MOTIVATION: Motifs play a crucial role in computational biology, as they provide valuable information about the binding specificity of proteins. However, conventional motif discovery methods typically rely on simple combinatoric or probabilistic approaches, which can be biased by heuristics such as substring-masking for multiple motif discovery. In recent years, deep neural networks have become increasingly popular for motif discovery, as they are capable of capturing complex patterns in data. Nonetheless, inferring motifs from neural networks remains a challenging problem, both from a modeling and computational standpoint, despite the success of these networks in supervised learning tasks. RESULTS: We present a principled representation learning approach based on a hierarchical sparse representation for motif discovery. Our method effectively discovers gapped, long, or overlapping motifs that we show to commonly exist in next-generation sequencing datasets, in addition to the short and enriched primary binding sites. Our model is fully interpretable, fast, and capable of capturing motifs in a large number of DNA strings. A key concept emerged from our approach-enumerating at the image level-effectively overcomes the k-mers paradigm, enabling modest computational resources for capturing the long and varied but conserved patterns, in addition to capturing the primary binding sites. AVAILABILITY AND IMPLEMENTATION: Our method is available as a Julia package under the MIT license at https://github.com/kchu25/MOTIFs.jl, and the results on experimental data can be found at https://zenodo.org/record/7783033.

BACKGROUND: Transcription factor (TF) binding site specificity is commonly represented by some form of matrix model in which the positions in the binding site are assumed to contribute independently to the site's activity. The independence assumption is known to be an approximation, often a good one but sometimes poor. Alternative approaches have been developed that use k-mers (DNA "words" of length k) to account for the non-independence, and more recently DNA structural parameters have been incorporated into the models. ChIP-seq data are often used to assess the discriminatory power of motifs and to compare different models. However, to measure the improvement due to using more complex models, one must compare to optimized matrix models. RESULTS: We describe a program "Discriminative Additive Model Optimization" (DAMO) that uses positive and negative examples, as in ChIP-seq data, and finds the additive position weight matrix (PWM) that maximizes the Area Under the Receiver Operating Characteristic Curve (AUROC). We compare to a recent study where structural parameters, serving as features in a gradient boosting classifier algorithm, are shown to improve the AUROC over JASPAR position frequency matrices (PFMs). In agreement with the previous results, we find that adding structural parameters gives the largest improvement, but most of the gain can be obtained by an optimized PWM and nearly all of the gain can be obtained with a di-nucleotide extension to the PWM. CONCLUSION: To appropriately compare different models for TF bind sites, optimized models must be used. PWMs and their extensions are good representations of binding specificity for most TFs, and more complex models, including the incorporation of DNA shape features and gradient boosting classifiers, provide only moderate improvements for a few TFs.

MOTIVATION: Characterizing the binding specificities of transcription factors (TFs) is crucial to the study of gene expression regulation. Recently developed high-throughput experimental methods, including protein binding microarrays (PBM) and high-throughput SELEX (HT-SELEX), have enabled rapid measurements of the specificities for hundreds of TFs. However, few studies have developed efficient algorithms for estimating binding motifs based on HT-SELEX data. Also the simple method of constructing a position weight matrix (PWM) by comparing the frequency of the preferred sequence with single-nucleotide variants has the risk of generating motifs with higher information content than the true binding specificity. RESULTS: We developed an algorithm called BEESEM that builds on a comprehensive biophysical model of protein-DNA interactions, which is trained using the expectation maximization method. BEESEM is capable of selecting the optimal motif length and calculating the confidence intervals of estimated parameters. By comparing BEESEM with the published motifs estimated using the same HT-SELEX data, we demonstrate that BEESEM provides significant improvements. We also evaluate several motif discovery algorithms on independent PBM and ChIP-seq data. BEESEM provides significantly better fits to in vitro data, but its performance is similar to some other methods on in vivo data under the criterion of the area under the receiver operating characteristic curve (AUROC). This highlights the limitations of the purely rank-based AUROC criterion. Using quantitative binding data to assess models, however, demonstrates that BEESEM improves on prior models. AVAILABILITY AND IMPLEMENTATION: Freely available on the web at http://stormo.wustl.edu/resources.html . CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

The specificities of transcription factors are most commonly represented with probabilistic models. These models provide a probability for each base occurring at each position within the binding site and the positions are assumed to contribute independently. The model is simple and intuitive and is the basis for many motif discovery algorithms. However, the model also has inherent limitations that prevent it from accurately representing true binding probabilities, especially for the highest affinity sites under conditions of high protein concentration. The limitations are not due to the assumption of independence between positions but rather are caused by the non-linear relationship between binding affinity and binding probability and the fact that independent normalization at each position skews the site probabilities. Generally probabilistic models are reasonably good approximations, but new high-throughput methods allow for biophysical models with increased accuracy that should be used whenever possible.

MOTIVATION: Generating accurate transcription factor (TF) binding site motifs from data generated using the next-generation sequencing, especially ChIP-seq, is challenging. The challenge arises because a typical experiment reports a large number of sequences bound by a TF, and the length of each sequence is relatively long. Most traditional motif finders are slow in handling such enormous amount of data. To overcome this limitation, tools have been developed that compromise accuracy with speed by using heuristic discrete search strategies or limited optimization of identified seed motifs. However, such strategies may not fully use the information in input sequences to generate motifs. Such motifs often form good seeds and can be further improved with appropriate scoring functions and rapid optimization. RESULTS: We report a tool named discriminative motif optimizer (DiMO). DiMO takes a seed motif along with a positive and a negative database and improves the motif based on a discriminative strategy. We use area under receiver-operating characteristic curve (AUC) as a measure of discriminating power of motifs and a strategy based on perceptron training that maximizes AUC rapidly in a discriminative manner. Using DiMO, on a large test set of 87 TFs from human, drosophila and yeast, we show that it is possible to significantly improve motifs identified by nine motif finders. The motifs are generated/optimized using training sets and evaluated on test sets. The AUC is improved for almost 90% of the TFs on test sets and the magnitude of increase is up to 39%. AVAILABILITY AND IMPLEMENTATION: DiMO is available at http://stormo.wustl.edu/DiMO

MOTIVATION: Recognition models for protein-DNA interactions, which allow the prediction of specificity for a DNA-binding domain based only on its sequence or the alteration of specificity through rational design, have long been a goal of computational biology. There has been some progress in constructing useful models, especially for C(2)H(2) zinc finger proteins, but it remains a challenging problem with ample room for improvement. For most families of transcription factors the best available methods utilize k-nearest neighbor (KNN) algorithms to make specificity predictions based on the average of the specificities of the k most similar proteins with defined specificities. Homeodomain (HD) proteins are the second most abundant family of transcription factors, after zinc fingers, in most metazoan genomes, and as a consequence an effective recognition model for this family would facilitate predictive models of many transcriptional regulatory networks within these genomes. RESULTS: Using extensive experimental data, we have tested several machine learning approaches and find that both support vector machines and random forests (RFs) can produce recognition models for HD proteins that are significant improvements over KNN-based methods. Cross-validation analyses show that the resulting models are capable of predicting specificities with high accuracy. We have produced a web-based prediction tool, PreMoTF (Predicted Motifs for Transcription Factors) (http://stormo.wustl.edu/PreMoTF), for predicting position frequency matrices from protein sequence using a RF-based model.

MicroRNAs (miRNAs) are post-transcriptional regulators that bind to their target mRNAs through base complementarity. Predicting miRNA targets is a challenging task and various studies showed that existing algorithms suffer from high number of false predictions and low to moderate overlap in their predictions. Until recently, very few algorithms considered the dynamic nature of the interactions, including the effect of less specific interactions, the miRNA expression level, and the effect of combinatorial miRNA binding. Addressing these issues can result in a more accurate miRNA:mRNA modeling with many applications, including efficient miRNA-related SNP evaluation. We present a novel thermodynamic model based on the Fermi-Dirac equation that incorporates miRNA expression in the prediction of target occupancy and we show that it improves the performance of two popular single miRNA target finders. Modeling combinatorial miRNA targeting is a natural extension of this model. Two other algorithms show improved prediction efficiency when combinatorial binding models were considered. ComiR (Combinatorial miRNA targeting), a novel algorithm we developed, incorporates the improved predictions of the four target finders into a single probabilistic score using ensemble learning. Combining target scores of multiple miRNAs using ComiR improves predictions over the naïve method for target combination. ComiR scoring scheme can be used for identification of SNPs affecting miRNA binding. As proof of principle, ComiR identified rs17737058 as disruptive to the miR-488-5p:NCOA1 interaction, which we confirmed in vitro. We also found rs17737058 to be significantly associated with decreased bone mineral density (BMD) in two independent cohorts indicating that the miR-488-5p/NCOA1 regulatory axis is likely critical in maintaining BMD in women. With increasing availability of comprehensive high-throughput datasets from patients ComiR is expected to become an essential tool for miRNA-related studies.

Sequence similarity based protein clustering methods organize proteins into families of similar sequences, a task that continues to be critical for automated protein characterization. However, many protein families cannot be automatically characterized further because little is known about the function of any protein in a family of similar sequences. We present a novel phylogenetic profile comparison (PPC) method called Automated Protein Annotation by Coordinate Evolution (APACE) that facilitates the automated characterization of proteins beyond their homology to other similar sequences. Our method implements a new approach for the normalization of similarity scores among multiple species and automates the characterization of proteins by their patterns of co-evolution with other proteins that do not necessarily share a similar sequence. We demonstrate that our method is able to recapitulate the topology of the latest, unresolved, composite deep eukaryotic phylogeny and is able to quantify the as yet unresolved branch lengths. We further demonstrate that our method is able to detect more functionally related proteins, given the same starting data, than existing methods. Finally, we demonstrate that our method can be successfully applied to much larger comparative genomic problem instances where existing methods often fail.

MOTIVATION: Computational techniques for microbial genomic sequence analysis are becoming increasingly important. With next-generation sequencing technology and the human microbiome project underway, current sequencing capacity is significantly greater than the speed at which organisms of interest can be studied experimentally. Most related computational work has been focused on sequence assembly, gene annotation and metabolic network reconstruction. We have developed a method that will primarily use available sequence data in order to determine prokaryotic transcription factor (TF) binding specificities. RESULTS: Specificity determining residues (critical residues) were identified from crystal structures of DNA-protein complexes and TFs with the same critical residues were grouped into specificity classes. The putative binding regions for each class were defined as the set of promoters for each TF itself (autoregulatory) and the immediately upstream and downstream operons. MEME was used to find putative motifs within each separate class. Tests on the LacI and TetR TF families, using RegulonDB annotated sites, showed the sensitivity of prediction 86% and 80%, respectively. AVAILABILITY: http://ural.wustl.edu/∼gsahota/HTHmotif/

An increasing number of cis-regulatory RNA elements have been found to regulate gene expression post-transcriptionally in various biological processes in bacterial systems. Effective computational tools for large-scale identification of novel regulatory RNAs are strongly desired to facilitate our exploration of gene regulation mechanisms and regulatory networks. We present a new computational program named RSSVM (RNA Sampler+Support Vector Machine), which employs Support Vector Machines (SVMs) for efficient identification of functional RNA motifs from random RNA secondary structures. RSSVM uses a set of distinctive features to represent the common RNA secondary structure and structural alignment predicted by RNA Sampler, a tool for accurate common RNA secondary structure prediction, and is trained with functional RNAs from a variety of bacterial RNA motif/gene families covering a wide range of sequence identities. When tested on a large number of known and random RNA motifs, RSSVM shows a significantly higher sensitivity than other leading RNA identification programs while maintaining the same false positive rate. RSSVM performs particularly well on sets with low sequence identities. The combination of RNA Sampler and RSSVM provides a new, fast, and efficient pipeline for large-scale discovery of regulatory RNA motifs. We applied RSSVM to multiple Shewanella genomes and identified putative regulatory RNA motifs in the 5' untranslated regions (UTRs) in S. oneidensis, an important bacterial organism with extraordinary respiratory and metal reducing abilities and great potential for bioremediation and alternative energy generation. From 1002 sets of 5'-UTRs of orthologous operons, we identified 166 putative regulatory RNA motifs, including 17 of the 19 known RNA motifs from Rfam, an additional 21 RNA motifs that are supported by literature evidence, 72 RNA motifs overlapping predicted transcription terminators or attenuators, and other candidate regulatory RNA motifs. Our study provides a list of promising novel regulatory RNA motifs potentially involved in post-transcriptional gene regulation. Combined with the previous cis-regulatory DNA motif study in S. oneidensis, this genome-wide discovery of cis-regulatory RNA motifs may offer more comprehensive views of gene regulation at a different level in this organism. The RSSVM software, predictions, and analysis results on Shewanella genomes are available at http://ural.wustl.edu/resources.html#RSSVM.

We employ a biophysical model that accounts for the non-linear relationship between binding energy and the statistics of selected binding sites. The model includes the chemical potential of the transcription factor, non-specific binding affinity of the protein for DNA, as well as sequence-specific parameters that may include non-independent contributions of bases to the interaction. We obtain maximum likelihood estimates for all of the parameters and compare the results to standard probabilistic methods of parameter estimation. On simulated data, where the true energy model is known and samples are generated with a variety of parameter values, we show that our method returns much more accurate estimates of the true parameters and much better predictions of the selected binding site distributions. We also introduce a new high-throughput SELEX (HT-SELEX) procedure to determine the binding specificity of a transcription factor in which the initial randomized library and the selected sites are sequenced with next generation methods that return hundreds of thousands of sites. We show that after a single round of selection our method can estimate binding parameters that give very good fits to the selected site distributions, much better than standard motif identification algorithms.

MOTIVATION: Modeling and identifying the DNA-protein recognition code is one of the most challenging problems in computational biology. Several quantitative methods have been developed to model DNA-protein interactions with specific focus on the C(2)H(2) zinc-finger proteins, the largest transcription factor family in eukaryotic genomes. In many cases, they performed well. But the overall the predictive accuracy of these methods is still limited. One of the major reasons is all these methods used weight matrix models to represent DNA-protein interactions, assuming all base-amino acid contacts contribute independently to the total free energy of binding. RESULTS: We present a context-dependent model for DNA-zinc-finger protein interactions that allows us to identify inter-positional dependencies in the DNA recognition code for C(2)H(2) zinc-finger proteins. The degree of non-independence was detected by comparing the linear perceptron model with the non-linear neural net (NN) model for their predictions of DNA-zinc-finger protein interactions. This dependency is supported by the complex base-amino acid contacts observed in DNA-zinc-finger interactions from structural analyses. Using extensive published qualitative and quantitative experimental data, we demonstrated that the context-dependent model developed in this study can significantly improves predictions of DNA binding profiles and free energies of binding for both individual zinc fingers and proteins with multiple zinc fingers when comparing to previous positional-independent models. This approach can be extended to other protein families with complex base-amino acid residue interactions that would help to further understand the transcriptional regulation in eukaryotic genomes. AVAILABILITY: The software implemented as c programs and are available by request. http://ural.wustl.edu/softwares.html

MOTIVATION: Non-coding RNA genes and RNA structural regulatory motifs play important roles in gene regulation and other cellular functions. They are often characterized by specific secondary structures that are critical to their functions and are often conserved in phylogenetically or functionally related sequences. Predicting common RNA secondary structures in multiple unaligned sequences remains a challenge in bioinformatics research. METHODS AND RESULTS: We present a new sampling based algorithm to predict common RNA secondary structures in multiple unaligned sequences. Our algorithm finds the common structure between two sequences by probabilistically sampling aligned stems based on stem conservation calculated from intrasequence base pairing probabilities and intersequence base alignment probabilities. It iteratively updates these probabilities based on sampled structures and subsequently recalculates stem conservation using the updated probabilities. The iterative process terminates upon convergence of the sampled structures. We extend the algorithm to multiple sequences by a consistency-based method, which iteratively incorporates and reinforces consistent structure information from pairwise comparisons into consensus structures. The algorithm has no limitation on predicting pseudoknots. In extensive testing on real sequence data, our algorithm outperformed other leading RNA structure prediction methods in both sensitivity and specificity with a reasonably fast speed. It also generated better structural alignments than other programs in sequences of a wide range of identities, which more accurately represent the RNA secondary structure conservations. AVAILABILITY: The algorithm is implemented in a C program, RNA Sampler, which is available at http://ural.wustl.edu/software.html

MOTIVATION: Currently most of the methods for identifying differentially expressed genes fall into the category of so called single-gene-analysis, performing hypothesis testing on a gene-by-gene basis. In a single-gene-analysis approach, estimating the variability of each gene is required to determine whether a gene is differentially expressed or not. Poor accuracy of variability estimation makes it difficult to identify genes with small fold-changes unless a very large number of replicate experiments are performed. RESULTS: We propose a method that can avoid the difficult task of estimating variability for each gene, while reliably identifying a group of differentially expressed genes with low false discovery rates, even when the fold-changes are very small. In this article, a new characterization of differentially expressed genes is established based on a theorem about the distribution of ranks of genes sorted by (log) ratios within each array. This characterization of differentially expressed genes based on rank is an example of all-gene-analysis instead of single gene analysis. We apply the method to a cDNA microarray dataset and many low fold-changed genes (as low as 1.3 fold-changes) are reliably identified without carrying out hypothesis testing on a gene-by-gene basis. The false discovery rate is estimated in two different ways reflecting the variability from all the genes without the complications related to multiple hypothesis testing. We also provide some comparisons between our approach and single-gene-analysis based methods. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Computational gene prediction methods are an important component of whole genome analyses. While ab initio gene finders have demonstrated major improvements in accuracy, the most reliable methods are evidence-based gene predictors. These algorithms can rely on several different sources of evidence including predictions from multiple ab initio gene finders, matches to known proteins, sequence conservation and partial cDNAs to predict the final product. Despite the success of these algorithms, prediction of complete gene structures, especially for alternatively spliced products, remains a difficult task. RESULTS: LOCUS (Length Optimized Characterization of Unknown Spliceforms) is a new evidence-based gene finding algorithm which integrates a length-constraint into a dynamic programming-based framework for prediction of gene products. On a Caenorhabditis elegans test set of alternatively spliced internal exons, its performance exceeds that of current ab initio gene finders and in most cases can accurately predict the correct form of all the alternative products. As the length information used by the algorithm can be obtained in a high-throughput fashion, we propose that integration of such information into a gene-prediction pipeline is feasible and doing so may improve our ability to fully characterize the complete set of mRNAs for a genome. AVAILABILITY: LOCUS is available from http://ural.wustl.edu/software.html

BACKGROUND: The regulatory map of a genome consists of the binding sites for proteins that determine the transcription of nearby genes. An initial regulatory map for S. cerevisiae was recently published using six motif discovery programs to analyze genome-wide chromatin immunoprecipitation data for 203 transcription factors. The programs were used to identify sequence motifs that were likely to correspond to the DNA-binding specificity of the immunoprecipitated proteins. We report improved versions of two conservation-based motif discovery algorithms, PhyloCon and Converge. Using these programs, we create a refined regulatory map for S. cerevisiae by reanalyzing the same chromatin immunoprecipitation data. RESULTS: Applying the same conservative criteria that were applied in the original study, we find that PhyloCon and Converge each separately discover more known specificities than the combination of all six programs in the previous study. Combining the results of PhyloCon and Converge, we discover significant sequence motifs for 36 transcription factors that were previously missed. The new set of motifs identifies 636 more regulatory interactions than the previous one. The new network contains 28% more regulatory interactions among transcription factors, evidence of greater cross-talk between regulators. CONCLUSION: Combining two complementary computational strategies for conservation-based motif discovery improves the ability to identify the specificity of transcriptional regulators from genome-wide chromatin immunoprecipitation data. The increased sensitivity of these methods significantly expands the map of yeast regulatory sites without the need to alter any of the thresholds for statistical significance. The new map of regulatory sites reveals a more elaborate and complex view of the yeast genetic regulatory network than was observed previously.

MOTIVATION: Searching for non-coding RNA (ncRNA) genes and structural RNA elements (eleRNA) are major challenges in gene finding today as these often are conserved in structure rather than in sequence. Even though the number of available methods is growing, it is still of interest to pairwise detect two genes with low sequence similarity, where the genes are part of a larger genomic region. RESULTS: Here we present such an approach for pairwise local alignment which is based on foldalign and the Sankoff algorithm for simultaneous structural alignment of multiple sequences. We include the ability to conduct mutual scans of two sequences of arbitrary length while searching for common local structural motifs of some maximum length. This drastically reduces the complexity of the algorithm. The scoring scheme includes structural parameters corresponding to those available for free energy as well as for substitution matrices similar to RIBOSUM. The new foldalign implementation is tested on a dataset where the ncRNAs and eleRNAs have sequence similarity <40% and where the ncRNAs and eleRNAs are energetically indistinguishable from the surrounding genomic sequence context. The method is tested in two ways: (1) its ability to find the common structure between the genes only and (2) its ability to locate ncRNAs and eleRNAs in a genomic context. In case (1), it makes sense to compare with methods like Dynalign, and the performances are very similar, but foldalign is substantially faster. The structure prediction performance for a family is typically around 0.7 using Matthews correlation coefficient. In case (2), the algorithm is successful at locating RNA families with an average sensitivity of 0.8 and a positive predictive value of 0.9 using a BLAST-like hit selection scheme. AVAILABILITY: The program is available online at http://foldalign.kvl.dk/

UNLABELLED: Each organism has traits that are shared with some, but not all, organisms. Identification of genes needed for a particular trait can be accomplished by a comparative genomics approach using three or more organisms. Genes that occur in organisms without the trait are removed from the set of genes in common among organisms with the trait. To facilitate these comparisons, a web-based server, Procom, was developed to identify the subset of genes that may be needed for a trait. AVAILABILITY: The Procom program is freely available with documentation and examples at http://ural.wustl.edu/~billy/Procom/ CONTACT: [email protected].

BACKGROUND: Recognition codes for protein-DNA interactions typically assume that the interacting positions contribute additively to the binding energy. While this is known to not be precisely true, an additive model over the DNA positions can be a good approximation, at least for some proteins. Much less information is available about whether the protein positions contribute additively to the interaction. RESULTS: Using EGR zinc finger proteins, we measure the binding affinity of six different variants of the protein to each of six different variants of the consensus binding site. Both the protein and binding site variants include single and double mutations that allow us to assess how well additive models can account for the data. For each protein and DNA alone we find that additive models are good approximations, but over the combined set of data there are context effects that limit their accuracy. However, a small modification to the purely additive model, with only three additional parameters, improves the fit significantly. CONCLUSION: The additive model holds very well for every DNA site and every protein included in this study, but clear context dependence in the interactions was detected. A simple modification to the independent model provides a better fit to the complete data.

MOTIVATION: Is protein secondary structure primarily determined by local interactions between residues closely spaced along the amino acid backbone or by non-local tertiary interactions? To answer this question, we measure the entropy densities of primary and secondary structure sequences, and the local inter-sequence mutual information density. RESULTS: We find that the important inter-sequence interactions are short ranged, that correlations between neighboring amino acids are essentially uninformative and that only one-fourth of the total information needed to determine the secondary structure is available from local inter-sequence correlations. These observations support the view that the majority of most proteins fold via a cooperative process where secondary and tertiary structure form concurrently. Moreover, existing single-sequence secondary structure prediction algorithms are almost optimal, and we should not expect a dramatic improvement in prediction accuracy. AVAILABILITY: Both the data sets and analysis code are freely available from our Web site at http://compbio.berkeley.edu/

MOTIVATION: Pseudoknots have generally been excluded from the prediction of RNA secondary structures due to its difficulty in modeling. Although, several dynamic programming algorithms exist for the prediction of pseudoknots using thermodynamic approaches, they are neither reliable nor efficient. On the other hand, comparative methods are more reliable, but are often done in an ad hoc manner and require expert intervention. Maximum weighted matching, an algorithm for pseudoknot prediction with comparative analysis, suffers from low-prediction accuracy in many cases. RESULTS: Here we present an algorithm, iterated loop matching, for reliably and efficiently predicting RNA secondary structures including pseudoknots. The method can utilize either thermodynamic or comparative information or both, thus is able to predict pseudoknots for both aligned and individual sequences. We have tested the algorithm on a number of RNA families. Using 8-12 homologous sequences, the algorithm correctly identifies more than 90% of base-pairs for short sequences and 80% overall. It correctly predicts nearly all pseudoknots and produces very few spurious base-pairs for sequences without pseudoknots. Comparisons show that our algorithm is both more sensitive and more specific than the maximum weighted matching method. In addition, our algorithm has high-prediction accuracy on individual sequences, comparable with the PKNOTS algorithm, while using much less computational resources. AVAILABILITY: The program has been implemented in ANSI C and is freely available for academic use at http://www.cse.wustl.edu/~zhang/projects/rna/ilm/ SUPPLEMENTARY INFORMATION: http://www.cse.wustl.edu/~zhang/projects/rna/ilm/

In 2002-2003, the American College of Medical Informatics (ACMI) undertook a study of the future of informatics training. This project capitalized on the rapidly expanding interest in the role of computation in basic biological research, well characterized in the National Institutes of Health (NIH) Biomedical Information Science and Technology Initiative (BISTI) report. The defining activity of the project was the three-day 2002 Annual Symposium of the College. A committee, comprised of the authors of this report, subsequently carried out activities, including interviews with a broader informatics and biological sciences constituency, collation and categorization of observations, and generation of recommendations. The committee viewed biomedical informatics as an interdisciplinary field, combining basic informational and computational sciences with application domains, including health care, biological research, and education. Consequently, effective training in informatics, viewed from a national perspective, should encompass four key elements: (1). curricula that integrate experiences in the computational sciences and application domains rather than just concatenating them; (2). diversity among trainees, with individualized, interdisciplinary cross-training allowing each trainee to develop key competencies that he or she does not initially possess; (3). direct immersion in research and development activities; and (4). exposure across the wide range of basic informational and computational sciences. Informatics training programs that implement these features, irrespective of their funding sources, will meet and exceed the challenges raised by the BISTI report, and optimally prepare their trainees for careers in a field that continues to evolve.

MOTIVATION: Discovery of regulatory motifs in unaligned DNA sequences remains a fundamental problem in computational biology. Two categories of algorithms have been developed to identify common motifs from a set of DNA sequences. The first can be called a 'multiple genes, single species' approach. It proposes that a degenerate motif is embedded in some or all of the otherwise unrelated input sequences and tries to describe a consensus motif and identify its occurrences. It is often used for co-regulated genes identified through experimental approaches. The second approach can be called 'single gene, multiple species'. It requires orthologous input sequences and tries to identify unusually well conserved regions by phylogenetic footprinting. Both approaches perform well, but each has some limitations. It is tempting to combine the knowledge of co-regulation among different genes and conservation among orthologous genes to improve our ability to identify motifs. RESULTS: Based on the Consensus algorithm previously established by our group, we introduce a new algorithm called PhyloCon (Phylogenetic Consensus) that takes into account both conservation among orthologous genes and co-regulation of genes within a species. This algorithm first aligns conserved regions of orthologous sequences into multiple sequence alignments, or profiles, then compares profiles representing non-orthologous sequences. Motifs emerge as common regions in these profiles. Here we present a novel statistic to compare profiles of DNA sequences and a greedy approach to search for common subprofiles. We demonstrate that PhyloCon performs well on both synthetic and biological data. AVAILABILITY: Software available upon request from the authors. http://ural.wustl.edu/softwares.html

Mark Borodovsky, Eugene Koonin, Chris Burge, Jim Fickett, John Logsdon, Andrej Sali, Gary Stormo, Igor Zhulin; THE THIRD GEORGIA TECH–EMORY INTERNATIONAL C

MOTIVATION: High density DNA oligo microarrays are widely used in biomedical research. Selection of optimal DNA oligos that are deposited on the microarrays is critical. Based on sequence information and hybridization free energy, we developed a new algorithm to select optimal short (20-25 bases) or long (50 or 70 bases) oligos from genes or open reading frames (ORFs) and predict their hybridization behavior. Having optimized probes for each gene is valuable for two reasons. By minimizing background hybridization they provide more accurate determinations of true expression levels. Having optimum probes minimizes the number of probes needed per gene, thereby decreasing the cost of each microarray, raising the number of genes on each chip and increasing its usage. RESULTS: In this paper we describe algorithms to optimize the selection of specific probes for each gene in an entire genome. The criteria for truly optimum probes are easily stated but they are not computable at all levels currently. We have developed an heuristic approach that is efficiently computable at all levels and should provide a good approximation to the true optimum set. We have run the program on the complete genomes for several model organisms and deposited the results in a database that is available on-line (http://ural.wustl.edu/~lif/probe.pl). AVAILABILITY: The program is available upon request.

MOTIVATION: Transcriptional activation in eukaryotic organisms normally requires combinatorial interactions of multiple transcription factors. Though several methods exist for identification of individual protein binding site patterns in DNA sequences, there are few methods for discovery of binding site patterns for cooperatively acting factors. Here we present an algorithm, Co-Bind (for COperative BINDing), for discovering DNA target sites for cooperatively acting transcription factors. The method utilizes a Gibbs sampling strategy to model the cooperativity between two transcription factors and defines position weight matrices for the binding sites. Sequences from both the training set and the entire genome are taken into account, in order to discriminate against commonly occurring patterns in the genome, and produce patterns which are significant only in the training set. RESULTS: We have tested Co-Bind on semi-synthetic and real data sets to show it can efficiently identify DNA target site patterns for cooperatively binding transcription factors. In cases where binding site patterns are weak and cannot be identified by other available methods, Co-Bind, by virtue of modeling the cooperativity between factors, can identify those sites efficiently. Though developed to model protein-DNA interactions, the scope of Co-Bind may be extended to combinatorial, sequence specific, interactions in other macromolecules. AVAILABILITY: The program is available upon request from the authors or may be downloaded from http://ural.wustl.edu.

High-density DNA oligonucleotide microarrays are widely used in biomedical research. In this paper, we describe algorithms to optimize the selection of specific probes for each gene in an entire genome. Having optimized probes for each gene is valuable for two reasons: (1) by minimizing background hybridization, they provide more accurate determinations of true expression levels, and (2) having optimum probes eliminates the need for multiple probes per gene, as is usually done now, thereby decreasing the cost of each microarray and increasing their usage. The criteria for truly optimum probes is easily stated, but they are not computable at present. We have developed a heuristic approach that is efficiently computable and should provide a good approximation to the true optimum set. We have run the program on the complete genomes for several model organisms and deposited the results in a database that is available online ().

MOTIVATION: Methods that predict the structure of molecules by looking for statistical correlation have been quite effective. Unfortunately, these methods often disregard phylogenetic information in the sequences they analyze. Here, we present a number of statistics for RNA molecular-structure prediction. Besides common pair-wise comparisons, we consider a few reasonable statistics for base-triple predictions, and present an elaborate analysis of these methods. All these statistics incorporate phylogenetic relationships of the sequences in the analysis to varying degrees, and the different nature of these tests gives a wide choice of statistical tools for RNA structure prediction. RESULTS: Starting from statistics that incorporate phylogenetic information only as independent sequence evolution models for each position of a multiple alignment, and extending this idea to a joint evolution model of two positions, we enhance the usual purely statistical methods (e.g. methods based on the Mutual Information statistic) with the use of phylogenetic information available in the sequences. In particular, we present a joint model based on the HKY evolution model, and consequently a X(2) test of independence for two positions. A significant part of this work is devoted to some mathematical analysis of these methods. We tested these statistics on regions of 16S and 23S rRNA, and tRNA.

The purpose of this article is to provide a brief history of the development and application of computer algorithms for the analysis and prediction of DNA binding sites. This problem can be conveniently divided into two subproblems. The first is, given a collection of known binding sites, develop a representation of those sites that can be used to search new sequences and reliably predict where additional binding sites occur. The second is, given a set of sequences known to contain binding sites for a common factor, but not knowing where the sites are, discover the location of the sites in each sequence and a representation for the specificity of the protein.

MOTIVATION: Molecular biologists frequently can obtain interesting insight by aligning a set of related DNA, RNA or protein sequences. Such alignments can be used to determine either evolutionary or functional relationships. Our interest is in identifying functional relationships. Unless the sequences are very similar, it is necessary to have a specific strategy for measuring-or scoring-the relatedness of the aligned sequences. If the alignment is not known, one can be determined by finding an alignment that optimizes the scoring scheme. RESULTS: We describe four components to our approach for determining alignments of multiple sequences. First, we review a log-likelihood scoring scheme we call information content. Second, we describe two methods for estimating the P value of an individual information content score: (i) a method that combines a technique from large-deviation statistics with numerical calculations; (ii) a method that is exclusively numerical. Third, we describe how we count the number of possible alignments given the overall amount of sequence data. This count is multiplied by the P value to determine the expected frequency of an information content score and, thus, the statistical significance of the corresponding alignment. Statistical significance can be used to compare alignments having differing widths and containing differing numbers of sequences. Fourth, we describe a greedy algorithm for determining alignments of functionally related sequences. Finally, we test the accuracy of our P value calculations, and give an example of using our algorithm to identify binding sites for the Escherichia coli CRP protein. AVAILABILITY: Programs were developed under the UNIX operating system and are available by anonymous ftp from ftp://beagle.colorado.edu/pub/consensus.

MOTIVATION: To provide a graphical interface for the generation, display and manipulation of a sequence landscape that will run on all X-windows-based Unix workstations. RESULTS: The sequence landscape approach enables the representation of the frequency of occurrence of all query sequence sub-words within a database. The landscape approach can detect tandem and other repeating word motifs, specific sub-words that are over-represented words in a particular database using Markov probability and the preference for sub-words belonging to either one of two databases. All these features aid in the classification of a query sequence. Given the open-text format for sequences and databases, the Xlandscape tool can be applied to a wide range of problems.

MOTIVATION: A large number of new DNA sequences with virtually unknown functions are generated as the Human Genome Project progresses. Therefore, it is essential to develop computer algorithms that can predict the functionality of DNA segments according to their primary sequences, including algorithms that can predict promoters. Although several promoter-predicting algorithms are available, they have high false-positive detections and the rate of promoter detection needs to be improved further. RESULTS: In this research, PromFD, a computer program to recognize vertebrate RNA polymerase II promoters, has been developed. Both vertebrate promoters and non-promoter sequences are used in the analysis. The promoters are obtained from the Eukaryotic Promoter Database. Promoters are divided into a training set and a test set. Non-promoter sequences are obtained from the GenBank sequence databank, and are also divided into a training set and a test set. The first step is to search out, among all possible permutations, patterns of strings 5-10 bp long, that are significantly over-represented in the promoter set. The program also searches IMD (Information Matrix Database) matrices that have a significantly higher presence in the promoter set. The results of the searches are stored in the PromFD database, and the program PromFD scores input DNA sequences according to their content of the database entries. PromFD predicts promoters-their locations and the location of potential TATA boxes, if found. The program can detect 71% of promoters in the training set with a false-positive rate of under 1 in every 13,000 bp, and 47% of promoters in the test set with a false-positive rate of under 1 in every 9800 bp. PromFD uses a new approach and its false-positive identification rate is better compared with other available promoter recognition algorithms. The source code for PromFD is in the 'c+2' language.

MOTIVATION: We extend the standard 'Sequence Logo' method of Schneider and Stevens (Nucleic Acids Res., 18, 6097-6100, 1990) to incorporate prior frequencies on the bases, allow for gaps in the alignments, and indicate the mutual information of base-paired regions in RNA. RESULTS: Given an alignment of RNA sequences with the base pairings indicated, the program will calculate the information at each position, including the mutual information of the base pairs, and display the results in a 'Structure Logo'. Alignments without base pairing can also be displayed in a 'Sequence Logo', but still allowing gaps and incorporating prior frequencies if desired. AVAILABILITY: The code is available from, and an Internet server can be used to run the program at, http://www.cbs.dtu.dk/gorodkin/appl/slogo. html.

The information matrix database (IMD), a database of weight matrices of transcription factor binding sites, is developed. MATRIX SEARCH, a program which can find potential transcription factor binding sites in DNA sequences using the IMD database, is also developed and accompanies the IMD database. MATRIX SEARCH adopts a user interface very similar to that of the SIGNAL SCAN program. MATRIX SEARCH allows the user to search an input sequence with the IMD automatically, to visualize the matrix representations of sites for particular factors, and to retrieve journal citations. The source code for MATRIX SEARCH is in the 'C' language, and the program is available for unix platforms.

SIGNAL SCAN is a program that utilizes a transcription factor database to find potential transcription factor binding sites in DNA sequences. The program is now in its third version. The SIGNAL SCAN transcription factor database format has changed and the program output format has been improved. New features allow the user to update the SIGNAL SCAN database automatically, to retrieve original journal citations and to develop user signal databases. The program now uses an indexing algorithm, improving scanning speed by a factor of 3. SIGNAL SCAN is now network compatible and is available for IBM-compatible PC, Unix and VMS platforms.

We have developed a method for identifying consensus patterns in a set of unaligned DNA sequences known to bind a common protein or to have some other common biochemical function. The method is based on a matrix representation of binding site patterns. Each row of the matrix represents one of the four possible bases, each column represents one of the positions of the binding site and each element is determined by the frequency the indicated base occurs at the indicated position. The goal of the method is to find the most significant matrix--i.e. the one with the lowest probability of occurring by chance--out of all the matrices that can be formed from the set of related sequences. The reliability of the method improves with the number of sequences, while the time required increases only linearly with the number of sequences. To test this method, we analysed 11 DNA sequences containing promoters regulated by the Escherichia coli LexA protein. The matrices we found were consistent with the known consensus sequence, and could distinguish the generally accepted LexA binding sites from other DNA sequences.