NetTIME: a multitask and base-pair resolution framework for improved transcription factor binding site prediction

A flexible transcription initiation architecture and a uniform regulatory grammar buffer expression variability and the effects of genetic variants on essential genes while a rigid transcription initiation architecture and a diverse regulatory grammar ensure condition-specific responsiveness of other genes.

SNP2TFBS is a computational resource intended to support researchers investigating the molecular mechanisms underlying regulatory variation in the human genome. The database essentially consists of a collection of text files providing specific annotations for human single nucleotide polymorphisms (SNPs), namely whether they are predicted to abolish, create or change the affinity of one or several transcription factor (TF) binding sites. A SNP's effect on TF binding is estimated based on a position weight matrix (PWM) model for the binding specificity of the corresponding factor. These data files are regenerated at regular intervals by an automatic procedure that takes as input a reference genome, a comprehensive SNP catalogue and a collection of PWMs. SNP2TFBS is also accessible over a web interface, enabling users to view the information provided for an individual SNP, to extract SNPs based on various search criteria, to annotate uploaded sets of SNPs or to display statistics about the frequencies of binding sites affected by selected SNPs. Homepage: http://ccg.vital-it.ch/snp2tfbs/.

Transcription factor (TF) binding at specific DNA sequences is the fundamental step in transcriptional regulation and is highly dependent on the chromatin structure context, which may be affected by specific histone modifications and variants, known as histone marks. The lack of a global binding map for hundreds of TFs means that previous studies have focused mainly on histone marks at binding sites for several specific TFs. We therefore studied 11 histone marks around computationally-inferred and experimentally-determined TF binding sites (TFBSs), based on 164 and 34 TFs, respectively, in human lymphoblastoid cell lines. For H2A.Z, methylation of H3K4, and acetylation of H3K27 and H3K9, the mark patterns exhibited bimodal distributions and strong pairwise correlations in the 600-bp region around enriched TFBSs, suggesting that these marks mainly coexist within the two nucleosomes proximal to the TF sites. TFs competing with nucleosomes to access DNA at most binding sites, contributes to the bimodal distribution, which is a common feature of histone marks for TF binding. Mark H3K79me2 showed a unimodal distribution on one side of TFBSs and the signals extended up to 4000 bp, indicating a longer-distance pattern. Interestingly, H4K20me1, H3K27me3, H3K36me3 and H3K9me3, which were more diffuse and less enriched surrounding TFBSs, showed unimodal distributions around the enriched TFBSs, suggesting that some TFs may bind to nucleosomal DNA. Besides, asymmetrical distributions of H3K36me3 and H3K9me3 indicated that repressors might establish a repressive chromatin structure in one direction to repress gene expression. In conclusion, this study demonstrated the ranges of histone marks associated with TF binding, and the common features of these marks around the binding sites. These findings have epigenetic implications for future analysis of regulatory elements.

Transcription factor (TF) binding is a key component of genomic regulation. Therefore, experimental methods exist for their identification. These, however, remain exacting and consequently their use is limited. For instance, the binding specificities of 25% human TFs remain uncharacterized; they do not have a known (i.e. determined experimentally) or computationally inferred motif. Here, we introduce ModCRE, a web server implementing a structure homology-modelling approach to predict TF binding motifs in order to identify cis-regulatory elements and automatically model higher-order TF regulatory complexes. ModCRE predicts a set of TF binding PWMs of a submitted TF structure or sequence and scans DNA for these. Then, matches are either profiled with a binding score or collected for their integration into a model containing DNA, TFs and transcription co-factors. In this paper we demonstrate how integrating protein-binding microarray (PBM) data addresses the protein-DNA structure scarcity issue in calculating statistical potentials, which in turn are shown to serve as capable predictors of TF binding PWMs. We reveal the conditional advantage of a structure-based over the nearest-neighbor approach in predicting TF binding sites as well as the improvement in both cases by means of rank-enrichment prediction. Finally, we apply ModCRE to automatically model the interferon beta enhanceosome, and the complex of the transcription factors SOX11 and SOX2 with the nucleosome.

The ENCyclopedia of DNA Elements (ENCODE) consortium has generated transcription factor (TF) binding ChIP-seq data covering hundreds of TF proteins and cell types; however, due to limits on time and resources, only a small fraction of all possible TF-cell type pairs have been profiled. One solution is to build machine learning models trained on currently available epigenomic data sets that can be applied to the remaining missing pairs. A major challenge is that TF binding sites are cell-type–specific, which can be attributed to cellular contexts such as chromatin accessibility. Meanwhile, indirect TF-DNA binding and interactions between TFs complicate this regulatory process. Technical issues such as sequencing biases and batch effects render the prediction task even more challenging. Many pioneering efforts have been made to predict TF binding profiles based on DNA sequence and DNase-seq footprints, but to what extent a model can be generalized to completely untested cell conditions remains unknown. In this study, we describe our first place solution to the 2017 ENCODE-DREAM in vivo TF binding site prediction challenge. By carefully addressing multisource biases and information imbalance across cell types, we created a pipeline that significantly outperforms the current state-of-the-art methods. The proposed method is sufficiently complex enough to model nonlinear interactions between TF binding motifs and chromatin accessibility information up to 1500 bp from the genomic region of interest.

Acetylcholinesterase in man is encoded by a single gene, ACHE, located on chromosome 7q22. In this study, the transcription start sites and major DNA promoter elements controlling the expression of this gene have been characterized by structural and functional studies. Immediately upstream of the first untranslated exon of the gene are GC-rich sequences containing consensus binding sites for several transcription factors, including Sp1, EGR-1 and AP2. In vitro transcription studies and RNase protection analyses of mRNA isolated from human NT2/D1 teratocarcinoma cells reveal that two closely spaced transcription cap sites are located at a consensus initiator (Inr) element similar to that found in the terminal transferase gene. Transient transfection of mutant genes shows that removal of three bases of this initiator sequence reduces promoter activity by 98% in NT2/D1 cells. In vitro transcription studies and transient transfection of a series of 5' deletion mutants of the ACHE promoter linked to a luciferase reporter show an Sp1 site at -71 to be essential for promoter activity. Purified Sp1 protein protects this site from DNase cleavage during in vitro footprinting experiments. A conserved AP2 consensus binding site, located between the GC box elements and the Inr, is protected by recombinant AP2 protein in DNase footprinting experiments, induces a mobility shift with AP2 protein and AP2-containing cell extracts, and fosters inhibition of transcription by AP2 as measured by transient transfection in mouse and human cell lines and in in vitro transcription reactions. These results indicate that AP2 functions as a repressor of human ACHE and mouse Ache transcription.

Detecting in vivo transcription factor (TF) binding is important for understanding gene regulatory circuitries. ChIP-seq is a powerful technique to empirically define TF binding in vivo. However, the multitude of distinct TFs makes genome-wide profiling for them all labor-intensive and costly. Algorithms for in silico prediction of TF binding have been developed, based mostly on histone modification or DNase I hypersensitivity data in conjunction with DNA motif and other genomic features. However, technical limitations of these methods prevent them from being applied broadly, especially in clinical settings. We conducted a comprehensive survey involving multiple cell lines, TFs, and methylation types and found that there are intimate relationships between TF binding and methylation level changes around the binding sites. Exploiting the connection between DNA methylation and TF binding, we proposed a novel supervised learning approach to predict TF–DNA interaction using data from base-resolution whole-genome methylation sequencing experiments. We devised beta-binomial models to characterize methylation data around TF binding sites and the background. Along with other static genomic features, we adopted a random forest framework to predict TF–DNA interaction. After conducting comprehensive tests, we saw that the proposed method accurately predicts TF binding and performs favorably versus competing methods.

Transcription factor (TF) binding site (TFBS) models are crucial for computational reconstruction of transcription regulatory networks. In existing repositories, a TF often has several models (also called binding profiles or motifs), obtained from different experimental data. Having a single TFBS model for a TF is more pragmatic for practical applications. We show that integration of TFBS data from various types of experiments into a single model typically results in the improved model quality probably due to partial correction of source specific technique bias.We present the Homo sapiens comprehensive model collection (HOCOMOCO, http://autosome.ru/HOCOMOCO/, http://cbrc.kaust.edu.sa/hocomoco/) containing carefully hand-curated TFBS models constructed by integration of binding sequences obtained by both low- and high-throughput methods. To construct position weight matrices to represent these TFBS models, we used ChIPMunk software in four computational modes, including newly developed periodic positional prior mode associated with DNA helix pitch. We selected only one TFBS model per TF, unless there was a clear experimental evidence for two rather distinct TFBS models. We assigned a quality rating to each model. HOCOMOCO contains 426 systematically curated TFBS models for 401 human TFs, where 172 models are based on more than one data source.

Many DNA motif finding algorithms that use Consensus (or any of its variants) in its motif model implicitly impose some restrictive assumptions over transcription factor (TF) binding sites (TFBS). Examples include all binding sites being of equal length, or having exactly one core region with fixed format, etc. In this paper, we have constructed a generalized consensus model (called Mixed-Length-Consensus, or ML-Consensus) without such constraints through multiple sequence alignment of known TFBS. We have extended this model with Information Content (IC) and Pairwise nucleotide correlation Score (PS), and have experimented with using multiple ML-Consensus for a set of binding sites. We have performed leave-one-out cross validation for training and testing of this algorithm over real binding site data of human, mouse, fruit fly, and yeast. We have produced ROC curves (True Positive Rate against False Positive Rate) for these experiments, and have used Wilcoxon Matched-Pair Signed Ranks Test to determine their statistical significance. Our results show that using IC and PS together with ML-Consensus consistently leads to better performance. We have experimented with various scopes for PS, and have found that scope values of 3-5 yield significantly better performance for different configurations. We have also found that using multiple ML-Consensus for one TF significantly improves recognition performance, but single ML-Consensus does better in yeast than in human data. Finally, we have found that using different multiple sequence alignment strategies for ML-Consensus yields varied performance across different species; a naive sorting based multiple sequence alignment outperformed CLUSTAL W2 alignment in yeast data.

More accurate and more complete predictions of cis-regulatory modules (CRMs) and constituent transcription factor (TF) binding sites (TFBSs) in genomes can facilitate characterizing functions of regulatory sequences. Here, we developed a database predicted cis-regulatory modules (PCRMS) (https://cci-bioinfo.uncc.edu) that stores highly accurate and unprecedentedly complete maps of predicted CRMs and TFBSs in the human and mouse genomes. The web interface allows the user to browse CRMs and TFBSs in an organism, find the closest CRMs to a gene, search CRMs around a gene and find all TFBSs of a TF. PCRMS can be a useful resource for the research community to characterize regulatory genomes.Database URL: https://cci-bioinfo.uncc.edu/

We performed a systematic analysis of gene expression features in early (10–21 days) development of human vs mouse embryonic cells (hESCs vs mESCs). Many development features were found to be conserved, and a majority of differentially regulated genes have similar expression change in both organisms. The similarity is especially evident, when gene expression profiles are clustered together and properties of clustered groups of genes are compared. First 10 days of mESC development match the features of hESC development within 21 days, in accordance with the differences in population doubling time in human and mouse ESCs. At the same time, several important differences are seen. There is a clear difference in initial expression change of transcription factors and stimulus responsive genes, which may be caused by the difference in experimental procedures. However, we also found that some biological processes develop differently; this can clearly be shown, for example, for neuron and sensory organ development. Some groups of genes show peaks of the expression levels during the development and these peaks cannot be claimed to happen at the same time points in the two organisms, as well as for the same groups of (orthologous) genes. We also detected a larger number of upregulated genes during development of mESCs as compared to hESCs. The differences were quantified by comparing promoters of related genes. Most of gene groups behave similarly and have similar transcription factor (TF) binding sites on their promoters. A few groups of genes have similar promoters, but are expressed differently in two species. Interestingly, there are groups of genes expressed similarly, although they have different promoters, which can be shown by comparing their TF binding sites. Namely, a large group of similarly expressed cell cycle-related genes is found to have discrepant TF binding properties in mouse vs human.

Transcription factor binding sites (TFBS) in gene promoter regions are often predicted by using position specific scoring matrices (PSSMs), which summarize sequence patterns of experimentally determined TF binding sites. Although PSSMs are more reliable than simple consensus string matching in predicting a true binding site, they generally result in high numbers of false positive hits. This study attempts to reduce the number of false positive matches and generate new predictions by integrating various types of genomic data by two methods: a Bayesian allocation procedure, and support vector machine classification. Several methods will be explored to strengthen the prediction of a true TFBS in the Saccharomyces cerevisiae genome: binding site degeneracy, binding site conservation, phylogenetic profiling, TF binding site clustering, gene expression profiles, GO functional annotation, and k-mer counts in promoter regions. Binding site degeneracy (or redundancy) refers to the number of times a particular transcription factor's binding motif is discovered in the upstream region of a gene. Phylogenetic conservation takes into account the number of orthologous upstream regions in other genomes that contain a particular binding site. Phylogenetic profiling refers to the presence or absence of a gene across a large set of genomes. Binding site clusters are statistically significant clusters of TF binding sites detected by the algorithm ClusterBuster. Gene expression takes into account the idea that when the gene expression profiles of a transcription factor and a potential target gene are correlated, then it is more likely that the gene is a genuine target. Also, genes with highly correlated expression profiles are often regulated by the same TF(s). The GO annotation data takes advantage of the idea that common transcription targets often have related function. Finally, the distribution of the counts of all k-mers of length 4, 5, and 6 in gene's promoter region were examined as means to predict TF binding. In each case the data are compared to known true positives taken from ChIP-chip data, Transfac, and the Saccharomyces Genome Database. First, degeneracy, conservation, expression, and binding site clusters were examined independently and in combination via Bayesian allocation. Then, binding sites were predicted with a support vector machine (SVM) using all methods alone and in combination. The SVM works best when all genomic data are combined, but can also identify which methods contribute the most to accurate classification. On average, a support vector machine can classify binding sites with high sensitivity and an accuracy of almost 80%.

Transcription Factor Binding Affinities and DNA Shape Readout.

Genomic studies have revealed elevated damage and mutation rates in active transcription factor (TF) binding sites in UV-linked cancers. Previous investigations into the relationship between TF activity and UV DNA damage have primarily focused on select TFs or been done in aggregate across large cohorts of TFs at kilobase resolution. While collectively, there is evidence that TFs contribute to UV-induced mutagenesis by both enhancing initial damage formation and attenuating repair, there has yet to be a comprehensive characterization of these mechanisms on a per-TF basis. Using genome-wide maps of UV damage from human skin fibroblasts, we developed a scalable statistical framework to analyze TF-mediated mutagenic mechanisms across hundreds of TFs. We identify numerous previously unreported TFs that significantly enhance and / or inhibit damage formation in their binding sites. A systematic survey of TF-DNA complexes further revealed that positions of UV damage modulation coincide with TF-induced structural distortions that either protect or predispose DNA to photodimer formation. Additionally, we analyzed repair efficiency in TF binding sites with unprecedented resolution, identifying specific TFs and binding site positions likely to compete with repair factors. By comparing these results with skin cancer mutations, we distinguish mutation peaks driven by increased damage susceptibility versus attenuated repair, illustrating that TF-mediated mutagenesis is highly contextual and dependent on the TF, binding site position, and sequence context of the damaged locus. Our approach provides a robust statistical framework for elucidating mechanisms of mutagenic TF-binding and offers novel insights into the complex interplay between protein interactions, DNA damage, and repair.

This work addresses a central question in medical genetics - the distinction between disease-causing SNPs and neutral variations. Unlike previous studies that focused mainly on coding SNPs, our efforts were centered around variations in regulatory regions and specifically within transcription factor (TF) binding sites. We have compiled a comprehensive collection of genome wide TF binding sites and developed computational measures to estimate the effects of binding site variations on the expression profiles of the regulated genes. Applying these measures to binding sites of known TFs, we were able to make predictions that were in line with published experimental evidence and with structural data on DNA-protein interactions. We attempted to generalize the properties of expression-altering substitutions by accumulating statistics from many substitutions across multiple binding sites. We found that in the yeast genome substitutions that abolish a G or a C are on average more severe than substitutions that abolish an A or a T. This may be attributed to the low GC content of the yeast genome, in which G and C may be important for conferring specificity. We found additional factors that are correlated with the severity of a substitution. Such factors can be integrated in order to create a set of rules for the prioritization of regulatory SNPs according to their disease-causing potential.