SUMMARY: GSEL is a computational framework for calculating the enrichment of signatures of diverse evolutionary forces in a set of genomic regions. GSEL can flexibly integrate any sequence-based evolutionary metric and analyze sets of human genomic regions identified by genome-wide assays (e.g. GWAS, eQTL, *-seq). The core of GSEL's approach is the generation of empirical null distributions tailored to the allele frequency and linkage disequilibrium structure of the regions of interest. We illustrate the application of GSEL to variants identified from a GWAS of body mass index, a highly polygenic trait. AVAILABILITY AND IMPLEMENTATION: GSEL is implemented as a fast, flexible and user-friendly python package. It is available with demonstration data at https://github.com/abraham-abin13/gsel_vec. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Deep neural networks (DNNs) have achieved state-of-the-art performance in identifying gene regulatory sequences, but they have provided limited insight into the biology of regulatory elements due to the difficulty of interpreting the complex features they learn. Several models of how combinatorial binding of transcription factors, i.e. the regulatory grammar, drives enhancer activity have been proposed, ranging from the flexible TF billboard model to the stringent enhanceosome model. However, there is limited knowledge of the prevalence of these (or other) sequence architectures across enhancers. Here we perform several hypothesis-driven analyses to explore the ability of DNNs to learn the regulatory grammar of enhancers. We created synthetic datasets based on existing hypotheses about combinatorial transcription factor binding site (TFBS) patterns, including homotypic clusters, heterotypic clusters, and enhanceosomes, from real TF binding motifs from diverse TF families. We then trained deep residual neural networks (ResNets) to model the sequences under a range of scenarios that reflect real-world multi-label regulatory sequence prediction tasks. We developed a gradient-based unsupervised clustering method to extract the patterns learned by the ResNet models. We demonstrated that simulated regulatory grammars are best learned in the penultimate layer of the ResNets, and the proposed method can accurately retrieve the regulatory grammar even when there is heterogeneity in the enhancer categories and a large fraction of TFBS outside of the regulatory grammar. However, we also identify common scenarios where ResNets fail to learn simulated regulatory grammars. Finally, we applied the proposed method to mouse developmental enhancers and were able to identify the components of a known heterotypic TF cluster. Our results provide a framework for interpreting the regulatory rules learned by ResNets, and they demonstrate that the ability and efficiency of ResNets in learning the regulatory grammar depends on the nature of the prediction task.

Predicting mutation-induced changes in protein thermodynamic stability (ΔΔG) is of great interest in protein engineering, variant interpretation, and protein biophysics. We introduce ThermoNet, a deep, 3D-convolutional neural network (3D-CNN) designed for structure-based prediction of ΔΔGs upon point mutation. To leverage the image-processing power inherent in CNNs, we treat protein structures as if they were multi-channel 3D images. In particular, the inputs to ThermoNet are uniformly constructed as multi-channel voxel grids based on biophysical properties derived from raw atom coordinates. We train and evaluate ThermoNet with a curated data set that accounts for protein homology and is balanced with direct and reverse mutations; this provides a framework for addressing biases that have likely influenced many previous ΔΔG prediction methods. ThermoNet demonstrates performance comparable to the best available methods on the widely used Ssym test set. In addition, ThermoNet accurately predicts the effects of both stabilizing and destabilizing mutations, while most other methods exhibit a strong bias towards predicting destabilization. We further show that homology between Ssym and widely used training sets like S2648 and VariBench has likely led to overestimated performance in previous studies. Finally, we demonstrate the practical utility of ThermoNet in predicting the ΔΔGs for two clinically relevant proteins, p53 and myoglobin, and for pathogenic and benign missense variants from ClinVar. Overall, our results suggest that 3D-CNNs can model the complex, non-linear interactions perturbed by mutations, directly from biophysical properties of atoms.

BACKGROUND: Next-generation sequencing of individuals with genetic diseases often detects candidate rare variants in numerous genes, but determining which are causal remains challenging. We hypothesized that the spatial distribution of missense variants in protein structures contains information about function and pathogenicity that can help prioritize variants of unknown significance (VUS) and elucidate the structural mechanisms leading to disease. RESULTS: To illustrate this approach in a clinical application, we analyzed 13 candidate missense variants in regulator of telomere elongation helicase 1 (RTEL1) identified in patients with Familial Interstitial Pneumonia (FIP). We curated pathogenic and neutral RTEL1 variants from the literature and public databases. We then used homology modeling to construct a 3D structural model of RTEL1 and mapped known variants into this structure. We next developed a pathogenicity prediction algorithm based on proximity to known disease causing and neutral variants and evaluated its performance with leave-one-out cross-validation. We further validated our predictions with segregation analyses, telomere lengths, and mutagenesis data from the homologous XPD protein. Our algorithm for classifying RTEL1 VUS based on spatial proximity to pathogenic and neutral variation accurately distinguished 7 known pathogenic from 29 neutral variants (ROC AUC = 0.85) in the N-terminal domains of RTEL1. Pathogenic proximity scores were also significantly correlated with effects on ATPase activity (Pearson r = -0.65, p = 0.0004) in XPD, a related helicase. Applying the algorithm to 13 VUS identified from sequencing of RTEL1 from patients predicted five out of six disease-segregating VUS to be pathogenic. We provide structural hypotheses regarding how these mutations may disrupt RTEL1 ATPase and helicase function. CONCLUSIONS: Spatial analysis of missense variation accurately classified candidate VUS in RTEL1 and suggests how such variants cause disease. Incorporating spatial proximity analyses into other pathogenicity prediction tools may improve accuracy for other genes and genetic diseases.

Genomic regions with gene regulatory enhancer activity turnover rapidly across mammals. In contrast, gene expression patterns and transcription factor binding preferences are largely conserved between mammalian species. Based on this conservation, we hypothesized that enhancers active in different mammals would exhibit conserved sequence patterns in spite of their different genomic locations. To investigate this hypothesis, we evaluated the extent to which sequence patterns that are predictive of enhancers in one species are predictive of enhancers in other mammalian species by training and testing two types of machine learning models. We trained support vector machine (SVM) and convolutional neural network (CNN) classifiers to distinguish enhancers defined by histone marks from the genomic background based on DNA sequence patterns in human, macaque, mouse, dog, cow, and opossum. The classifiers accurately identified many adult liver, developing limb, and developing brain enhancers, and the CNNs outperformed the SVMs. Furthermore, classifiers trained in one species and tested in another performed nearly as well as classifiers trained and tested on the same species. We observed similar cross-species conservation when applying the models to human and mouse enhancers validated in transgenic assays. This indicates that many short sequence patterns predictive of enhancers are largely conserved. The sequence patterns most predictive of enhancers in each species matched the binding motifs for a common set of TFs enriched for expression in relevant tissues, supporting the biological relevance of the learned features. Thus, despite the rapid change of active enhancer locations between mammals, cross-species enhancer prediction is often possible. Our results suggest that short sequence patterns encoding enhancer activity have been maintained across more than 180 million years of mammalian evolution.

MOTIVATION: DNA and protein patterns are usefully represented by sequence logos. However, the methods for logo generation in common use lack a proper statistical basis, and are non-optimal for recognizing functionally relevant alignment columns. RESULTS: We redefine the information at a logo position as a per-observation multiple alignment log-odds score. Such scores are positive or negative, depending on whether a column's observations are better explained as arising from relatedness or chance. Within this framework, we propose distinct normalized maximum likelihood and Bayesian measures of column information. We illustrate these measures on High Mobility Group B (HMGB) box proteins and a dataset of enzyme alignments. Particularly in the context of protein alignments, our measures improve the discrimination of biologically relevant positions. AVAILABILITY AND IMPLEMENTATION: Our new measures are implemented in an open-source Web-based logo generation program, which is available at http://www.ncbi.nlm.nih.gov/CBBresearch/Yu/logoddslogo/index.html. A stand-alone version of the program is also available from this site. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Methylation of CpG dinucleotides is a prevalent epigenetic modification that is required for proper development in vertebrates. Genome-wide DNA methylation assays have become increasingly common, and this has enabled characterization of DNA methylation in distinct stages across differentiating cellular lineages. Changes in CpG methylation are essential to cellular differentiation; however, current methods for modeling methylation dynamics do not account for the dependency structure between precursor and dependent cell types. RESULTS: We developed a continuous-time Markov chain approach, based on the observation that changes in methylation state over tissue differentiation can be modeled similarly to DNA nucleotide changes over evolutionary time. This model explicitly takes precursor to descendant relationships into account and enables inference of CpG methylation dynamics. To illustrate our method, we analyzed a high-resolution methylation map of the differentiation of mouse stem cells into several blood cell types. Our model can successfully infer unobserved CpG methylation states from observations at the same sites in related cell types (90% correct), and this approach more accurately reconstructs missing data than imputation based on neighboring CpGs (84% correct). Additionally, the single CpG resolution of our methylation dynamics estimates enabled us to show that DNA sequence context of CpG sites is informative about methylation dynamics across tissue differentiation. Finally, we identified genomic regions with clusters of highly dynamic CpGs and present a likely functional example. Our work establishes a framework for inference and modeling that is well suited to DNA methylation data, and our success suggests that other methods for analyzing DNA nucleotide substitutions will also translate to the modeling of epigenetic phenomena. AVAILABILITY AND IMPLEMENTATION: Source code is available at www.kostkalab.net/software.

Gene-regulatory enhancers have been identified using various approaches, including evolutionary conservation, regulatory protein binding, chromatin modifications, and DNA sequence motifs. To integrate these different approaches, we developed EnhancerFinder, a two-step method for distinguishing developmental enhancers from the genomic background and then predicting their tissue specificity. EnhancerFinder uses a multiple kernel learning approach to integrate DNA sequence motifs, evolutionary patterns, and diverse functional genomics datasets from a variety of cell types. In contrast with prediction approaches that define enhancers based on histone marks or p300 sites from a single cell line, we trained EnhancerFinder on hundreds of experimentally verified human developmental enhancers from the VISTA Enhancer Browser. We comprehensively evaluated EnhancerFinder using cross validation and found that our integrative method improves the identification of enhancers over approaches that consider a single type of data, such as sequence motifs, evolutionary conservation, or the binding of enhancer-associated proteins. We find that VISTA enhancers active in embryonic heart are easier to identify than enhancers active in several other embryonic tissues, likely due to their uniquely high GC content. We applied EnhancerFinder to the entire human genome and predicted 84,301 developmental enhancers and their tissue specificity. These predictions provide specific functional annotations for large amounts of human non-coding DNA, and are significantly enriched near genes with annotated roles in their predicted tissues and lead SNPs from genome-wide association studies. We demonstrate the utility of EnhancerFinder predictions through in vivo validation of novel embryonic gene regulatory enhancers from three developmental transcription factor loci. Our genome-wide developmental enhancer predictions are freely available as a UCSC Genome Browser track, which we hope will enable researchers to further investigate questions in developmental biology.

The evolutionary history of a protein reflects the functional history of its ancestors. Recent phylogenetic studies identified distinct evolutionary signatures that characterize proteins involved in cancer, Mendelian disease, and different ontogenic stages. Despite the potential to yield insight into the cellular functions and interactions of proteins, such comparative phylogenetic analyses are rarely performed, because they require custom algorithms. We developed ProteinHistorian to make tools for performing analyses of protein origins widely available. Given a list of proteins of interest, ProteinHistorian estimates the phylogenetic age of each protein, quantifies enrichment for proteins of specific ages, and compares variation in protein age with other protein attributes. ProteinHistorian allows flexibility in the definition of protein age by including several algorithms for estimating ages from different databases of evolutionary relationships. We illustrate the use of ProteinHistorian with three example analyses. First, we demonstrate that proteins with high expression in human, compared to chimpanzee and rhesus macaque, are significantly younger than those with human-specific low expression. Next, we show that human proteins with annotated regulatory functions are significantly younger than proteins with catalytic functions. Finally, we compare protein length and age in many eukaryotic species and, as expected from previous studies, find a positive, though often weak, correlation between protein age and length. ProteinHistorian is available through a web server with an intuitive interface and as a set of command line tools; this allows biologists and bioinformaticians alike to integrate these approaches into their analysis pipelines. ProteinHistorian's modular, extensible design facilitates the integration of new datasets and algorithms. The ProteinHistorian web server, source code, and pre-computed ages for 32 eukaryotic genomes are freely available under the GNU public license at http://lighthouse.ucsf.edu/ProteinHistorian/.

The DBA/2J mouse is not only the oldest inbred strain, but also one of the most widely used strains. DBA/2J exhibits many unique anatomical, physiological, and behavior traits. In addition, DBA/2J is one parent of the large BXD family of recombinant inbred strains [ 1 ]. The genome of the other parent of this BXD family—C57BL/6J—has been sequenced and serves as the mouse reference genome [ 2 ]. We sequenced the genome of DBA/2J using SOLiD and Illumina high throughput short read protocols to generate a comprehensive set of ~5 million sequence variants segregating in the BXD family that ultimately cause developmental, anatomical, functional and behavioral differences among these 80+ strains. We generated approximately 13.2 and 38.9× whole-genome short reads of DBA/2J females using Illumina GA2 and ABI SOLiD massively parallel DNA sequencing platforms. Comparing to the C57BL/6J reference genome sequence, we identified over 4.5 million single nucleotide polymorphisms (SNPs), including 84 nonsense and ~11,000 missense mutations, 78% of which are novel. We also detected ~568,000 insertions and deletions (indels) within single short reads and ~9,400 between mate-paired reads. Approximately 300 inversions were detected by SOLiD mate-pair reads, 46 of which span at least one exon. In addition, we identified ~22,000 copy number variants (CNVs) in the range of 1 Kb to 100 Kb (Figure 1 ). Concentric circles represent the sequence and structural variation across mouse chromosomes. Moving inward from the outer circle, circle 1 denotes each chromosome. Circle 2, read depth with 100kb window. Circle 3, SNP density with 100kb windows (black is lowest density and orange is highest density). Circle 4, Indels density with 100kb window. Circle 4, Inversion. Circle 5, CNVs, blue (outward) denotes loss of CNVs and green (inward) denotes gains of CNVs. Our study generates the first consensus sequence for the DBA/2J and creates a compendium of sequence and structural variations that will be used by the community of researchers who study complex traits in mouse models. The sequence data provide a novel resource with which to initiate reverse genetic analysis of complex traits, particularly by exploiting strong alleles (premature stop codons, frame-shift mutations, and deletion) that differentially affect members of the BXD strain family. The DBA/2J genome is also an essential prerequisite to unbiased alignment of RNA-seq and ChIP-seq data generated using BXD strains and any other cross involving these two common parental strains.

G-quadruplex DNA is a four-stranded DNA structure formed by non-Watson-Crick base pairing between stacked sets of four guanines. Many possible functions have been proposed for this structure, but its in vivo role in the cell is still largely unresolved. We carried out a genome-wide survey of the evolutionary conservation of regions with the potential to form G-quadruplex DNA structures (G4 DNA motifs) across seven yeast species. We found that G4 DNA motifs were significantly more conserved than expected by chance, and the nucleotide-level conservation patterns suggested that the motif conservation was the result of the formation of G4 DNA structures. We characterized the association of conserved and non-conserved G4 DNA motifs in Saccharomyces cerevisiae with more than 40 known genome features and gene classes. Our comprehensive, integrated evolutionary and functional analysis confirmed the previously observed associations of G4 DNA motifs with promoter regions and the rDNA, and it identified several previously unrecognized associations of G4 DNA motifs with genomic features, such as mitotic and meiotic double-strand break sites (DSBs). Conserved G4 DNA motifs maintained strong associations with promoters and the rDNA, but not with DSBs. We also performed the first analysis of G4 DNA motifs in the mitochondria, and surprisingly found a tenfold higher concentration of the motifs in the AT-rich yeast mitochondrial DNA than in nuclear DNA. The evolutionary conservation of the G4 DNA motif and its association with specific genome features supports the hypothesis that G4 DNA has in vivo functions that are under evolutionary constraint.

Identifying a protein's functional sites is an important step towards characterizing its molecular function. Numerous structure- and sequence-based methods have been developed for this problem. Here we introduce ConCavity, a small molecule binding site prediction algorithm that integrates evolutionary sequence conservation estimates with structure-based methods for identifying protein surface cavities. In large-scale testing on a diverse set of single- and multi-chain protein structures, we show that ConCavity substantially outperforms existing methods for identifying both 3D ligand binding pockets and individual ligand binding residues. As part of our testing, we perform one of the first direct comparisons of conservation-based and structure-based methods. We find that the two approaches provide largely complementary information, which can be combined to improve upon either approach alone. We also demonstrate that ConCavity has state-of-the-art performance in predicting catalytic sites and drug binding pockets. Overall, the algorithms and analysis presented here significantly improve our ability to identify ligand binding sites and further advance our understanding of the relationship between evolutionary sequence conservation and structural and functional attributes of proteins. Data, source code, and prediction visualizations are available on the ConCavity web site (http://compbio.cs.princeton.edu/concavity/).

MOTIVATION: Within a homologous protein family, proteins may be grouped into subtypes that share specific functions that are not common to the entire family. Often, the amino acids present in a small number of sequence positions determine each protein's particular functional specificity. Knowledge of these specificity determining positions (SDPs) aids in protein function prediction, drug design and experimental analysis. A number of sequence-based computational methods have been introduced for identifying SDPs; however, their further development and evaluation have been hindered by the limited number of known experimentally determined SDPs. RESULTS: We combine several bioinformatics resources to automate a process, typically undertaken manually, to build a dataset of SDPs. The resulting large dataset, which consists of SDPs in enzymes, enables us to characterize SDPs in terms of their physicochemical and evolutionary properties. It also facilitates the large-scale evaluation of sequence-based SDP prediction methods. We present a simple sequence-based SDP prediction method, GroupSim, and show that, surprisingly, it is competitive with a representative set of current methods. We also describe ConsWin, a heuristic that considers sequence conservation of neighboring amino acids, and demonstrate that it improves the performance of all methods tested on our large dataset of enzyme SDPs. AVAILABILITY: Datasets and GroupSim code are available online at http://compbio.cs.princeton.edu/specificity/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: All residues in a protein are not equally important. Some are essential for the proper structure and function of the protein, whereas others can be readily replaced. Conservation analysis is one of the most widely used methods for predicting these functionally important residues in protein sequences. RESULTS: We introduce an information-theoretic approach for estimating sequence conservation based on Jensen-Shannon divergence. We also develop a general heuristic that considers the estimated conservation of sequentially neighboring sites. In large-scale testing, we demonstrate that our combined approach outperforms previous conservation-based measures in identifying functionally important residues; in particular, it is significantly better than the commonly used Shannon entropy measure. We find that considering conservation at sequential neighbors improves the performance of all methods tested. Our analysis also reveals that many existing methods that attempt to incorporate the relationships between amino acids do not lead to better identification of functionally important sites. Finally, we find that while conservation is highly predictive in identifying catalytic sites and residues near bound ligands, it is much less effective in identifying residues in protein-protein interfaces. AVAILABILITY: Data sets and code for all conservation measures evaluated are available at http://compbio.cs.princeton.edu/conservation/