Statistical mechanical modeling of genome-wide transcription factor occupancy data by MatrixREDUCE

Knowing the sequence specificities of DNA-and RNA-binding proteins is essential for developing models of the regulatory processes in biological systems and for identifying causal disease variants. Here we show that sequence specificities can be ascertained from experimental data with 'deep learning' techniques, which offer a scalable, flexible and unified computational approach for pattern discovery. Using a diverse array of experimental data and evaluation metrics, we find that deep learning outperforms other state-of-the-art methods, even when training on in vitro data and testing on in vivo data. We call this approach DeepBind and have built a stand-alone software tool that is fully automatic and handles millions of sequences per experiment. Specificities determined by DeepBind are readily visualized as a weighted ensemble of position weight matrices or as a 'mutation map' that indicates how variations affect binding within a specific sequence.DNA-and RNA-binding proteins play a central role in gene regulation, including transcription and alternative splicing. The sequence specificities of a protein are most commonly characterized using position weight matrices 1 (PWMs), which are easy to interpret and can be scanned over a genomic sequence to detect potential binding sites. However, growing evidence indicates that sequence specificities can be more accurately captured by more complex techniques 2-5 . Recently, 'deep learning' has achieved record-breaking performance in a variety of information technology applications 6,7 . We adapted deep learning methods to the task of predicting sequence specificities and found that they compete favorably with the state of the art. Our approach, called DeepBind, is based on deep convolutional neural networks and can discover new patterns even when the locations of patterns within sequences are unknown-a task for which traditional neural networks require an exorbitant amount of training data.There are several challenging aspects in learning models of sequence specificity using modern high-throughput technologies. First, the data come in qualitatively different forms. Protein binding microarrays (PBMs) 8 and RNAcompete assays 9 provide a specificity coefficient for each probe sequence, whereas chromatin immunoprecipitation (ChIP)-seq 10 provides a ranked list of putatively bound sequences of varying length, and HT-SELEX 11 generates a set of very high affinity sequences. Second, the quantity of data is large. A typical high-throughput experiment measures between 10,000 and 100,000 sequences, and it is computationally demanding to incorporate them all. Third, each data acquisition technology has its own artifacts, biases and limitations, and we must discover the pertinent specificities despite these unwanted effects. For example, ChIP-seq reads often localize to "hyper-ChIPable" regions of the genome near highly expressed genes 12 .DeepBind (Fig. 1) addresses the above challenges. (i) It can be applied to both microarray and sequencing data; (ii) it can learn from millions of...

show abstract

“…DeepBind consistently had better overall performance than MartixREDUCE 19 and Ray et al's 9 PWM_align_Z method (Fig. 3c).…”

Section: A N a Ly S I Smentioning

confidence: 84%

Predicting the sequence specificities of DNA- and RNA-binding proteins by deep learning

et al. 2015

View full text Add to dashboard Cite

show abstract

“…Other models such as MatrixREDUCE (Foat et al 2006) or TRAP (Roider et al 2007) could also be used to compute the probability that a TF with a particular PWM binds a DNA sequence. However, both MatrixREDUCE and TRAP use parameters that need to be trained on the ChIP-chip data.…”

Section: Scoring a Dna Sequence According To A Pwmmentioning

confidence: 99%

Distinguishing Direct versus Indirect Transcription Factor-DNA Interactions

Gordân

Hartemink

Bulyk

2010

Lecture Notes in Computer Science

View full text Add to dashboard Cite

“…The key observations are (i) that the binding site specificity is largely imparted by the contacts between DNA bases and amino acid side chains and (ii) that the number of atomic contacts with a base pair is strongly correlated with the degree of conservation of that base pair in the binding site (13). Therefore, a structure of the TF-DNA complex and a single consensus (highest affinity) sequence can be used to predict the position-specific weight matrix (PWM) for the TF (14,15). We make PWM predictions for 67 S. cerevisiae TFs by using homology models (we use ''homology'' to imply similarity rather than evolutionary relationship).…”

mentioning

confidence: 99%

Connecting protein structure with predictions of regulatory sites

Morozov

Siggia²

2007

Proc. Natl. Acad. Sci. U.S.A.

View full text Add to dashboard Cite

A common task posed by microarray experiments is to infer the binding site preferences for a known transcription factor from a collection of genes that it regulates and to ascertain whether the factor acts alone or in a complex. The converse problem can also be posed: Given a collection of binding sites, can the regulatory factor or complex of factors be inferred? Both tasks are substantially facilitated by using relatively simple homology models for protein-DNA interactions, as well as the rapidly expanding protein structure database. For budding yeast, we are able to construct reliable structural models for 67 transcription factors and with them redetermine factor binding sites by using a Bayesian Gibbs sampling algorithm and an extensive protein localization data set. For 49 factors in common with a prior analysis of this data set (based largely on phylogenetic conservation), we find that half of the previously predicted binding motifs are in need of some revision. We also solve the inverse problem of ascertaining the factors from the binding sites by assigning a correct protein fold to 25 of the 49 cases from a previous study. Our approach is easily extended to other organisms, including higher eukaryotes. Our study highlights the utility of enlarging current structural genomics projects that exhaustively sample fold structure space to include all factors with significantly different DNA-binding specificities.protein-DNA interactions ͉ homology models of transcription factors ͉ weight matrix predictions T ranscription factors (TFs) are regulatory proteins used by the cell to activate or repress gene transcription. They interact with short nucleotide sequences, typically located upstream of a gene, by means of the DNA-binding domains that recognize their cognate binding sites. As a rule, regulation of gene transcription is analyzed by the bioinformatics methods designed to detect statistically overrepresented motifs in promoter sequences. Intergenic sequences bound by the TF can be identified by using DNA microarray technology, including chromatin immunoprecipitation (ChIPchip) (1, 2), protein binding (3), and DNA immunoprecipitation (DIP-chip) arrays (4). Of special note is a recent genome-wide study that used ChIP-chip analysis to profile in vivo genomic occupancies for 203 DNA-binding transcriptional regulators in Saccharomyces cerevisiae (2). Using these data, the authors predicted binding specificities for 65 TFs by using the genomes of related species; a number that was later increased to 98 by MacIsaac et al. (5) The DNA-binding domains of TFs can be classified into a limited number of structural families (6, 7). Structural studies of the protein-DNA complexes reveal that, within each family, the overall fold of the DNA-binding domain and its mode of interaction with the cognate binding site are remarkably conserved, resulting in a characteristic pattern of amino acid contacts with DNA bases. These interactions form the basis of the sequence-specific direct readout of nucleotide sequences by amino acid...

show abstract

Statistical mechanical modeling of genome-wide transcription factor occupancy data by MatrixREDUCE

Abstract: MatrixREDUCE source code is freely available for non-commercial use at http://www.bussemakerlab.org/. The software runs on Linux, Unix, and Mac OS X.

Cited by 268 publications

References 38 publications

Predicting the sequence specificities of DNA- and RNA-binding proteins by deep learning

Predicting the sequence specificities of DNA- and RNA-binding proteins by deep learning

Distinguishing Direct versus Indirect Transcription Factor-DNA Interactions

Connecting protein structure with predictions of regulatory sites

Contact Info

Product

Resources

About