DNCON2: improved protein contact prediction using two-level deep convolutional neural networks

Emerging frontiers in protein structure prediction following the AlphaFold revolution

Models of protein structures enable molecular understanding of biological processes. Current protein structure prediction tools lie at the interface of biology, chemistry and computer science. Millions of protein structure models have been generated in a very short space of time through a revolution in protein structure prediction driven by deep learning, led by AlphaFold. This has provided a wealth of new structural information. Interpreting these predictions is critical to determining where and when this information is useful. But proteins are not static nor do they act alone, and structures of proteins interacting with other proteins and other biomolecules are critical to a complete understanding of their biological function at the molecular level. This review focuses on the application of state-of-the-art protein structure prediction to these advanced applications. We also suggest a set of guidelines for reporting AlphaFold predictions.

Machine Learning Techniques to Infer Protein Structure and Function from Sequences: A Comprehensive Review

The elucidation of protein structure and function plays a pivotal role in understanding biological processes and facilitating drug discovery. With the exponential growth of protein sequence data, machine learning techniques have emerged as powerful tools for predicting protein characteristics from sequences alone. This review provides a comprehensive overview of the importance and application of machine learning in inferring protein structure and function. We discuss various machine learning approaches, primarily focusing on convolutional neural networks and natural language processing, and their utilization in predicting protein secondary and tertiary structures, residue-residue contacts, protein function, and subcellular localization. Furthermore, we highlight the challenges associated with using machine learning techniques in this context, such as the availability of high-quality training datasets and the interpretability of models. We also delve into the latest progress in the field concerning the advancements made in the development of intricate deep learning architectures. Overall, this review underscores the significance of machine learning in advancing our understanding of protein structure and function, and its potential to revolutionize drug discovery and personalized medicine.

The Historical Evolution and Significance of Multiple Sequence Alignment in Molecular Structure and Function Prediction

Multiple sequence alignment (MSA) has evolved into a fundamental tool in the biological sciences, playing a pivotal role in predicting molecular structures and functions. With broad applications in protein and nucleic acid modeling, MSAs continue to underpin advancements across a range of disciplines. MSAs are not only foundational for traditional sequence comparison techniques but also increasingly important in the context of artificial intelligence (AI)-driven advancements. Recent breakthroughs in AI, particularly in protein and nucleic acid structure prediction, rely heavily on the accuracy and efficiency of MSAs to enhance remote homology detection and guide spatial restraints. This review traces the historical evolution of MSA, highlighting its significance in molecular structure and function prediction. We cover the methodologies used for protein monomers, protein complexes, and RNA, while also exploring emerging AI-based alternatives, such as protein language models, as complementary or replacement approaches to traditional MSAs in application tasks. By discussing the strengths, limitations, and applications of these methods, this review aims to provide researchers with valuable insights into MSA's evolving role, equipping them to make informed decisions in structural prediction research.

Generative β-hairpin design using a residue-based physicochemical property landscape

De novo peptide design is a new frontier that has broad application potential in the biological and biomedical fields. Most existing models for de novo peptide design are largely based on sequence homology that can be restricted based on evolutionarily derived protein sequences and lack the physicochemical context essential in protein folding. Generative machine learning for de novo peptide design is a promising way to synthesize theoretical data that are based on, but unique from, the observable universe. In this study, we created and tested a custom peptide generative adversarial network intended to design peptide sequences that can fold into the β-hairpin secondary structure. This deep neural network model is designed to establish a preliminary foundation of the generative approach based on physicochemical and conformational properties of 20 canonical amino acids, for example, hydrophobicity and residue volume, using extant structure-specific sequence data from the PDB. The beta generative adversarial network model robustly distinguishes secondary structures of β hairpin from α helix and intrinsically disordered peptides with an accuracy of up to 96% and generates artificial β-hairpin peptide sequences with minimum sequence identities around 31% and 50% when compared against the current NCBI PDB and nonredundant databases, respectively. These results highlight the potential of generative models specifically anchored by physicochemical and conformational property features of amino acids to expand the sequence-to-structure landscape of proteins beyond evolutionary limits.

Using Attention-UNet Models to Predict Protein Contact Maps

Proteins are essential to life, and understanding their intrinsic roles requires determining their structure. The field of proteomics has opened up new opportunities by applying deep learning algorithms to large databases of solved protein structures. With the availability of large data sets and advanced machine learning methods, the prediction of protein residue interactions has greatly improved. Protein contact maps provide empirical evidence of the interacting residue pairs within a protein sequence. Template-free protein structure prediction systems rely heavily on this information. This article proposes UNet-CON, an attention-integrated UNet architecture, trained to predict residue-residue contacts in protein sequences. With the predicted contacts being more accurate than state-of-the-art methods on the PDB25 test set, the model paves the way for the development of more powerful deep learning algorithms for predicting protein residue interactions.

AffinityVAE: A multi-objective model for protein-ligand affinity prediction and drug design

In the prediction of protein-ligand affinity, the traditional methods require a large amount of computing resources, and have certain limitations in predicting and simulating the structural changes. Although employing data-driven approaches can yield favorable outcomes in deep learning, it entails a lack of interpretability. Some methods may require additional structural information or domain knowledge to support the interpretation, which may limit their applicability. This paper proposes an affinity variational autoencoder (AffinityVAE) using interaction feature mapping and a variational autoencoder, which consists of a multi-objective model capable of end-to-end affinity prediction and drug discovery. In this study, the limitations of affinity prediction in terms of interpretability are tackled by proposing the concept of a protein-ligand interaction feature map. This increases the diversity and quantity of protein-ligand binding data by designing an adaptive autoencoder of target chemical properties to generate new ligands similar to known ligands and adding them to the original training set. AffinityVAE is then retrained using this extended training set to further validate the protein-ligand binding affinity prediction. Comparisons were conducted between the AffinityVAE and recent methods to demonstrate the high efficiency of the proposed model. The experimental results show that AffinityVAE has very high prediction performance, and it has the potential to enhance the diversity and the amount of protein-ligand binding data, which promotes the drug development.

Deep Learning for Genomics: From Early Neural Nets to Modern Large Language Models

The data explosion driven by advancements in genomic research, such as high-throughput sequencing techniques, is constantly challenging conventional methods used in genomics. In parallel with the urgent demand for robust algorithms, deep learning has succeeded in various fields such as vision, speech, and text processing. Yet genomics entails unique challenges to deep learning, since we expect a superhuman intelligence that explores beyond our knowledge to interpret the genome from deep learning. A powerful deep learning model should rely on the insightful utilization of task-specific knowledge. In this paper, we briefly discuss the strengths of different deep learning models from a genomic perspective so as to fit each particular task with proper deep learning-based architecture, and we remark on practical considerations of developing deep learning architectures for genomics. We also provide a concise review of deep learning applications in various aspects of genomic research and point out current challenges and potential research directions for future genomics applications. We believe the collaborative use of ever-growing diverse data and the fast iteration of deep learning models will continue to contribute to the future of genomics.

ResCNNT-fold: Combining residual convolutional neural network and Transformer for protein fold recognition from language model embeddings

A comprehensive understanding of protein functions holds significant promise for disease research and drug development, and proteins with analogous tertiary structures tend to exhibit similar functions. Protein fold recognition stands as a classical approach in the realm of protein structure investigation. Despite significant advancements made by researchers in this field, the continuous updating of protein databases presents an ongoing challenge in accurately identifying protein fold types. In this study, we introduce a predictor, ResCNNT-fold, for protein fold recognition and employ the LE dataset for testing purpose. ResCNNT-fold leverages a pre-trained language model to obtain embedding representations for protein sequences, which are then processed by the ResCNNT feature extractor, a combination of residual convolutional neural network and Transformer, to derive fold-specific features. Subsequently, the query protein is paired with each protein whose structure is known in the template dataset. For each pair, the similarity score of their fold-specific features is calculated. Ultimately, the query protein is identified as the fold type of the template protein in the pair with the highest similarity score. To further validate the utility and efficacy of the proposed ResCNNT-fold predictor, we conduct a 2-fold cross-validation experiment on the fold level of the LE dataset. Remarkably, this rigorous evaluation yields an exceptional accuracy of 91.57%, which surpasses the best result among other state-of-the-art protein fold recognition methods by an approximate margin of 10%. The excellent performance unequivocally underscores the compelling advantages inherent to our proposed ResCNNT-fold predictor in the realm of protein fold recognition. The source code and data of ResCNNT-fold can be downloaded from https://github.com/Bioinformatics-Laboratory/ResCNNT-fold.

When will RNA get its AlphaFold moment?

The protein structure prediction problem has been solved for many types of proteins by AlphaFold. Recently, there has been considerable excitement to build off the success of AlphaFold and predict the 3D structures of RNAs. RNA prediction methods use a variety of techniques, from physics-based to machine learning approaches. We believe that there are challenges preventing the successful development of deep learning-based methods like AlphaFold for RNA in the short term. Broadly speaking, the challenges are the limited number of structures and alignments making data-hungry deep learning methods unlikely to succeed. Additionally, there are several issues with the existing structure and sequence data, as they are often of insufficient quality, highly biased and missing key information. Here, we discuss these challenges in detail and suggest some steps to remedy the situation. We believe that it is possible to create an accurate RNA structure prediction method, but it will require solving several data quality and volume issues, usage of data beyond simple sequence alignments, or the development of new less data-hungry machine learning methods.

ETLD: an encoder-transformation layer-decoder architecture for protein contact and mutation effects prediction

The latent features extracted from the multiple sequence alignments (MSAs) of homologous protein families are useful for identifying residue-residue contacts, predicting mutation effects, shaping protein evolution, etc. Over the past three decades, a growing body of supervised and unsupervised machine learning methods have been applied to this field, yielding fruitful results. Here, we propose a novel self-supervised model, called encoder-transformation layer-decoder (ETLD) architecture, capable of capturing protein sequence latent features directly from MSAs. Compared to the typical autoencoder model, ETLD introduces a transformation layer with the ability to learn inter-site couplings, which can be used to parse out the two-dimensional residue-residue contacts map after a simple mathematical derivation or an additional supervised neural network. ETLD retains the process of encoding and decoding sequences, and the predicted probabilities of amino acids at each site can be further used to construct the mutation landscapes for mutation effects prediction, outperforming advanced models such as GEMME, DeepSequence and EVmutation in general. Overall, ETLD is a highly interpretable unsupervised model with great potential for improvement and can be further combined with supervised methods for more extensive and accurate predictions.

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.

PROFASA-a web-based protein fragment and structure analysis workstation

Introduction: In the field of bioinformatics and computational biology, protein structure modelling and analysis is a crucial aspect. However, most existing tools require a high degree of technical expertise and lack a user-friendly interface. To address this problem, we developed a protein workstation called PROFASA. Methods: PROFASA is an innovative protein workstation that combines state-of-the-art protein structure visualisation techniques with cutting-edge tools and algorithms for protein analysis. Our goal is to provide users with a comprehensive platform for all protein sequence and structure analyses. PROFASA is designed with the idea of simplifying complex protein analysis workflows into one-click operations, while providing powerful customisation options to meet the needs of professional users. Results: PROFASA provides a one-stop solution that enables users to perform protein structure evaluation, parametric analysis and protein visualisation. Users can use I-TASSER or AlphaFold2 to construct protein models with one click, generate new protein sequences, models, and calculate protein parameters. In addition, PROFASA offers features such as real-time collaboration, note sharing, and shared projects, making it an ideal tool for researchers and teaching professionals. Discussion: PROFASA's innovation lies in its user-friendly interface and one-stop solution. It not only lowers the barrier to entry for protein computation, analysis and visualisation tools, but also opens up new possibilities for protein research and education. We expect PROFASA to advance the study of protein design and engineering and open up new research areas.

Immunogenicity and efficacy of a novel multi-patch SARS-CoV-2/COVID-19 vaccine candidate

Introduction

While there has been considerable progress in the development of vaccines against SARS-CoV-2, largely based on the S (spike) protein of the virus, less progress has been made with vaccines delivering different viral antigens with cross-reactive potential.

Methods

In an effort to develop an immunogen with the capacity to induce broad antigen presentation, we have designed a multi-patch synthetic candidate containing dominant and persistent B cell epitopes from conserved regions of SARS-CoV-2 structural proteins associated with long-term immunity, termed CoV2-BMEP. Here we describe the characterization, immunogenicity and efficacy of CoV2-BMEP using two delivery platforms: nucleic acid DNA and attenuated modified vaccinia virus Ankara (MVA).

Results

In cultured cells, both vectors produced a main protein of about 37 kDa as well as heterogeneous proteins with size ranging between 25-37 kDa. In C57BL/6 mice, both homologous and heterologous prime/boost combination of vectors induced the activation of SARS-CoV-2-specific CD4 and CD8 T cell responses, with a more balanced CD8⁺ T cell response detected in lungs. The homologous MVA/MVA immunization regimen elicited the highest specific CD8⁺ T cell responses in spleen and detectable binding antibodies (bAbs) to S and N antigens of SARS-CoV-2. In SARS-CoV-2 susceptible k18-hACE2 Tg mice, two doses of MVA-CoV2-BMEP elicited S- and N-specific bAbs as well as cross-neutralizing antibodies against different variants of concern (VoC). After SARS-CoV-2 challenge, all animals in the control unvaccinated group succumbed to the infection while vaccinated animals with high titers of neutralizing antibodies were fully protected against mortality, correlating with a reduction of virus infection in the lungs and inhibition of the cytokine storm.

Discussion

These findings revealed a novel immunogen with the capacity to control SARS-CoV-2 infection, using a broader antigen presentation mechanism than the approved vaccines based solely on the S antigen.

Learning to evolve structural ensembles of unfolded and disordered proteins using experimental solution data

The structural characterization of proteins with a disorder requires a computational approach backed by experiments to model their diverse and dynamic structural ensembles. The selection of conformational ensembles consistent with solution experiments of disordered proteins highly depends on the initial pool of conformers, with currently available tools limited by conformational sampling. We have developed a Generative Recurrent Neural Network (GRNN) that uses supervised learning to bias the probability distributions of torsions to take advantage of experimental data types such as nuclear magnetic resonance J-couplings, nuclear Overhauser effects, and paramagnetic resonance enhancements. We show that updating the generative model parameters according to the reward feedback on the basis of the agreement between experimental data and probabilistic selection of torsions from learned distributions provides an alternative to existing approaches that simply reweight conformers of a static structural pool for disordered proteins. Instead, the biased GRNN, DynamICE, learns to physically change the conformations of the underlying pool of the disordered protein to those that better agree with experiments.

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/.

Contact-Assisted Threading in Low-Homology Protein Modeling

The ability to successfully predict the three-dimensional structure of a protein from its amino acid sequence has made considerable progress in the recent past. The progress is propelled by the improved accuracy of deep learning-based inter-residue contact map predictors coupled with the rising growth of protein sequence databases. Contact map encodes interatomic interaction information that can be exploited for highly accurate prediction of protein structures via contact map threading even for the query proteins that are not amenable to direct homology modeling. As such, contact-assisted threading has garnered considerable research effort. In this chapter, we provide an overview of existing contact-assisted threading methods while highlighting the recent advances and discussing some of the current limitations and future prospects in the application of contact-assisted threading for improving the accuracy of low-homology protein modeling.

rrQNet: Protein contact map quality estimation by deep evolutionary reconciliation

Protein contact maps have proven to be a valuable tool in the deep learning revolution of protein structure prediction, ushering in the recent breakthrough by AlphaFold2. However, self-assessment of the quality of predicted structures are typically performed at the granularity of three-dimensional coordinates as opposed to directly exploiting the rotation- and translation-invariant two-dimensional (2D) contact maps. Here, we present rrQNet, a deep learning method for self-assessment in 2D by contact map quality estimation. Our approach is based on the intuition that for a contact map to be of high quality, the residue pairs predicted to be in contact should be mutually consistent with the evolutionary context of the protein. The deep neural network architecture of rrQNet implements this intuition by cascading two deep modules-one encoding the evolutionary context and the other performing evolutionary reconciliation. The penultimate stage of rrQNet estimates the quality scores at the interacting residue-pair level, which are then aggregated for estimating the quality of a contact map. This design choice offers versatility at varied resolutions from individual residue pairs to full-fledged contact maps. Trained on multiple complementary sources of contact predictors, rrQNet facilitates generalizability across various contact maps. By rigorously testing using publicly available datasets and comparing against several in-house baseline approaches, we show that rrQNet accurately reproduces the true quality score of a predicted contact map and successfully distinguishes between accurate and inaccurate contact maps predicted by a wide variety of contact predictors. The open-source rrQNet software package is freely available at https://github.com/Bhattacharya-Lab/rrQNet.

Prediction of inter-chain distance maps of protein complexes with 2D attention-based deep neural networks

Residue-residue distance information is useful for predicting tertiary structures of protein monomers or quaternary structures of protein complexes. Many deep learning methods have been developed to predict intra-chain residue-residue distances of monomers accurately, but few methods can accurately predict inter-chain residue-residue distances of complexes. We develop a deep learning method CDPred (i.e., Complex Distance Prediction) based on the 2D attention-powered residual network to address the gap. Tested on two homodimer datasets, CDPred achieves the precision of 60.94% and 42.93% for top L/5 inter-chain contact predictions (L: length of the monomer in homodimer), respectively, substantially higher than DeepHomo's 37.40% and 23.08% and GLINTER's 48.09% and 36.74%. Tested on the two heterodimer datasets, the top Ls/5 inter-chain contact prediction precision (Ls: length of the shorter monomer in heterodimer) of CDPred is 47.59% and 22.87% respectively, surpassing GLINTER's 23.24% and 13.49%. Moreover, the prediction of CDPred is complementary with that of AlphaFold2-multimer.

Collectively encoding protein properties enriches protein language models

Pre-trained natural language processing models on a large natural language corpus can naturally transfer learned knowledge to protein domains by fine-tuning specific in-domain tasks. However, few studies focused on enriching such protein language models by jointly learning protein properties from strongly-correlated protein tasks. Here we elaborately designed a multi-task learning (MTL) architecture, aiming to decipher implicit structural and evolutionary information from three sequence-level classification tasks for protein family, superfamily and fold. Considering the co-existing contextual relevance between human words and protein language, we employed BERT, pre-trained on a large natural language corpus, as our backbone to handle protein sequences. More importantly, the encoded knowledge obtained in the MTL stage can be well transferred to more fine-grained downstream tasks of TAPE. Experiments on structure- or evolution-related applications demonstrate that our approach outperforms many state-of-the-art Transformer-based protein models, especially in remote homology detection.

Res-Dom: predicting protein domain boundary from sequence using deep residual network and Bi-LSTM

Motivation

Protein domains are the basic units of proteins that can fold, function and evolve independently. Protein domain boundary partition plays an important role in protein structure prediction, understanding their biological functions, annotating their evolutionary mechanisms and protein design. Although there are many methods that have been developed to predict domain boundaries from protein sequence over the past two decades, there is still much room for improvement.

Results

In this article, a novel domain boundary prediction tool called Res-Dom was developed, which is based on a deep residual network, bidirectional long short-term memory (Bi-LSTM) and transfer learning. We used deep residual neural networks to extract higher-order residue-related information. In addition, we also used a pre-trained protein language model called ESM to extract sequence embedded features, which can summarize sequence context information more abundantly. To improve the global representation of these deep residual networks, a Bi-LSTM network was also designed to consider long-range interactions between residues. Res-Dom was then tested on an independent test set including 342 proteins and generated correct single-domain and multi-domain classifications with a Matthew's correlation coefficient of 0.668, which was 17.6% higher than the second-best compared method. For domain boundaries, the normalized domain overlapping score of Res-Dom was 0.849, which was 5% higher than the second-best compared method. Furthermore, Res-Dom required significantly less time than most of the recently developed state-of-the-art domain prediction methods.

Availability and implementation

All source code, datasets and model are available at http://isyslab.info/Res-Dom/.

Improved Protein Real-Valued Distance Prediction Using Deep Residual Dense Network (DRDN)

Three-dimensional protein structure prediction is one of the major challenges in bioinformatics. According to recent research findings, real-valued distance prediction plays a vital role in determining the unique three-dimensional protein structure. This paper proposes a novel methodology involving a deep residual dense network (DRDN) for predicting protein real-valued distance. The features extracted from the given query protein sequence and its corresponding homologous sequences are used for training the model. Multi-aligned homologous sequences for each query protein sequence are retrieved from five different databases using DeepMSA, HHblits, and HITS_PR_HHblits methods. The proposed method yielded outcomes of 3.89, 0.23, 0.45, and 0.63, respectively, corresponding to the evaluation metrics such as Absolute Error, Relative Error, High-accuracy Pairwise Distance Test (PDA), and Pairwise Distance Test (PDT). Further, the contact map is computed based on CASP criteria by converting the predicted real-valued distance, and it is evaluated using the precision metric. It is observed that precision of long-range top L/5 contact prediction on the CASP13 dataset by the proposed method, RaptorX, Zhang, trRosetta, JinboXu & JinLu, and Deepdist are 0.834, 0.657, 0.70, 0.785, 0.786, and 0.812, respectively. Also, Top-L/5 contact prediction on the CASP14 dataset evaluated using average precision resulted in 0.847, 0.707, 0.752, 0.783, 0.792, 0.817, and 0.825 respectively, corresponding to the proposed method, Zhang, RaptorX, trRosetta, Deepdist, JinboXu & JinLu, and Alphafold2.

Protein Science Meets Artificial Intelligence: A Systematic Review and a Biochemical Meta-Analysis of an Inter-Field

Proteins are some of the most fascinating and challenging molecules in the universe, and they pose a big challenge for artificial intelligence. The implementation of machine learning/AI in protein science gives rise to a world of knowledge adventures in the workhorse of the cell and proteome homeostasis, which are essential for making life possible. This opens up epistemic horizons thanks to a coupling of human tacit-explicit knowledge with machine learning power, the benefits of which are already tangible, such as important advances in protein structure prediction. Moreover, the driving force behind the protein processes of self-organization, adjustment, and fitness requires a space corresponding to gigabytes of life data in its order of magnitude. There are many tasks such as novel protein design, protein folding pathways, and synthetic metabolic routes, as well as protein-aggregation mechanisms, pathogenesis of protein misfolding and disease, and proteostasis networks that are currently unexplored or unrevealed. In this systematic review and biochemical meta-analysis, we aim to contribute to bridging the gap between what we call binomial artificial intelligence (AI) and protein science (PS), a growing research enterprise with exciting and promising biotechnological and biomedical applications. We undertake our task by exploring "the state of the art" in AI and machine learning (ML) applications to protein science in the scientific literature to address some critical research questions in this domain, including What kind of tasks are already explored by ML approaches to protein sciences? What are the most common ML algorithms and databases used? What is the situational diagnostic of the AI-PS inter-field? What do ML processing steps have in common? We also formulate novel questions such as Is it possible to discover what the rules of protein evolution are with the binomial AI-PS? How do protein folding pathways evolve? What are the rules that dictate the folds? What are the minimal nuclear protein structures? How do protein aggregates form and why do they exhibit different toxicities? What are the structural properties of amyloid proteins? How can we design an effective proteostasis network to deal with misfolded proteins? We are a cross-functional group of scientists from several academic disciplines, and we have conducted the systematic review using a variant of the PICO and PRISMA approaches. The search was carried out in four databases (PubMed, Bireme, OVID, and EBSCO Web of Science), resulting in 144 research articles. After three rounds of quality screening, 93 articles were finally selected for further analysis. A summary of our findings is as follows: regarding AI applications, there are mainly four types: 1) genomics, 2) protein structure and function, 3) protein design and evolution, and 4) drug design. In terms of the ML algorithms and databases used, supervised learning was the most common approach (85%). As for the databases used for the ML models, PDB and UniprotKB/Swissprot were the most common ones (21 and 8%, respectively). Moreover, we identified that approximately 63% of the articles organized their results into three steps, which we labeled pre-process, process, and post-process. A few studies combined data from several databases or created their own databases after the pre-process. Our main finding is that, as of today, there are no research road maps serving as guides to address gaps in our knowledge of the AI-PS binomial. All research efforts to collect, integrate multidimensional data features, and then analyze and validate them are, so far, uncoordinated and scattered throughout the scientific literature without a clear epistemic goal or connection between the studies. Therefore, our main contribution to the scientific literature is to offer a road map to help solve problems in drug design, protein structures, design, and function prediction while also presenting the "state of the art" on research in the AI-PS binomial until February 2021. Thus, we pave the way toward future advances in the synthetic redesign of novel proteins and protein networks and artificial metabolic pathways, learning lessons from nature for the welfare of humankind. Many of the novel proteins and metabolic pathways are currently non-existent in nature, nor are they used in the chemical industry or biomedical field.

Using multiple convolutional window scanning of convolutional neural network for an efficient prediction of ATP-binding sites in transport proteins

Protein multiple sequence alignment information has long been important features to know about functions of proteins inferred from related sequences with known functions. It is therefore one of the underlying ideas of Alpha fold 2, a breakthrough study and model for the prediction of three-dimensional structures of proteins from their primary sequence. Our study used protein multiple sequence alignment information in the form of position-specific scoring matrices as input. We also refined the use of a convolutional neural network, a well-known deep-learning architecture with impressive achievement on image and image-like data. Specifically, we revisited the study of prediction of adenosine triphosphate (ATP)-binding sites with more efficient convolutional neural networks. We applied multiple convolutional window scanning filters of a convolutional neural network on position-specific scoring matrices for as much as useful information as possible. Furthermore, only the most specific motifs are retained at each feature map output through the one-max pooling layer before going to the next layer. We assumed that this way could help us retain the most conserved motifs which are discriminative information for prediction. Our experiment results show that a convolutional neural network with not too many convolutional layers can be enough to extract the conserved information of proteins, which leads to higher performance. Our best prediction models were obtained after examining them with different hyper-parameters. Our experiment results showed that our models were superior to traditional use of convolutional neural networks on the same datasets as well as other machine-learning classification algorithms.

Inter-Residue Distance Prediction From Duet Deep Learning Models

Residue distance prediction from the sequence is critical for many biological applications such as protein structure reconstruction, protein–protein interaction prediction, and protein design. However, prediction of fine-grained distances between residues with long sequence separations still remains challenging. In this study, we propose DuetDis, a method based on duet feature sets and deep residual network with squeeze-and-excitation (SE), for protein inter-residue distance prediction. DuetDis embraces the ability to learn and fuse features directly or indirectly extracted from the whole-genome/metagenomic databases and, therefore, minimize the information loss through ensembling models trained on different feature sets. We evaluate DuetDis and 11 widely used peer methods on a large-scale test set (610 proteins chains). The experimental results suggest that 1) prediction results from different feature sets show obvious differences; 2) ensembling different feature sets can improve the prediction performance; 3) high-quality multiple sequence alignment (MSA) used for both training and testing can greatly improve the prediction performance; and 4) DuetDis is more accurate than peer methods for the overall prediction, more reliable in terms of model prediction score, and more robust against shallow multiple sequence alignment (MSA).

Refined Contact Map Prediction of Peptides Based on GCN and ResNet

Predicting peptide inter-residue contact maps plays an important role in computational biology, which determines the topology of the peptide structure. However, due to the limited number of known homologous structures, there is still much room for inter-residue contact map prediction. Current models are not sufficient for capturing the high accuracy relationship between the residues, especially for those with a long-range distance. In this article, we developed a novel deep neural network framework to refine the rough contact map produced by the existing methods. The rough contact map is used to construct the residue graph that is processed by the graph convolutional neural network (GCN). GCN can better capture the global information and is therefore used to grasp the long-range contact relationship. The residual convolutional neural network is also applied in the framework for learning local information. We conducted the experiments on four different test datasets, and the inter-residue long-range contact map prediction accuracy demonstrates the effectiveness of our proposed method.

Performing protein fold recognition by exploiting a stack convolutional neural network with the attention mechanism

Protein fold recognition is a critical step in protein structure and function prediction, and aims to ascertain the most likely fold type of the query protein. As a typical pattern recognition problem, designing a powerful feature extractor and metric function to extract relevant and representative fold-specific features from protein sequences is the key to improving protein fold recognition. In this study, we propose an effective sequence-based approach, called RattnetFold, to identify protein fold types. The basic concept of RattnetFold is to employ a stack convolutional neural network with the attention mechanism that acts as a feature extractor to extract fold-specific features from protein residue-residue contact maps. Moreover, based on the fold-specific features, we leverage metric learning to project fold-specific features into a subspace where similar proteins are closer together and name this approach RattnetFoldPro. Benchmarking experiments illustrate that RattnetFold and RattnetFoldPro enable the convolutional neural networks to efficiently learn the underlying subtle patterns in residue-residue contact maps, thereby improving the performance of protein fold recognition. An online web server of RattnetFold and the benchmark datasets are freely available at http://csbio.njust.edu.cn/bioinf/rattnetfold/.

Deep learning methods for 3D structural proteome and interactome modeling

Bolstered by recent methodological and hardware advances, deep learning has increasingly been applied to biological problems and structural proteomics. Such approaches have achieved remarkable improvements over traditional machine learning methods in tasks ranging from protein contact map prediction to protein folding, prediction of protein-protein interaction interfaces, and characterization of protein-drug binding pockets. In particular, emergence of ab initio protein structure prediction methods including AlphaFold2 has revolutionized protein structural modeling. From a protein function perspective, numerous deep learning methods have facilitated deconvolution of the exact amino acid residues and protein surface regions responsible for binding other proteins or small molecule drugs. In this review, we provide a comprehensive overview of recent deep learning methods applied in structural proteomics.

MFPS_CNN: Multi-filter pattern scanning from position-specific scoring matrix with convolutional neural network for efficient prediction of ion transporters

In cellular transportation mechanisms, the movement of ions across the cell membrane and its proper control are important for cells, especially for life processes. Ion transporters/pumps and ion channel proteins work as border guards controlling the incessant traffic of ions across cell membranes. We revisited the study of classification of transporters and ion channels from membrane proteins with a more efficient deep learning approach. Specifically, we applied multi-window scanning filters of convolutional neural networks on almost full-length position-specific scoring matrices for extracting useful information. In this way, we were able to retain important evolutionary information of the proteins. Our experiment results show that a convolutional neural network with a minimum number of convolutional layers can be enough to extract the conserved information of proteins which leads to higher performance. Our best prediction models were obtained after examining different data imbalanced handling techniques, and different protein encoding methods. We also showed that our models were superior to traditional deep learning approaches on the same datasets as well as other machine learning classification algorithms.

Membrane contact probability: An essential and predictive character for the structural and functional studies of membrane proteins

One of the unique traits of membrane proteins is that a significant fraction of their hydrophobic amino acids is exposed to the hydrophobic core of lipid bilayers rather than being embedded in the protein interior, which is often not explicitly considered in the protein structure and function predictions. Here, we propose a characteristic and predictive quantity, the membrane contact probability (MCP), to describe the likelihood of the amino acids of a given sequence being in direct contact with the acyl chains of lipid molecules. We show that MCP is complementary to solvent accessibility in characterizing the outer surface of membrane proteins, and it can be predicted for any given sequence with a machine learning-based method by utilizing a training dataset extracted from MemProtMD, a database generated from molecular dynamics simulations for the membrane proteins with a known structure. As the first of many potential applications, we demonstrate that MCP can be used to systematically improve the prediction precision of the protein contact maps and structures.

The distribution of residues on protein surfaces is largely determined by the surrounding environment. For soluble proteins, most of the residues on the outer surface are hydrophilic, and people use the quantity “solvent accessibility” to describe and predict these surface residues. In contrast, for membrane proteins that are embedded in a lipid bilayer, many of their surface residues are hydrophobic and membrane-contacting, but there is yet a widely-accepted quantity for the description or prediction of this characteristic property. Here, we propose a new quantity termed “membrane contact probability (MCP)”, which can be used to describe and predict the membrane-contacting surface residues of proteins. We also propose a machine learning-based method to predict MCP from protein sequences, utilizing the dataset generated by physics-based computer simulations. We demonstrate that a quantity such as MCP is helpful for protein structure prediction, and we believe that it will find broad applications in the structure and function studies of membrane proteins.

Computational modeling of mRNA degradation dynamics using deep neural networks

MOTIVATION

messenger RNA (mRNA) degradation plays critical roles in post-transcriptional gene regulation. A major component of mRNA degradation is determined by 3'-UTR elements. Hence, researchers are interested in studying mRNA dynamics as a function of 3'-UTR elements. A recent study measured the mRNA degradation dynamics of tens of thousands of 3'-UTR sequences using a massively parallel reporter assay. However, the computational approach used to model mRNA degradation was based on a simplifying assumption of a linear degradation rate. Consequently, the underlying mechanism of 3'-UTR elements is still not fully understood.

RESULTS

Here, we developed deep neural networks to predict mRNA degradation dynamics and interpreted the networks to identify regulatory elements in the 3'-UTR and their positional effect. Given an input of a 110 nt-long 3'-UTR sequence and an initial mRNA level, the model predicts mRNA levels of eight consecutive time points. Our deep neural networks significantly improved prediction performance of mRNA degradation dynamics compared with extant methods for the task. Moreover, we demonstrated that models predicting the dynamics of two identical 3'-UTR sequences, differing by their poly(A) tail, performed better than single-task models. On the interpretability front, by using Integrated Gradients, our convolutional neural networks (CNNs) models identified known and novel cis-regulatory sequence elements of mRNA degradation. By applying a novel systematic evaluation of model interpretability, we demonstrated that the recurrent neural network models are inferior to the CNN models in terms of interpretability and that random initialization ensemble improves both prediction and interoperability performance. Moreover, using a mutagenesis analysis, we newly discovered the positional effect of various 3'-UTR elements.

AVAILABILITY AND IMPLEMENTATION

All the code developed through this study is available at github.com/OrensteinLab/DeepUTR/.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

OUP accepted manuscript

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.

Co-evolutionary distance predictions contain flexibility information

MOTIVATION

Co-evolution analysis can be used to accurately predict residue-residue contacts from multiple sequence alignments. The introduction of machine-learning techniques has enabled substantial improvements in precision and a shift from predicting binary contacts to predict distances between pairs of residues. These developments have significantly improved the accuracy of de novo prediction of static protein structures. With AlphaFold2 lifting the accuracy of some predicted protein models close to experimental levels, structure prediction research will move on to other challenges. One of those areas is the prediction of more than one conformation of a protein. Here, we examine the potential of residue-residue distance predictions to be informative of protein flexibility rather than simply static structure.

RESULTS

We used DMPfold to predict distance distributions for every residue pair in a set of proteins that showed both rigid and flexible behaviour. Residue pairs that were in contact in at least one reference structure were classified as rigid, flexible or neither. The predicted distance distribution of each residue pair was analysed for local maxima of probability indicating the most likely distance or distances between a pair of residues. We found that rigid residue pairs tended to have only a single local maximum in their predicted distance distributions while flexible residue pairs more often had multiple local maxima. These results suggest that the shape of predicted distance distributions contains information on the rigidity or flexibility of a protein and its constituent residues.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Protein inter-residue contact and distance prediction by coupling complementary coevolution features with deep residual networks in CASP14

This article reports and analyzes the results of protein contact and distance prediction by our methods in the 14th Critical Assessment of techniques for protein Structure Prediction (CASP14). A new deep learning-based contact/distance predictor was employed based on the ensemble of two complementary coevolution features coupling with deep residual networks. We also improved our multiple sequence alignment (MSA) generation protocol with wholesale meta-genome sequence databases. On 22 CASP14 free modeling (FM) targets, the proposed model achieved a top-L/5 long-range precision of 63.8% and a mean distance bin error of 1.494. Based on the predicted distance potentials, 11 out of 22 FM targets and all of the 14 FM/template-based modeling (TBM) targets have correctly predicted folds (TM-score >0.5), suggesting that our approach can provide reliable distance potentials for ab initio protein folding.

Multi contact-based folding method for de novo protein structure prediction

Meta contact, which combines different contact maps into one to improve contact prediction accuracy and effectively reduce the noise from a single contact map, is a widely used method. However, protein structure prediction using meta contact cannot fully exploit the information carried by original contact maps. In this work, a multi contact-based folding method under the evolutionary algorithm framework, MultiCFold, is proposed. In MultiCFold, the thorough information of different contact maps is directly used by populations to guide protein structure folding. In addition, noncontact is considered as an effective supplement to contact information and can further assist protein folding. MultiCFold is tested on a set of 120 nonredundant proteins, and the average TM-score and average RMSD reach 0.617 and 5.815 Å, respectively. Compared with the meta contact-based method, MetaCFold, average TM-score and average RMSD have a 6.62 and 8.82% improvement. In particular, the import of noncontact information increases the average TM-score by 6.30%. Furthermore, MultiCFold is compared with four state-of-the-art methods of CASP13 on the 24 FM targets, and results show that MultiCFold is significantly better than other methods after the full-atom relax procedure.

Combining Cryo-EM Density Map and Residue Contact for Protein Secondary Structure Topologies

Although atomic structures have been determined directly from cryo-EM density maps with high resolutions, current structure determination methods for medium resolution (5 to 10 Å) cryo-EM maps are limited by the availability of structure templates. Secondary structure traces are lines detected from a cryo-EM density map for α-helices and β-strands of a protein. A topology of secondary structures defines the mapping between a set of sequence segments and a set of traces of secondary structures in three-dimensional space. In order to enhance accuracy in ranking secondary structure topologies, we explored a method that combines three sources of information: a set of sequence segments in 1D, a set of amino acid contact pairs in 2D, and a set of traces in 3D at the secondary structure level. A test of fourteen cases shows that the accuracy of predicted secondary structures is critical for deriving topologies. The use of significant long-range contact pairs is most effective at enriching the rank of the maximum-match topology for proteins with a large number of secondary structures, if the secondary structure prediction is fairly accurate. It was observed that the enrichment depends on the quality of initial topology candidates in this approach. We provide detailed analysis in various cases to show the potential and challenge when combining three sources of information.

Improved protein contact prediction using dimensional hybrid residual networks and singularity enhanced loss function

Deep residual learning has shown great success in protein contact prediction. In this study, a new deep residual learning-based protein contact prediction model was developed. Comparing with previous models, a new type of residual block hybridizing 1D and 2D convolutions was designed to increase the effective receptive field of the residual network, and a new loss function emphasizing the easily misclassified residue pairs was proposed to enhance the model training. The developed protein contact prediction model referred to as DRN-1D2D was first evaluated on 105 CASP11 targets, 76 CAMEO hard targets and 398 membrane proteins together with two in house-developed reference models based on either the standard 2D residual block or the traditional BCE loss function, from which we confirmed that both the dimensional hybrid residual block and the singularity enhanced loss function can be employed to improve the model performance for protein contact prediction. DRN-1D2D was further evaluated on 39 CASP13 and CASP14 free modeling targets together with the two reference models and six state-of-the-art protein contact prediction models including DeepCov, DeepCon, DeepConPred2, SPOT-Contact, RaptorX-Contact and TripleRes. The result shows that DRN-1D2D consistently achieved the best performance among all these models.

Improved 3-D Protein Structure Predictions using Deep ResNet Model

Protein Structure Prediction (PSP) is considered to be a complicated problem in computational biology. In spite of, the remarkable progress made by the co-evolution-based method in PSP, it is still a challenging and unresolved problem. Recently, along with co-evolutionary relationships, deep learning approaches have been introduced in PSP that lead to significant progress. In this paper a novel methodology using deep ResNet architecture for predicting inter-residue distance and dihedral angles is proposed, that aims to generate 125 homologous sequences in an average from a set of customized sequence database. These sequences are used to generate input features. As an outcome of neural networks, a pool of structures is generated from which the lowest potential structure is chosen as the final predicted 3-D protein structure. The proposed method is trained using 6521 protein sequences extracted from Protein Data Bank (PDB). For testing 48 protein sequences whose residue length is less than 400 residues are chosen from the 13th Critical Assessment of protein Structure Prediction (CASP 13) dataset are used. The model is compared with Alphafold, Zhang, and RaptorX. The template modeling (TM) score is used to evaluate the accuracy of the estimated structure. The proposed method produces better performances for 52% of the target sequences while that of Alphafold, Zhang, RaptorX were 10%, 22.9%, and 6% respectively. Additionally, for 37.5% target sequences, the proposed method was able to achieve accuracy greater than or equal to 0.80. The TM score obtained for the sequences under consideration were 0.69, 0.67, 0.65, and 0.58 respectively for the proposed method, Alphafold, Zhang, and RaptorX.

Accurate prediction of inter-protein residue-residue contacts for homo-oligomeric protein complexes

Protein-protein interactions play a fundamental role in all cellular processes. Therefore, determining the structure of protein-protein complexes is crucial to understand their molecular mechanisms and develop drugs targeting the protein-protein interactions. Recently, deep learning has led to a breakthrough in intra-protein contact prediction, achieving an unusual high accuracy in recent Critical Assessment of protein Structure Prediction (CASP) structure prediction challenges. However, due to the limited number of known homologous protein-protein interactions and the challenge to generate joint multiple sequence alignments of two interacting proteins, the advances in inter-protein contact prediction remain limited. Here, we have proposed a deep learning model to predict inter-protein residue-residue contacts across homo-oligomeric protein interfaces, named as DeepHomo. Unlike previous deep learning approaches, we integrated intra-protein distance map and inter-protein docking pattern, in addition to evolutionary coupling, sequence conservation, and physico-chemical information of monomers. DeepHomo was extensively tested on both experimentally determined structures and realistic CASP-Critical Assessment of Predicted Interaction (CAPRI) targets. It was shown that DeepHomo achieved a high precision of >60% for the top predicted contact and outperformed state-of-the-art direct-coupling analysis and machine learning-based approaches. Integrating predicted inter-chain contacts into protein-protein docking significantly improved the docking accuracy on the benchmark dataset of realistic homo-dimeric targets from CASP-CAPRI experiments. DeepHomo is available at http://huanglab.phys.hust.edu.cn/DeepHomo/.

Modeling protein structures with the coarse-grained UNRES force field in the CASP14 experiment

The UNited RESidue (UNRES) force field was tested in the 14th Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction (CASP14), in which larger oligomeric and multimeric targets were present compared to previous editions. Three prediction modes were tested (i) ab initio (the UNRES group), (ii) contact-assisted (the UNRES-contact group), and (iii) template-assisted (the UNRES-template group). For most of the targets, the contact restraints were derived from the server models top-ranked by the DeepQA method, while the DNCON2 method was used for 11 targets. Our consensus-fragment procedure was used to run template-assisted predictions. Each group also processed the Nuclear Magnetic Resonance (NMR)- and Small Angle X-Ray Scattering (SAXS)-data assisted targets. The average Global Distance Test Total Score (GDT_TS) of the 'Model 1' predictions were 29.17, 39.32, and 56.37 for the UNRES, UNRES-contact, and UNRES-template predictions, respectively, increasing by 0.53, 2.24, and 3.76, respectively, compared to CASP13. It was also found that the GDT_TS of the UNRES models obtained in ab initio mode and in the contact-assisted mode decreases with the square root of chain length, while the exponent in this relationship is 0.20 for the UNRES-template group models and 0.11 for the best performing AlphaFold2 models, which suggests that incorporation of database information, which stems from protein evolution, brings in long-range correlations, thus enabling the correction of force-field inaccuracies.

An interpretable machine learning method for homo-trimeric protein interface residue-residue interaction prediction

Protein-protein interaction plays an important role in life activities. A more fine-grained analysis, such as residues and atoms level, will better benefit us to understand the mechanism for inter-protein interaction and drug design. The development of efficient computational methods to reduce trials and errors, as well as assisting experimental researchers to determine the complex structure are some of the ongoing studies in the field. The research of trimer protein interface, especially homotrimer, has been rarely studied. In this paper, we proposed an interpretable machine learning method for homo-trimeric protein interface residue pairs prediction. The structure, sequence, and physicochemical information are intergraded as feature input fed to model for training. Graph model is utilized to present spatial information for intra-protein. Matrix factorization captures the different features' interactions. Kernel function is designed to auto-acquire the adjacent information of our target residue pairs. The accuracy rate achieves 54.5% in an independent test set. Sequence and structure alignment exhibit the ability of model self-study. Our model indicates the biological significance between sequence and structure, and could be auxiliary for reducing trials and errors in the fields of protein complex determination and protein-protein docking, etc. SIGNIFICANCE: Protein complex structures are significant for understanding protein function and promising functional protein design. With data increasing, some computational tools have been developed for protein complex residue contact prediction, which is one of the most significant steps for complex structure prediction. But for homo-trimeric protein, the sequence-based deep learning predictors are infeasible for homologous sequences, and the algorithm black box prevents us from understanding of each step operation. In this way, we propose an interpreting machine learning method for homo-trimeric protein interface residue-residue interaction prediction, and the predictor shows a good performance. Our work provides a computational auxiliary way for determining the homo-trimeric proteins interface residue pairs which will be further verified by wet experiments, and and gives a hand for the downstream works, such as protein-protein docking, protein complex structure prediction and drug design.

Improving fragment-based ab initio protein structure assembly using low-accuracy contact-map predictions

Sequence-based contact prediction has shown considerable promise in assisting non-homologous structure modeling, but it often requires many homologous sequences and a sufficient number of correct contacts to achieve correct folds. Here, we developed a method, C-QUARK, that integrates multiple deep-learning and coevolution-based contact-maps to guide the replica-exchange Monte Carlo fragment assembly simulations. The method was tested on 247 non-redundant proteins, where C-QUARK could fold 75% of the cases with TM-scores (template-modeling scores) ≥0.5, which was 2.6 times more than that achieved by QUARK. For the 59 cases that had either low contact accuracy or few homologous sequences, C-QUARK correctly folded 6 times more proteins than other contact-based folding methods. C-QUARK was also tested on 64 free-modeling targets from the 13th CASP (critical assessment of protein structure prediction) experiment and had an average GDT_TS (global distance test) score that was 5% higher than the best CASP predictors. These data demonstrate, in a robust manner, the progress in modeling non-homologous protein structures using low-accuracy and sparse contact-map predictions.

Folding non-homologous proteins by coupling deep-learning contact maps with I-TASSER assembly simulations

Structure prediction for proteins lacking homologous templates in the Protein Data Bank (PDB) remains a significant unsolved problem. We developed a protocol, C-I-TASSER, to integrate interresidue contact maps from deep neural-network learning with the cutting-edge I-TASSER fragment assembly simulations. Large-scale benchmark tests showed that C-I-TASSER can fold more than twice the number of non-homologous proteins than the I-TASSER, which does not use contacts. When applied to a folding experiment on 8,266 unsolved Pfam families, C-I-TASSER successfully folded 4,162 domain families, including 504 folds that are not found in the PDB. Furthermore, it created correct folds for 85% of proteins in the SARS-CoV-2 genome, despite the quick mutation rate of the virus and sparse sequence profiles. The results demonstrated the critical importance of coupling whole-genome and metagenome-based evolutionary information with optimal structure assembly simulations for solving the problem of non-homologous protein structure prediction.

Improving protein tertiary structure prediction by deep learning and distance prediction in CASP14

Substantial progresses in protein structure prediction have been made by utilizing deep‐learning and residue‐residue distance prediction since CASP13. Inspired by the advances, we improve our CASP14 MULTICOM protein structure prediction system by incorporating three new components: (a) a new deep learning‐based protein inter‐residue distance predictor to improve template‐free (ab initio) tertiary structure prediction, (b) an enhanced template‐based tertiary structure prediction method, and (c) distance‐based model quality assessment methods empowered by deep learning. In the 2020 CASP14 experiment, MULTICOM predictor was ranked seventh out of 146 predictors in tertiary structure prediction and ranked third out of 136 predictors in inter‐domain structure prediction. The results demonstrate that the template‐free modeling based on deep learning and residue‐residue distance prediction can predict the correct topology for almost all template‐based modeling targets and a majority of hard targets (template‐free targets or targets whose templates cannot be recognized), which is a significant improvement over the CASP13 MULTICOM predictor. Moreover, the template‐free modeling performs better than the template‐based modeling on not only hard targets but also the targets that have homologous templates. The performance of the template‐free modeling largely depends on the accuracy of distance prediction closely related to the quality of multiple sequence alignments. The structural model quality assessment works well on targets for which enough good models can be predicted, but it may perform poorly when only a few good models are predicted for a hard target and the distribution of model quality scores is highly skewed. MULTICOM is available at https://github.com/jianlin-cheng/MULTICOM_Human_CASP14/tree/CASP14_DeepRank3 and https://github.com/multicom-toolbox/multicom/tree/multicom_v2.0.

TopDomain: Exhaustive Protein Domain Boundary Metaprediction Combining Multisource Information and Deep Learning

Protein domains are independent, functional, and stable structural units of proteins. Accurate protein domain boundary prediction plays an important role in understanding protein structure and evolution, as well as for protein structure prediction. Current domain boundary prediction methods differ in terms of boundary definition, methodology, and training databases resulting in disparate performance for different proteins. We developed TopDomain, an exhaustive metapredictor, that uses deep neural networks to combine multisource information from sequence- and homology-based features of over 50 primary predictors. For this purpose, we developed a new domain boundary data set termed the TopDomain data set, in which the true annotations are informed by SCOPe annotations, structural domain parsers, human inspection, and deep learning. We benchmark TopDomain against 2484 targets with 3354 boundaries from the TopDomain test set and achieve F1 scores of 78.4% and 73.8% for multidomain boundary prediction within ±20 residues and ±10 residues of the true boundary, respectively. When examined on targets from CASP11-13 competitions, TopDomain achieves F1 scores of 47.5% and 42.8% for multidomain proteins. TopDomain significantly outperforms 15 widely used, state-of-the-art ab initio and homology-based domain boundary predictors. Finally, we implemented TopDomain_TMC, which accurately predicts whether domain parsing is necessary for the target protein.

COMTOP: Protein Residue-Residue Contact Prediction through Mixed Integer Linear Optimization

Protein contact prediction helps reconstruct the tertiary structure that greatly determines a protein’s function; therefore, contact prediction from the sequence is an important problem. Recently there has been exciting progress on this problem, but many of the existing methods are still low quality of prediction accuracy. In this paper, we present a new mixed integer linear programming (MILP)-based consensus method: a Consensus scheme based On a Mixed integer linear opTimization method for prOtein contact Prediction (COMTOP). The MILP-based consensus method combines the strengths of seven selected protein contact prediction methods, including CCMpred, EVfold, DeepCov, NNcon, PconsC4, plmDCA, and PSICOV, by optimizing the number of correctly predicted contacts and achieving a better prediction accuracy. The proposed hybrid protein residue–residue contact prediction scheme was tested in four independent test sets. For 239 highly non-redundant proteins, the method showed a prediction accuracy of 59.68%, 70.79%, 78.86%, 89.04%, 94.51%, and 97.35% for top-5L, top-3L, top-2L, top-L, top-L/2, and top-L/5 contacts, respectively. When tested on the CASP13 and CASP14 test sets, the proposed method obtained accuracies of 75.91% and 77.49% for top-L/5 predictions, respectively. COMTOP was further tested on 57 non-redundant α-helical transmembrane proteins and achieved prediction accuracies of 64.34% and 73.91% for top-L/2 and top-L/5 predictions, respectively. For all test datasets, the improvement of COMTOP in accuracy over the seven individual methods increased with the increasing number of predicted contacts. For example, COMTOP performed much better for large number of contact predictions (such as top-5L and top-3L) than for small number of contact predictions such as top-L/2 and top-L/5. The results and analysis demonstrate that COMTOP can significantly improve the performance of the individual methods; therefore, COMTOP is more robust against different types of test sets. COMTOP also showed better/comparable predictions when compared with the state-of-the-art predictors.

DNCON2_Inter: predicting interchain contacts for homodimeric and homomultimeric protein complexes using multiple sequence alignments of monomers and deep learning

Deep learning methods that achieved great success in predicting intrachain residue-residue contacts have been applied to predict interchain contacts between proteins. However, these methods require multiple sequence alignments (MSAs) of a pair of interacting proteins (dimers) as input, which are often difficult to obtain because there are not many known protein complexes available to generate MSAs of sufficient depth for a pair of proteins. In recognizing that multiple sequence alignments of a monomer that forms homomultimers contain the co-evolutionary signals of both intrachain and interchain residue pairs in contact, we applied DNCON2 (a deep learning-based protein intrachain residue-residue contact predictor) to predict both intrachain and interchain contacts for homomultimers using multiple sequence alignment (MSA) and other co-evolutionary features of a single monomer followed by discrimination of interchain and intrachain contacts according to the tertiary structure of the monomer. We name this tool DNCON2_Inter. Allowing true-positive predictions within two residue shifts, the best average precision was obtained for the Top-L/10 predictions of 22.9% for homodimers and 17.0% for higher-order homomultimers. In some instances, especially where interchain contact densities are high, DNCON2_Inter predicted interchain contacts with 100% precision. We also developed Con_Complex, a complex structure reconstruction tool that uses predicted contacts to produce the structure of the complex. Using Con_Complex, we show that the predicted contacts can be used to accurately construct the structure of some complexes. Our experiment demonstrates that monomeric multiple sequence alignments can be used with deep learning to predict interchain contacts of homomeric proteins.

Structural discrimination analysis for constraint selection in protein modeling

MOTIVATION

Protein structure modeling can be improved by the use of distance constraints between amino acid residues, provided such data reflects-at least partially-the native tertiary structure of the target system. In fact, only a small subset of the native contact map is necessary to successfully drive the model conformational search, so one important goal is to obtain the set of constraints with the highest true-positive rate, lowest redundancy, and greatest amount of information. In this work, we introduce a constraint evaluation and selection method based on the point-biserial correlation coefficient, which utilizes structural information from an ensemble of models to indirectly measure the power of each constraint in biasing the conformational search towards consensus structures.

RESULTS

Residue contact maps obtained by direct coupling analysis are systematically improved by means of discriminant analysis, reaching in some cases accuracies often seen only in modern deep-learning based approaches. When combined with an iterative modeling workflow, the proposed constraint classification optimizes the selection of the constraint set and maximizes the probability of obtaining successful models. The use of discriminant analysis for the valorization of the information of constraint data sets is a general concept with possible applications to other constraint types and modeling problems.

AVAILABILITY AND IMPLEMENTATION

scripts and procedures to implement the methodology presented herein are available at https://github.com/m3g/2021_Bottino_Biserial.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.