Fast and effective protein model refinement using deep graph neural networks

Recent advances and challenges in protein complex model accuracy estimation

Estimation of model accuracy plays a crucial role in protein structure prediction, aiming to evaluate the quality of predicted protein structure models accurately and objectively. This process is not only key to screening candidate models that are close to the real structure, but also provides guidance for further optimization of protein structures. With the significant advancements made by AlphaFold2 in monomer structure, the problem of single-domain protein structure prediction has been widely solved. Correspondingly, the importance of assessing the quality of single-domain protein models decreased, and the research focus has shifted to estimation of model accuracy of protein complexes. In this review, our goal is to provide a comprehensive overview of the reference and statistical metrics, as well as representative methods, and the current challenges within four distinct facets (Topology Global Score, Interface Total Score, Interface Residue-Wise Score, and Tertiary Residue-Wise Score) in the field of complex EMA.

Deep Learning-Assisted Spectrum-Structure Correlation: State-of-the-Art and Perspectives

Spectrum-structure correlation is playing an increasingly crucial role in spectral analysis and has undergone significant development in recent decades. With the advancement of spectrometers, the high-throughput detection triggers the explosive growth of spectral data, and the research extension from small molecules to biomolecules accompanies massive chemical space. Facing the evolving landscape of spectrum-structure correlation, conventional chemometrics becomes ill-equipped, and deep learning assisted chemometrics rapidly emerges as a flourishing approach with superior ability of extracting latent features and making precise predictions. In this review, the molecular and spectral representations and fundamental knowledge of deep learning are first introduced. We then summarize the development of how deep learning assist to establish the correlation between spectrum and molecular structure in the recent 5 years, by empowering spectral prediction (i.e., forward structure-spectrum correlation) and further enabling library matching and de novo molecular generation (i.e., inverse spectrum-structure correlation). Finally, we highlight the most important open issues persisted with corresponding potential solutions. With the fast development of deep learning, it is expected to see ultimate solution of establishing spectrum-structure correlation soon, which would trigger substantial development of various disciplines.

Selection among site-dependent structurally constrained substitution models of protein evolution by approximate Bayesian computation

MOTIVATION

The selection among substitution models of molecular evolution is fundamental for obtaining accurate phylogenetic inferences. At the protein level, evolutionary analyses are traditionally based on empirical substitution models but these models make unrealistic assumptions and are being surpassed by structurally constrained substitution (SCS) models. The SCS models often consider site-dependent evolution, a process that provides realism but complicates their implementation into likelihood functions that are commonly used for substitution model selection.

RESULTS

We present a method to perform selection among site-dependent SCS models, also among empirical and site-dependent SCS models, based on the approximate Bayesian computation (ABC) approach and its implementation into the computational framework ProteinModelerABC. The framework implements ABC with and without regression adjustments and includes diverse empirical and site-dependent SCS models of protein evolution. Using extensive simulated data, we found that it provides selection among SCS and empirical models with acceptable accuracy. As illustrative examples, we applied the framework to analyze a variety of protein families observing that SCS models fit them better than the corresponding best-fitting empirical substitution models.

AVAILABILITY AND IMPLEMENTATION

ProteinModelerABC is freely available from https://github.com/DavidFerreiro/ProteinModelerABC, can run in parallel and includes a graphical user interface. The framework is distributed with detailed documentation and ready-to-use examples.

Identification novel salt-enhancing peptides from largemouth bass and exploration their action mechanism with transmembrane channel-like 4 (TMC4) by molecular simulation

The purpose of this study was to screen and verify salt-enhancing peptides that can effectively reduce sodium consumption from Largemouth bass myosin through virtual hydrolysis, molecular simulation, and sensory evaluation. The human transmembrane channel-like 4 (TMC4) was constructed using Alphafold2, with 93.3 % of amino acids falling within allowed regions. A total of 19 peptides were predicted through virtual hydrolysis and screening. DAF, QIF, RPAL, and IPVM significantly enhanced the saltiness perception, and QIF exhibited the most pronounced effect in enhancing saltiness (P < 0.05). The residues Ala258, Ser546, Ser603, Phe259, Cys265, Glu539, Lys278 and Ser585 were identified as key binding sites. The TMC4-DAF complex achieved stability after 20, 000 ps, exhibiting an average RMSD value of 0.84 nm. DAF consistently displayed fluctuations at approximately 3.05 nm, and the number of hydrogen bonds varied between 3 and 5. These results suggested that Alphafold2 modelling can be used for predicting salt-enhancing peptides.

Amalgamated Pharmacoinformatics Study to Investigate the Mechanism of Xiao Jianzhong Tang against Chronic Atrophic Gastritis

BACKGROUND

Traditional Chinese medicine (TCM) Xiao Jianzhong Tang (XJZ) has a favorable efficacy in the treatment of chronic atrophic gastritis (CAG). However, its pharmacological mechanism has not been fully explained.

OBJECTIVE

The purpose of this study was to find the potential mechanism of XJZ in the treatment of CAG using pharmacocoinformatics approaches.

METHODS

Network pharmacology was used to screen out the key compounds and key targets, MODELLER and GNNRefine were used to repair and refine proteins, Autodock vina was employed to perform molecular docking, Δ Lin_F9XGB was used to score the docking results, and Gromacs was used to perform molecular dynamics simulations (MD).

RESULTS

Kaempferol, licochalcone A, and naringenin, were obtained as key compounds, while AKT1, MAPK1, MAPK14, RELA, STAT1, and STAT3 were acquired as key targets. Among docking results, 12 complexes scored greater than five. They were run for 50ns MD. The free binding energy of AKT1-licochalcone A and MAPK1-licochalcone A was less than -15 kcal/mol and AKT1-naringenin and STAT3-licochalcone A was less than -9 kcal/mol. These complexes were crucial in XJZ treating CAG.

CONCLUSION

Our findings suggest that licochalcone A could act on AKT1, MAPK1, and STAT3, and naringenin could act on AKT1 to play the potential therapeutic effect on CAG. The work also provides a powerful approach to interpreting the complex mechanism of TCM through the amalgamation of network pharmacology, deep learning-based protein refinement, molecular docking, machine learning-based binding affinity estimation, MD simulations, and MM-PBSA-based estimation of binding free energy.

Assessment of the Performances of the Protein Modeling Techniques Participating in CASP15 Using a Structure-Based Functional Site Prediction Approach: ResiRole

BACKGROUND

Model quality assessments via computational methods which entail comparisons of the modeled structures to the experimentally determined structures are essential in the field of protein structure prediction. The assessments provide means to benchmark the accuracies of the modeling techniques and to aid with their development. We previously described the ResiRole method to gauge model quality principally based on the preservation of the structural characteristics described in SeqFEATURE functional site prediction models.

METHODS

We apply ResiRole to benchmark modeling group performances in the Critical Assessment of Structure Prediction experiment, round 15. To gauge model quality, a normalized Predicted Functional site Similarity Score (PFSS) was calculated as the average of one minus the absolute values of the differences of the functional site prediction probabilities, as found for the experimental structures versus those found at the corresponding sites in the structure models.

RESULTS

The average PFSS per modeling group (gPFSS) correlates with standard quality metrics, and can effectively be used to rank the accuracies of the groups. For the free modeling (FM) category, correlation coefficients of the Local Distance Difference Test (LDDT) and Global Distance Test-Total Score (GDT-TS) metrics with gPFSS were 0.98239 and 0.87691, respectively. An example finding for a specific group is that the gPFSS for EMBER3D was higher than expected based on the predictive relationship between gPFSS and LDDT. We infer the result is due to the use of constraints imprinted by function that are a part of the EMBER3D methodology. Also, we find functional site predictions that may guide further functional characterizations of the respective proteins.

CONCLUSION

The gPFSS metric provides an effective means to assess and rank the performances of the structure prediction techniques according to their abilities to accurately recount the structural features at predicted functional sites.

Low-Data Drug Design with Few-Shot Generative Domain Adaptation

Developing new drugs for emerging diseases, such as COVID-19, is crucial for promoting public health. In recent years, the application of artificial intelligence (AI) has significantly advanced drug discovery pipelines. Generative models, such as generative adversarial networks (GANs), exhibit the potential for discovering novel drug molecules by relying on a vast number of training samples. However, for new diseases, only a few samples are typically available, posing a significant challenge to learning a generative model that produces both high-quality and diverse molecules under limited supervision. To address this low-data drug generation issue, we propose a novel molecule generative domain adaptation paradigm (Mol-GenDA), which transfers a pre-trained GAN on a large-scale drug molecule dataset to a new disease domain using only a few references. Specifically, we introduce a molecule adaptor into the GAN generator during the fine tuning, allowing the generator to reuse prior knowledge learned in pre-training to the greatest extent and maintain the quality and diversity of the generated molecules. Comprehensive downstream experiments demonstrate that Mol-GenDA can produce high-quality and diverse drug candidates. In summary, the proposed approach offers a promising solution to expedite drug discovery for new diseases, which could lead to the timely development of effective drugs to combat emerging outbreaks.

iQDeep: an integrated web server for protein scoring using multiscale deep learning models

The remarkable recent advances in protein structure prediction have enabled computational modeling of protein structures with considerably higher accuracy than ever before. While state-of-the-art structure prediction methods provide self-assessment confidence scores of their own predictions, an independent and open-access system for protein scoring is still needed that can be applied to a broad range of predictive modeling scenarios. Here, we present iQDeep, an integrated and highly customizable web server for protein scoring, freely available at http://fusion.cs.vt.edu/iQDeep. The underlying method of iQDeep employs multiscale deep residual neural networks (ResNets) to perform residue-level error classifications, and then probabilistically combines the error classifications for protein scoring. By adjusting the error resolutions, our method can reliably estimate the standard- or high-accuracy variants of the Global Distance Test metric for versatile protein scoring. The performance of the method has been extensively tested and compared against the state-of-the-art approaches in multiple rounds of Critical Assessment of Techniques for Protein Structure Prediction (CASP) experiments including benchmark assessment in CASP12 and CASP13 as well as blind evaluation in CASP14. The iQDeep web server offers a number of convenient features, including (i) the choice of individual and batch processing modes; (ii) an interactive and privacy-preserving web interface for automated job submission, tracking, and results retrieval; (iii) web-based quantitative and visual analyses of the results including overall estimated score and its residue-wise breakdown along with agreements between various sequence- and structural-level features; (iv) extensive help information on job submission and results interpretation via web-based tutorial and help tooltips.

Different Recognition of Protein Features Depending on Deep Learning Models: A Case Study of Aromatic Decarboxylase UbiD

The number of unannotated protein sequences is explosively increasing due to genome sequence technology. A more comprehensive understanding of protein functions for protein annotation requires the discovery of new features that cannot be captured from conventional methods. Deep learning can extract important features from input data and predict protein functions based on the features. Here, protein feature vectors generated by 3 deep learning models are analyzed using Integrated Gradients to explore important features of amino acid sites. As a case study, prediction and feature extraction models for UbiD enzymes were built using these models. The important amino acid residues extracted from the models were different from secondary structures, conserved regions and active sites of known UbiD information. Interestingly, the different amino acid residues within UbiD sequences were regarded as important factors depending on the type of models and sequences. The Transformer models focused on more specific regions than the other models. These results suggest that each deep learning model understands protein features with different aspects from existing knowledge and has the potential to discover new laws of protein functions. This study will help to extract new protein features for the other protein annotations.

Atomic protein structure refinement using all-atom graph representations and SE(3)-equivariant graph transformer

MOTIVATION

The state-of-art protein structure prediction methods such as AlphaFold are being widely used to predict structures of uncharacterized proteins in biomedical research. There is a significant need to further improve the quality and nativeness of the predicted structures to enhance their usability. In this work, we develop ATOMRefine, a deep learning-based, end-to-end, all-atom protein structural model refinement method. It uses a SE(3)-equivariant graph transformer network to directly refine protein atomic coordinates in a predicted tertiary structure represented as a molecular graph.

RESULTS

The method is first trained and tested on the structural models in AlphaFoldDB whose experimental structures are known, and then blindly tested on 69 CASP14 regular targets and 7 CASP14 refinement targets. ATOMRefine improves the quality of both backbone atoms and all-atom conformation of the initial structural models generated by AlphaFold. It also performs better than two state-of-the-art refinement methods in multiple evaluation metrics including an all-atom model quality score-the MolProbity score based on the analysis of all-atom contacts, bond length, atom clashes, torsion angles, and side-chain rotamers. As ATOMRefine can refine a protein structure quickly, it provides a viable, fast solution for improving protein geometry and fixing structural errors of predicted structures through direct coordinate refinement.

AVAILABILITY AND IMPLEMENTATION

The source code of ATOMRefine is available in the GitHub repository (https://github.com/BioinfoMachineLearning/ATOMRefine). All the required data for training and testing are available at https://doi.org/10.5281/zenodo.6944368.

High-accuracy protein model quality assessment using attention graph neural networks

Great improvement has been brought to protein tertiary structure prediction through deep learning. It is important but very challenging to accurately rank and score decoy structures predicted by different models. CASP14 results show that existing quality assessment (QA) approaches lag behind the development of protein structure prediction methods, where almost all existing QA models degrade in accuracy when the target is a decoy of high quality. How to give an accurate assessment to high-accuracy decoys is particularly useful with the available of accurate structure prediction methods. Here we propose a fast and effective single-model QA method, QATEN, which can evaluate decoys only by their topological characteristics and atomic types. Our model uses graph neural networks and attention mechanisms to evaluate global and amino acid level scores, and uses specific loss functions to constrain the network to focus more on high-precision decoys and protein domains. On the CASP14 evaluation decoys, QATEN performs better than other QA models under all correlation coefficients when targeting average LDDT. QATEN shows promising performance when considering only high-accuracy decoys. Compared to the embedded evaluation modules of predicted ${C}_{\alpha^{-}} RMSD$ (pRMSD) in RosettaFold and predicted LDDT (pLDDT) in AlphaFold2, QATEN is complementary and capable of achieving better evaluation on some decoy structures generated by AlphaFold2 and RosettaFold. These results suggest that the new QATEN approach can be used as a reliable independent assessment algorithm for high-accuracy protein structure decoys.

BayeStab: Predicting effects of mutations on protein stability with uncertainty quantification

Predicting protein thermostability change upon mutation is crucial for understanding diseases and designing therapeutics. However, accurately estimating Gibbs free energy change of the protein remained a challenge. Some methods struggle to generalize on examples with no homology and produce uncalibrated predictions. Here we leverage advances in graph neural networks for protein feature extraction to tackle this structure-property prediction task. Our method, BayeStab, is then tested on four test datasets, including S669, S611, S350, and Myoglobin, showing high generalization and symmetry performance. Meanwhile, we apply concrete dropout enabled Bayesian neural networks to infer plausible models and estimate uncertainty. By decomposing the uncertainty into parts induced by data noise and model, we demonstrate that the probabilistic method allows insights into the inherent noise of the training datasets, which is closely relevant to the upper bound of the task. Finally, the BayeStab web server is created and can be found at: http://www.bayestab.com. The code for this work is available at: https://github.com/HongzhouTang/BayeStab.

Chemistry-informed molecular graph as reaction descriptor for machine-learned retrosynthesis planning

Infusing "chemical wisdom" should improve the data-driven approaches that rely exclusively on historical synthetic data for automatic retrosynthesis planning. For this purpose, we designed a chemistry-informed molecular graph (CIMG) to describe chemical reactions. A collection of key information that is most relevant to chemical reactions is integrated in CIMG:NMR chemical shifts as vertex features, bond dissociation energies as edge features, and solvent/catalyst information as global features. For any given compound as a target, a product CIMG is generated and exploited by a graph neural network (GNN) model to choose reaction template(s) leading to this product. A reactant CIMG is then inferred and used in two GNN models to select appropriate catalyst and solvent, respectively. Finally, a fourth GNN model compares the two CIMG descriptors to check the plausibility of the proposed reaction. A reaction vector is obtained for every molecule in training these models. The chemical wisdom of reaction propensity contained in the pretrained reaction vectors is exploited to autocategorize molecules/reactions and to accelerate Monte Carlo tree search (MCTS) for multistep retrosynthesis planning. Full synthetic routes with recommended catalysts/solvents are predicted efficiently using this CIMG-based approach.

NanoNet: Rapid and accurate end-to-end nanobody modeling by deep learning

Antibodies are a rapidly growing class of therapeutics. Recently, single domain camelid VHH antibodies, and their recognition nanobody domain (Nb) appeared as a cost-effective highly stable alternative to full-length antibodies. There is a growing need for high-throughput epitope mapping based on accurate structural modeling of the variable domains that share a common fold and differ in the Complementarity Determining Regions (CDRs). We develop a deep learning end-to-end model, NanoNet, that given a sequence directly produces the 3D coordinates of the backbone and Cβ atoms of the entire VH domain. For the Nb test set, NanoNet achieves 3.16Å average RMSD for the most variable CDR3 loops and 2.65Å, 1.73Å for the CDR1, CDR2 loops, respectively. The accuracy for antibody VH domains is even higher: 2.38Å RMSD for CDR3 and 0.89Å, 0.96Å for the CDR1, CDR2 loops, respectively. NanoNet run times allow generation of ∼1M nanobody structures in less than 4 hours on a standard CPU computer enabling high-throughput structure modeling. NanoNet is available at GitHub: https://github.com/dina-lab3D/NanoNet

DeepUMQA: ultrafast shape recognition-based protein model quality assessment using deep learning

MOTIVATION

Protein model quality assessment is a key component of protein structure prediction. In recent research, the voxelization feature was used to characterize the local structural information of residues, but it may be insufficient for describing residue-level topological information. Design features that can further reflect residue-level topology when combined with deep learning methods are therefore crucial to improve the performance of model quality assessment.

RESULTS

We developed a deep-learning method, DeepUMQA, based on Ultrafast Shape Recognition (USR) for the residue-level single-model quality assessment. In the framework of the deep residual neural network, the residue-level USR feature was introduced to describe the topological relationship between the residue and overall structure by calculating the first moment of a set of residue distance sets and then combined with 1D, 2D and voxelization features to assess the quality of the model. Experimental results on the CASP13, CASP14 test datasets and CAMEO blind test show that USR could supplement the voxelization features to comprehensively characterize residue structure information and significantly improve model assessment accuracy. The performance of DeepUMQA ranks among the top during the state-of-the-art single-model quality assessment methods, including ProQ2, ProQ3, ProQ3D, Ornate, VoroMQA, ProteinGCN, ResNetQA, QDeep, GraphQA, ModFOLD6, ModFOLD7, ModFOLD8, QMEAN3, QMEANDisCo3 and DeepAccNet.

AVAILABILITY AND IMPLEMENTATION

The DeepUMQA server is freely available at http://zhanglab-bioinf.com/DeepUMQA/.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Deep graph learning of inter-protein contacts

MOTIVATION

Inter-protein (interfacial) contact prediction is very useful for in silico structural characterization of protein-protein interactions. Although deep learning has been applied to this problem, its accuracy is not as good as intra-protein contact prediction.

RESULTS

We propose a new deep learning method GLINTER (Graph Learning of INTER-protein contacts) for interfacial contact prediction of dimers, leveraging a rotational invariant representation of protein tertiary structures and a pretrained language model of multiple sequence alignments. Tested on the 13th and 14th CASP-CAPRI datasets, the average top L/10 precision achieved by GLINTER is 54% on the homodimers and 52% on all the dimers, much higher than 30% obtained by the latest deep learning method DeepHomo on the homodimers and 15% obtained by BIPSPI on all the dimers. Our experiments show that GLINTER-predicted contacts help improve selection of docking decoys.

AVAILABILITY AND IMPLEMENTATION

The software is available at https://github.com/zw2x/glinter. The datasets are available at https://github.com/zw2x/glinter/data.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

NPI-RGCNAE: Fast predicting ncRNA-protein interactions using the Relational Graph Convolutional Network Auto-Encoder

- ncRNAs play important roles in a variety of biological processes by interacting with RNA-binding proteins. Therefore, identifying ncRNA-protein interactions is important to understanding the biological functions of ncRNAs. Since experimental methods to determine ncRNA-protein interactions are always costly and time-consuming, computational methods have been proposed as alternative approaches. We developed a novel method NPI-RGCNAE (predicting ncRNA-Protein Interactions by the Relational Graph Convolutional Network Auto-Encoder). With a reliable negative sample selection strategy, we applied the Relational Graph Convolutional Network encoder and the DistMult decoder to predict ncRNA-protein interactions in an accurate and efficient way. By using the 5-fold cross-validation, we found that our method achieved a comparable performance to all state-of-the-art methods. Our method requires less than 10% training time of all state-of-the-art methods. It is a more efficient choice with large datasets in practice. All datasets and source codes of NPI-RGCNAE have been deposited in a public Github repository (https://github.com/Angelia0hh/NPI-RGCNAE).

Protein sequence-to-structure learning: Is this the end(-to-end revolution)?

The potential of deep learning has been recognized in the protein structure prediction community for some time, and became indisputable after CASP13. In CASP14, deep learning has boosted the field to unanticipated levels reaching near-experimental accuracy. This success comes from advances transferred from other machine learning areas, as well as methods specifically designed to deal with protein sequences and structures, and their abstractions. Novel emerging approaches include (i) geometric learning, that is, learning on representations such as graphs, three-dimensional (3D) Voronoi tessellations, and point clouds; (ii) pretrained protein language models leveraging attention; (iii) equivariant architectures preserving the symmetry of 3D space; (iv) use of large meta-genome databases; (v) combinations of protein representations; and (vi) finally truly end-to-end architectures, that is, differentiable models starting from a sequence and returning a 3D structure. Here, we provide an overview and our opinion of the novel deep learning approaches developed in the last 2 years and widely used in CASP14.

Fast and effective protein model refinement using deep graph neural networks

Protein model refinement is the last step applied to improve the quality of a predicted protein model. Currently the most successful refinement methods rely on extensive conformational sampling and thus, take hours or days to refine even a single protein model. Here we propose a fast and effective model refinement method that applies GNN (graph neural networks) to predict refined inter-atom distance probability distribution from an initial model and then rebuilds 3D models from the predicted distance distribution. Tested on the CASP (Critical Assessment of Structure Prediction) refinement targets, our method has comparable accuracy as two leading human groups Feig and Baker, but runs substantially faster. Our method may refine one protein model within ~11 minutes on 1 CPU while Baker needs ~30 hours on 60 CPUs and Feig needs ~16 hours on 1 GPU. Finally, our study shows that GNN outperforms ResNet (convolutional residual neural networks) for model refinement when very limited conformational sampling is allowed.