Proteins with desired functions and properties are important in fields like nanotechnology and biomedicine. De novo protein design enables the production of previously unseen proteins from the ground up and is believed as a key point for handling real social challenges. Recent introduction of deep learning into design methods exhibits a transformative influence and is expected to represent a promising and exciting future direction. In this review, we retrospect the major aspects of current advances in deep-learning-based design procedures and illustrate their novelty in comparison with conventional knowledge-based approaches through noticeable cases. We not only describe deep learning developments in structure-based protein design and direct sequence design, but also highlight recent applications of deep reinforcement learning in protein design. The future perspectives on design goals, challenges and opportunities are also comprehensively discussed.

MOTIVATION: As one of the most important post-translational modifications (PTMs), protein lysine crotonylation (Kcr) has attracted wide attention, which involves in important physiological activities, such as cell differentiation and metabolism. However, experimental methods are expensive and time-consuming for Kcr identification. Instead, computational methods can predict Kcr sites in silico with high efficiency and low cost. RESULTS: In this study, we proposed a novel predictor, BERT-Kcr, for protein Kcr sites prediction, which was developed by using a transfer learning method with pre-trained bidirectional encoder representations from transformers (BERT) models. These models were originally used for natural language processing (NLP) tasks, such as sentence classification. Here, we transferred each amino acid into a word as the input information to the pre-trained BERT model. The features encoded by BERT were extracted and then fed to a BiLSTM network to build our final model. Compared with the models built by other machine learning and deep learning classifiers, BERT-Kcr achieved the best performance with AUROC of 0.983 for 10-fold cross validation. Further evaluation on the independent test set indicates that BERT-Kcr outperforms the state-of-the-art model Deep-Kcr with an improvement of about 5% for AUROC. The results of our experiment indicate that the direct use of sequence information and advanced pre-trained models of NLP could be an effective way for identifying PTM sites of proteins. AVAILABILITY AND IMPLEMENTATION: The BERT-Kcr model is publicly available on http://zhulab.org.cn/BERT-Kcr_models/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Gradient descent-based protein modeling is a popular protein structure prediction approach that takes as input the predicted inter-residue distances and other necessary constraints and folds protein structures by minimizing protein-specific energy potentials. The constraints from multiple predicted protein properties provide redundant and sometime conflicting information that can trap the optimization process into local minima and impairs the modeling efficiency. RESULTS: To address these issues, we developed a self-adaptive protein modeling framework, SAMF. It eliminates redundancy of constraints and resolves conflicts, folds protein structures in an iterative way, and picks up the best structures by a deep quality analysis system. Without a large amount of complicated domain knowledge and numerous patches as barriers, SAMF achieves the state-of-the-art performance by exploiting the power of cutting-edge techniques of deep learning. SAMF has a modular design and can be easily customized and extended. As the quality of input constraints is ever growing, the superiority of SAMF will be amplified over time. AVAILABILITY AND IMPLEMENTATION: The source code and data for reproducing the results is available at https://msracb.blob.core.windows.net/pub/psp/SAMF.zip. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Abstract Background Despite the great advance of protein structure prediction, accurate prediction of the structures of mainly β proteins is still highly challenging, but could be assisted by the knowledge of residue-residue pairing in β strands. Previously, we proposed a ridge-detection-based algorithm RDb2C that adopted a multi-stage random forest framework to predict the β-β pairing given the amino acid sequence of a protein. Results In this work, we developed a second version of this algorithm, RDb2C2, by employing the residual neural network to further enhance the prediction accuracy. In the benchmark test, this new algorithm improves the F1-score by > 10 percentage points, reaching impressively high values of ~ 72% and ~ 73% in the BetaSheet916 and BetaSheet1452 sets, respectively. Conclusion Our new method promotes the prediction accuracy of β-β pairing to a new level and the prediction results could better assist the structure modeling of mainly β proteins. We prepared an online server of RDb2C2 at http://structpred.life.tsinghua.edu.cn/rdb2c2.html .

Despite the rapid progress of protein residue contact prediction, predicted residue contact maps frequently contain many errors. However, information of residue pairing in β strands could be extracted from a noisy contact map, due to the presence of characteristic contact patterns in β-β interactions. This information may benefit the tertiary structure prediction of mainly β proteins. In this work, we propose a novel ridge-detection-based β-β contact predictor to identify residue pairing in β strands from any predicted residue contact map. Our algorithm RDb2C adopts ridge detection, a well-developed technique in computer image processing, to capture consecutive residue contacts, and then utilizes a novel multi-stage random forest framework to integrate the ridge information and additional features for prediction. Starting from the predicted contact map of CCMpred, RDb2C remarkably outperforms all state-of-the-art methods on two conventional test sets of β proteins (BetaSheet916 and BetaSheet1452), and achieves F1-scores of ~ 62% and ~ 76% at the residue level and strand level, respectively. Taking the prediction of the more advanced RaptorX-Contact as input, RDb2C achieves impressively higher performance, with F1-scores reaching ~ 76% and ~ 86% at the residue level and strand level, respectively. In a test of structural modeling using the top 1 L predicted contacts as constraints, for 61 mainly β proteins, the average TM-score achieves 0.442 when using the raw RaptorX-Contact prediction, but increases to 0.506 when using the improved prediction by RDb2C. Our method can significantly improve the prediction of β-β contacts from any predicted residue contact maps. Prediction results of our algorithm could be directly applied to effectively facilitate the practical structure prediction of mainly β proteins. All source data and codes are available at http://166.111.152.91/Downloads.html or the GitHub address of https://github.com/wzmao/RDb2C .

Motivation: The quality of fragment library determines the efficiency of fragment assembly, an approach that is widely used in most de novo protein-structure prediction algorithms. Conventional fragment libraries are constructed mainly based on the identities of amino acids, sometimes facilitated by predicted information including dihedral angles and secondary structures. However, it remains challenging to identify near-native fragment structures with low sequence homology. Results: We introduce a novel fragment-library-construction algorithm, LRFragLib, to improve the detection of near-native low-homology fragments of 7-10 residues, using a multi-stage, flexible selection protocol. Based on logistic regression scoring models, LRFragLib outperforms existing techniques by achieving a significantly higher precision and a comparable coverage on recent CASP protein sets in sampling near-native structures. The method also has a comparable computational efficiency to the fastest existing techniques with substantially reduced memory usage. Availability and Implementation: The source code is available for download at http://166.111.152.91/Downloads.html. Contact: [email protected]. Supplementary information: Supplementary data are available at Bioinformatics online.

MOTIVATION: Residue-residue contacts are of great value for protein structure prediction, since contact information, especially from those long-range residue pairs, can significantly reduce the complexity of conformational sampling for protein structure prediction in practice. Despite progresses in the past decade on protein targets with abundant homologous sequences, accurate contact prediction for proteins with limited sequence information is still far from satisfaction. Methodologies for these hard targets still need further improvement. RESULTS: We presented a computational program DeepConPred, which includes a pipeline of two novel deep-learning-based methods (DeepCCon and DeepRCon) as well as a contact refinement step, to improve the prediction of long-range residue contacts from primary sequences. When compared with previous prediction approaches, our framework employed an effective scheme to identify optimal and important features for contact prediction, and was only trained with coevolutionary information derived from a limited number of homologous sequences to ensure robustness and usefulness for hard targets. Independent tests showed that 59.33%/49.97%, 64.39%/54.01% and 70.00%/59.81% of the top L/5, top L/10 and top 5 predictions were correct for CASP10/CASP11 proteins, respectively. In general, our algorithm ranked as one of the best methods for CASP targets. AVAILABILITY AND IMPLEMENTATION: All source data and codes are available at http://166.111.152.91/Downloads.html . CONTACT: [email protected] or [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

GLUT1 facilitates the down-gradient translocation of D-glucose across cell membrane in mammals. XylE, an Escherichia coli homolog of GLUT1, utilizes proton gradient as an energy source to drive uphill D-xylose transport. Previous studies of XylE and GLUT1 suggest that the variation between an acidic residue (Asp27 in XylE) and a neutral one (Asn29 in GLUT1) is a key element for their mechanistic divergence. In this work, we combined computational and biochemical approaches to investigate the mechanism of proton coupling by XylE and the functional divergence between GLUT1 and XylE. Using molecular dynamics simulations, we evaluated the free energy profiles of the transition between inward- and outward-facing conformations for the apo proteins. Our results revealed the correlation between the protonation state and conformational preference in XylE, which is supported by the crystal structures. In addition, our simulations suggested a thermodynamic difference between XylE and GLUT1 that cannot be explained by the single residue variation at the protonation site. To understand the molecular basis, we applied Bayesian network models to analyze the alteration in the architecture of the hydrogen bond networks during conformational transition. The models and subsequent experimental validation suggest that multiple residue substitutions are required to produce the thermodynamic and functional distinction between XylE and GLUT1. Despite the lack of simulation studies with substrates, these computational and biochemical characterizations provide unprecedented insight into the mechanistic difference between proton symporters and uniporters.

Protein phosphorylation is one of the most pervasive post-translational modifications, regulating diverse cellular processes in various organisms. As mass spectrometry-based experimental approaches for identifying phosphorylation events are resource-intensive, many computational methods have been proposed, in which phosphorylation site prediction is formulated as a classification problem. They differ in several ways, and one crucial issue is the construction of training data and test data for unbiased performance evaluation. In this article, we categorize the existing data construction methods and try to answer three questions: (i) Is it equivalent to use different data construction methods in the assessment of phosphorylation site prediction algorithms? (ii) What kind of test data set is unbiased for assessing the prediction performance of a trained algorithm in different real world scenarios? (iii) Among the summarized training data construction methods, which one(s) has better generalization performance for most scenarios? To answer these questions, we conduct comprehensive experimental studies for both non-kinase-specific and kinase-specific prediction tasks. The experimental results show that: (i) different data construction methods can lead to significantly different prediction performance; (ii) there can be different test data construction methods that are unbiased with respect to different real world scenarios; and (iii) different data construction methods have different generalization performance in different real world scenarios. Therefore, when developing new algorithms in future research, people should concentrate on what kind of scenario their algorithm will work for, what the corresponding unbiased test data are and which training data construction method can generate best generalization performance.

Phosphorylation motifs represent position-specific amino acid patterns around the phosphorylation sites in the set of phosphopeptides. Several algorithms have been proposed to uncover phosphorylation motifs, whereas the problem of efficiently discovering a set of significant motifs with sufficiently high coverage and non-redundancy still remains unsolved. Here we present a novel notion called conditional phosphorylation motifs. Through this new concept, the motifs whose over-expressiveness mainly benefits from its constituting parts can be filtered out effectively. To discover conditional phosphorylation motifs, we propose an algorithm called C-Motif for a non-redundant identification of significant phosphorylation motifs. C-Motif is implemented under the Apriori framework, and it tests the statistical significance together with the frequency of candidate motifs in a single stage. Experiments demonstrate that C-Motif outperforms some current algorithms such as MMFPh and Motif-All in terms of coverage and non-redundancy of the results and efficiency of the execution. The source code of C-Motif is available at: https://sourceforge. net/projects/cmotif/.

Abstract Contact: [email protected]