Single-stranded DNA binding proteins (SSBs) represent one important type of nucleic acid binding proteins and are crucial for many cellular functions, such as DNA replication, transcriptional regulation, and maintaining genome stability and integrity. Compared to other types of nucleic acid binding proteins, SSBs are underexplored due to a relatively small number of annotated SSBs and experimentally solved SSB structures. Only a few computational approaches have been developed so far to classify SSBs from double-stranded DNA binding proteins (DSBs), which typically assume the inputs are known DNA binding proteins (DBPs). However, our recent study reveals that a large number of SSBs are predicted as RNA binding proteins (RBPs) for any given nucleic acid binding proteins, suggesting that SSB/DSB classification alone is not enough and SSB/RBP classification is equally important in accurately predicting novel SSBs from proteins with unknown functions. In this study, in addition to a much improved SSB/DSB classifier, we developed the first SSB/RBP classifier with machine learning models. Besides the widely used PSSM (position specific scoring matrix), we tested a structural feature and a feature extracted from a protein language model ESM2 for training and testing. Our results show that ESM2 improves prediction accuracy dramatically, up to 95% for the SSB/DSB classifier and 92% for the SSB/RBP classifier respectively with a support vector machine (SVM) approach.

Nucleic acid-binding proteins (NABPs), including DNA-binding proteins (DBPs) and RNA-binding proteins (RBPs), play important roles in essential biological processes. To facilitate functional annotation and accurate prediction of different types of NABPs, many machine learning-based computational approaches have been developed. However, the datasets used for training and testing as well as the prediction scopes in these studies have limited their applications. In this paper, we developed new strategies to overcome these limitations by generating more accurate and robust datasets and developing deep learning-based methods including both hierarchical and multi-class approaches to predict the types of NABPs for any given protein. The deep learning models employ two layers of convolutional neural network and one layer of long short-term memory. Our approaches outperform existing DBP and RBP predictors with a balanced prediction between DBPs and RBPs, and are more practically useful in identifying novel NABPs. The multi-class approach greatly improves the prediction accuracy of DBPs and RBPs, especially for the DBPs with ~12% improvement. Moreover, we explored the prediction accuracy of single-stranded DNA binding proteins and their effect on the overall prediction accuracy of NABP predictions.

BACKGROUND: Atomic details of protein-DNA complexes can provide insightful information for better understanding of the function and binding specificity of DNA binding proteins. In addition to experimental methods for solving protein-DNA complex structures, protein-DNA docking can be used to predict native or near-native complex models. A docking program typically generates a large number of complex conformations and predicts the complex model(s) based on interaction energies between protein and DNA. However, the prediction accuracy is hampered by current approaches to model assessment, especially when docking simulations fail to produce any near-native models. RESULTS: We present here a Support Vector Machine (SVM)-based approach for quality assessment of the predicted transcription factor (TF)-DNA complex models. Besides a knowledge-based protein-DNA interaction potential DDNA3, we applied several structural features that have been shown to play important roles in binding specificity between transcription factors and DNA molecules to quality assessment of complex models. To address the issue of unbalanced positive and negative cases in the training dataset, we applied hard-negative mining, an iterative training process that selects an initial training dataset by combining all of the positive cases and a random sample from the negative cases. Results show that the SVM model greatly improves prediction accuracy (84.2%) over two knowledge-based protein-DNA interaction potentials, orientation potential (60.8%) and DDNA3 (68.4%). The improvement is achieved through reducing the number of false positive predictions, especially for the hard docking cases, in which a docking algorithm fails to produce any near-native complex models. CONCLUSIONS: A learning-based SVM scoring model with structural features for specific protein-DNA binding and an atomic-level protein-DNA interaction potential DDNA3 significantly improves prediction accuracy of complex models by successfully identifying cases without near-native structural models.

Protein-DNA docking is an important computational technique for generating native or near-native complex models. A docking program typically generates a number of complex conformations and predicts the docking solution based on interaction energies. However, incomplete sampling and energy function deficiencies can result in false positive protein-DNA complex models, which hampers its application in biology or medicine. Built upon our investigation of structural features for binding specificity between protein and DNA molecules, we present here a Support Vector Machine (SVM)-based approach for quality assessment of the docked transcription factor-DNA complex models by combining structural features and a knowledge-based protein-DNA interaction potential. Our results show that the SVM scoring model greatly improves the prediction accuracy by successfully identifying the false positive cases, in which the docking algorithm fails to produce any near-native complex models.

BACKGROUND: Gene expression is regulated by transcription factors binding to specific target DNA sites. Understanding how and where transcription factors bind at genome scale represents an essential step toward our understanding of gene regulation networks. Previously we developed a structure-based method for prediction of transcription factor binding sites using an integrative energy function that combines a knowledge-based multibody potential and two atomic energy terms. While the method performs well, it is not computationally efficient due to the exponential increase in the number of binding sequences to be evaluated for longer binding sites. In this paper, we present an efficient pentamer algorithm by splitting DNA binding sequences into overlapping fragments along with a simplified integrative energy function for transcription factor binding site prediction. RESULTS: A DNA binding sequence is split into overlapping pentamers (5 base pairs) for calculating transcription factor-pentamer interaction energy. To combine the results from overlapping pentamer scores, we developed two methods, Kmer-Sum and PWM (Position Weight Matrix) stacking, for full-length binding motif prediction. Our results show that both Kmer-Sum and PWM stacking in the new pentamer approach along with a simplified integrative energy function improved transcription factor binding site prediction accuracy and dramatically reduced computation time, especially for longer binding sites. CONCLUSION: Our new fragment-based pentamer algorithm and simplified energy function improve both efficiency and accuracy. To our knowledge, this is the first fragment-based method for structure-based transcription factor binding sites prediction.

UNLABELLED: Transcription factors (TFs) regulate gene expression through binding to specific target DNA sites. Accurate annotation of transcription factor binding sites (TFBSs) at genome scale represents an essential step toward our understanding of gene regulation networks. In this article, we present a structure-based method for computational prediction of TFBSs using a novel, integrative energy (IE) function. The new energy function combines a multibody (MB) knowledge-based potential and two atomic energy terms (hydrogen bond and π interaction) that might not be accurately captured by the knowledge-based potential owing to the mean force nature and low count problem. We applied the new energy function to the TFBS prediction using a non-redundant dataset that consists of TFs from 12 different families. Our results show that the new IE function improves the prediction accuracy over the knowledge-based, statistical potentials, especially for homeodomain TFs, the second largest TF family in mammals. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Computational modeling of protein-DNA complexes remains a challenging problem in structural bioinformatics. One of the key factors for a successful protein-DNA docking is a potential function that can accurately discriminate the near-native structures from decoy complexes and at the same time make conformational sampling more efficient. Here, we developed a novel orientation-dependent, knowledge-based, residue-level potential for improving transcription factor (TF)-DNA docking. RESULTS: We demonstrated the performance of this new potential in TF-DNA binding affinity prediction, discrimination of native protein-DNA complex from decoy structures, and most importantly in rigid TF-DNA docking. The rigid TF-DNA docking with the new orientation potential, on a benchmark of 38 complexes, successfully predicts 42% of the cases with root mean square deviations lower than 1 Å and 55% of the cases with root mean square deviations lower than 3 Å. The results suggest that docking with this new orientation-dependent, coarse-grained statistical potential can achieve high-docking accuracy and can serve as a crucial first step in multi-stage flexible protein-DNA docking. AVAILABILITY AND IMPLEMENTATION: The new potential is available at http://bioinfozen.uncc.edu/Protein_DNA_orientation_potential.tar.

BACKGROUND: One of the crucial steps in regulation of gene expression is the binding of transcription factor(s) to specific DNA sequences. Knowledge of the binding affinity and specificity at a structural level between transcription factors and their target sites has important implications in our understanding of the mechanism of gene regulation. Due to their unique functions and binding specificity, there is a need for a transcription factor-specific, structure-based database and corresponding web service to facilitate structural bioinformatics studies of transcription factor-DNA interactions, such as development of knowledge-based interaction potential, transcription factor-DNA docking, binding induced conformational changes, and the thermodynamics of protein-DNA interactions. DESCRIPTION: TFinDit is a relational database and a web search tool for studying transcription factor-DNA interactions. The database contains annotated transcription factor-DNA complex structures and related data, such as unbound protein structures, thermodynamic data, and binding sequences for the corresponding transcription factors in the complex structures. TFinDit also provides a user-friendly interface and allows users to either query individual entries or generate datasets through culling the database based on one or more search criteria. CONCLUSIONS: TFinDit is a specialized structural database with annotated transcription factor-DNA complex structures and other preprocessed data. We believe that this database/web service can facilitate the development and testing of TF-DNA interaction potentials and TF-DNA docking algorithms, and the study of protein-DNA recognition mechanisms.

Protein-DNA docking is a very challenging problem in bioinformatics and has important implications in a number of applications (e.g. rational drug design). This paper presents a computational approach to improve the prediction accuracy of protein-DNA docking. One of the major difficulties in protein-DNA docking is the high cost of sampling the conformational space. To address this problem, we develop a graphics processing unit (GPU)-based approach to accelerate the sampling process, and thereby improving the quality of conformational space sampling. The effectiveness of the our approach is validated against a non-redundant set of 75 protein-DNA complexes, and the results demonstrate improved performance in finding near-native protein-DNA complex structures. To the best of our knowledge, this is the first ad hoc effort of applying GPU or GPU clusters to the protein-DNA docking problem.