be used to treat neurodegenerative diseases. Finally, EnGDD was used to find possible drug targets for Parkinson's disease and Alzheimer's disease after confirming its DTI identification performance. The results show that D01277, D04641, and D08969 may be applied to the treatment of Parkinson's disease through targeting hsa1813 (dopamine receptor D2) and D02173, D02558, and D03822 may be the clues of treatment for patients with Alzheimer's disease through targeting hsa5743 (prostaglandinendoperoxide synthase 2). The above prediction results need further biomedical validation.

Discussion

We anticipate that our proposed EnGDD model can help discover potential therapeutic clues for various diseases including neurodegenerative diseases.

GraphGPSM: a global scoring model for protein structure using graph neural networks

The scoring models used for protein structure modeling and ranking are mainly divided into unified field and protein-specific scoring functions. Although protein structure prediction has made tremendous progress since CASP14, the modeling accuracy still cannot meet the requirements to a certain extent. Especially, accurate modeling of multi-domain and orphan proteins remains a challenge. Therefore, an accurate and efficient protein scoring model should be developed urgently to guide the protein structure folding or ranking through deep learning. In this work, we propose a protein structure global scoring model based on equivariant graph neural network (EGNN), named GraphGPSM, to guide protein structure modeling and ranking. We construct an EGNN architecture, and a message passing mechanism is designed to update and transmit information between nodes and edges of the graph. Finally, the global score of the protein model is output through a multilayer perceptron. Residue-level ultrafast shape recognition is used to describe the relationship between residues and the overall structure topology, and distance and direction encoded by Gaussian radial basis functions are designed to represent the overall topology of the protein backbone. These two features are combined with Rosetta energy terms, backbone dihedral angles and inter-residue distance and orientations to represent the protein model and embedded into the nodes and edges of the graph neural network. The experimental results on the CASP13, CASP14 and CAMEO test sets show that the scores of our developed GraphGPSM have a strong correlation with the TM-score of the models, which are significantly better than those of the unified field score function REF2015 and the state-of-the-art local lDDT-based scoring models ModFOLD8, ProQ3D and DeepAccNet, etc. The modeling experimental results on 484 test proteins demonstrate that GraphGPSM can greatly improve the modeling accuracy. GraphGPSM is further used to model 35 orphan proteins and 57 multi-domain proteins. The results show that the average TM-score of the models predicted by GraphGPSM is 13.2 and 7.1% higher than that of the models predicted by AlphaFold2. GraphGPSM also participates in CASP15 and achieves competitive performance in global accuracy estimation.

AlphaFold, allosteric, and orthosteric drug discovery: Ways forward

Drug discovery is arguably a highly challenging and significant interdisciplinary aim. The stunning success of the artificial intelligence-powered AlphaFold, whose latest version is buttressed by an innovative machine-learning approach that integrates physical and biological knowledge about protein structures, raised drug discovery hopes that unsurprisingly, have not come to bear. Even though accurate, the models are rigid, including the drug pockets. AlphaFold's mixed performance poses the question of how its power can be harnessed in drug discovery. Here we discuss possible ways of going forward wielding its strengths, while bearing in mind what AlphaFold can and cannot do. For kinases and receptors, an input enriched in active (ON) state models can better AlphaFold's chance of rational drug design success.