Search

The GeneMANIA Cytoscape app enables users to construct a composite gene-gene functional interaction network from a gene list. The resulting network includes the genes most related to the original list, and functional annotations from Gene Ontology. The edges are annotated with details about the publication or data source the interactions were derived from. The app leverages GeneMANIA’s database of 1800+ networks, containing over 500 million interactions spanning 8 organisms: A. thaliana, C. elegans, D. melanogaster, D. rerio, H. sapiens, M. musculus, R. norvegicus, and S. cerevisiae. Users may also import their own organisms, networks, and expression profiles. The app is compatible with Cytoscape versions 2 and 3.

RNA binding proteins (RBPs) are key players in the regulation of gene expression. In this chapter we discuss the main protein-RNA recognition modes used by RBPs in order to regulate multiple steps of RNA processing. We discuss traditional and state-of-the-art technologies that can be used to study RNAs bound by individual RBPs, or vice versa, for both in vitro and in vivo methodologies. To help highlight the biological significance of RBP mediated regulation, online resources on experimentally verified protein-RNA interactions are briefly presented. Finally, we present the major tools to computationally infer RNA binding sites according to the modeling features and to the unsupervised or supervised frameworks that are adopted. Since some RNA binding site search algorithms are derived from DNA binding site search algorithms, we discuss the commonalities and novelties introduced to handle both sequence and structural features uniquely characterizing protein-RNA interactions.

Networks of gene-gene interactions (or, functional interactions or more generally, associations) have proven very useful for predicting gene function. Association networks have proven useful in other biological domains to predict properties of nodes representing, e.g., patients, based on their connectivity with other nodes with pre-established properties. The quality of these predictions depends on the quality and relevance of the association data. For predicting gene function, there are hundreds of different networks that can be used and a plethora of different algorithms to use them—validating prediction performance can be time consuming and error prone. Here we describe methodology and software to automatically evaluate the contribution of an individual association network to predicting gene function (and more generally, predicting node function). This software is implemented in Network Assessor, which is part of the GeneMANIA command line tools. We also describe its use in the GeneMANIA quality control system.Availability: The software is available in Java JAR format at http://pages.genemania.org/tools/.

Comprehensive functional annotation of vertebrate genomes is fundamental to biological discovery. Reverse genetic screening has been highly useful for determination of gene function, but is untenable as a systematic approach in vertebrate model organisms given the number of surveyable genes and observable phenotypes. Unbiased prediction of gene-phenotype relationships offers a strategy to direct finite experimental resources towards likely phenotypes, thus maximizing de novo discovery of gene functions. Here we prioritized genes for phenotypic assay in zebrafish through machine learning, predicting the effect of loss of function of each of 15,106 zebrafish genes on 338 distinct embryonic anatomical processes. Focusing on cardiovascular phenotypes, the learning procedure predicted known knockdown and mutant phenotypes with high precision. In proof-of-concept studies we validated 16 high-confidence cardiac predictions using targeted morpholino knockdown and initial blinded phenotyping in embryonic zebrafish, confirming a significant enrichment for cardiac phenotypes as compared with morpholino controls. Subsequent detailed analyses of cardiac function confirmed these results, identifying novel physiological defects for 11 tested genes. Among these we identified tmem88a, a recently described attenuator of Wnt signaling, as a discrete regulator of the patterning of intercellular coupling in the zebrafish cardiac epithelium. Thus, we show that systematic prioritization in zebrafish can accelerate the pace of developmental gene function discovery.

RNA-binding proteins are key regulators of gene expression, yet only a small fraction have been functionally characterized. Here we report a systematic analysis of the RNA motifs recognized by RNA-binding proteins, encompassing 205 distinct genes from 24 diverse eukaryotes. The sequence specificities of RNA-binding proteins display deep evolutionary conservation, and the recognition preferences for a large fraction of metazoan RNA-binding proteins can thus be inferred from their RNA-binding domain sequence. The motifs that we identify in vitro correlate well with in vivo RNA-binding data. Moreover, we can associate them with distinct functional roles in diverse types of post-transcriptional regulation, enabling new insights into the functions of RNA-binding proteins both in normal physiology and in human disease. These data provide an unprecedented overview of RNA-binding proteins and their targets, and constitute an invaluable resource for determining post-transcriptional regulatory mechanisms in eukaryotes.

Tumor heterogeneity is a limiting factor in cancer treatment and in the discovery of biomarkers to personalize it. We describe a computational purification tool, ISOpure, which directly addresses the effects of variable contamination by normal tissue in clinical tumor specimens. ISOpure uses a set of tumor expression profiles and a panel of healthy tissue expression profiles to generate a purified cancer profile for each tumor sample, and an estimate of the proportion of RNA originating from cancerous cells. Applying ISOpure before identifying gene signatures leads to significant improvements in the prediction of prognosis and other clinical variables in lung and prostate cancer.

RNA binding proteins (RBPs) are key players in the regulation of gene expression. In this chapter we discuss the main protein-RNA recognition modes used by RBPs in order to regulate multiple steps of RNA processing. We discuss traditional and state-of-the-art technologies that can be used to study RNAs bound by individual RBPs, or vice versa, for both in vitro and in vivo methodologies. To help highlight the biological significance of RBP mediated regulation, online resources on experimentally verified protein-RNA interactions are briefly presented. Finally, we present the major tools to computationally infer RNA binding sites according to the modeling features and to the unsupervised or supervised frameworks that are adopted. Since some RNA binding site search algorithms are derived from DNA binding site search algorithms, we discuss the commonalities and novelties introduced to handle both sequence and structural features uniquely characterizing protein-RNA interactions.

There has been unprecedented public investment in sequencing human and cancer genomes in the hopes of understanding disease [1, 2]. At the same time, large genome-wide association studies have helped elucidating the genetic underpinning of common diseases, identifying thousands of putative disease relevant loci [7, 8]. Complementary molecular profiling studies have revealed that several of these loci are co-associated with individual mRNA levels, suggesting candidate pathways that are putative mediators of genetic signals [7]. Coupled with the public investment, there is considerable personal investment in genetic profiling, being offered by companies such as 23andMe, deCODEme and others. Although this work has led to amazing discoveries, such as the surprising genetic, subclonal diversity within tumor populations (e.g., [3-5]), it’s not clear how much how these insights will improve personalization of medicine.

Background High-throughput sequencing allows the detection and quantification of frequencies of somatic single nucleotide variants (SNV) in heterogeneous tumor cell populations. In some cases, the evolutionary history and population frequency of the subclonal lineages of tumor cells present in the sample can be reconstructed from these SNV frequency measurements. But automated methods to do this reconstruction are not available and the conditions under which reconstruction is possible have not been described. Results We describe the conditions under which the evolutionary history can be uniquely reconstructed from SNV frequencies from single or multiple samples from the tumor population and we introduce a new statistical model, PhyloSub, that infers the phylogeny and genotype of the major subclonal lineages represented in the population of cancer cells. It uses a Bayesian nonparametric prior over trees that groups SNVs into major subclonal lineages and automatically estimates the number of lineages and their ancestry. We sample from the joint posterior distribution over trees to identify evolutionary histories and cell population frequencies that have the highest probability of generating the observed SNV frequency data. When multiple phylogenies are consistent with a given set of SNV frequencies, PhyloSub represents the uncertainty in the tumor phylogeny using a “partial order plot”. Experiments on a simulated dataset and two real datasets comprising tumor samples from acute myeloid leukemia and chronic lymphocytic leukemia patients demonstrate that PhyloSub can infer both linear (or chain) and branching lineages and its inferences are in good agreement with ground truth, where it is available. Conclusions PhyloSub can be applied to frequencies of any “binary” somatic mutation, including SNVs as well as small insertions and deletions. The PhyloSub and partial order plot software is available from https://github.com/morrislab/phylosub/.

467 Background: There is an unmet need for a comprehensive genomic characterization of non muscle invasive bladder cancer (NMIBC). NMIBC comprise over 70% of all bladder cancers at presentation. They have highly variable clinical behavior that is not always adequately predicted on the basis of their histological grade (2004 World Health Organization low and high grade, LG-HG). The discrepancy between phenotype and genotype is compounded further by interobserver variability in pathological grading. We have previously established methods for whole transcriptome RNAseq from formalin fixed paraffin embedded tissues (FFPE). Methods: Whole transcriptomic analysis of 110 NMI FFPE BC both LG and HG was performed incorporating messenger RNA expression, splice variants, gene fusion, and pathway perturbation. We used a discovery (n = 40) and a validation cohort (n = 70). These data were integrated and tested for correlation with both pathological grading and clinical outcomes. Grade Risk Index (GRI) score quantifying how closely a patient's transcriptome is related to a reference set of LG NMIBC samples was established. Conventional pathological grading was reviewed by 3 different expert uro-pathologists and interobserver variability calculated. Results: Unsupervised clustering of data from RNA sequencing uncovered classification of three robust - - nonoverlapping, prognostically significant subtypes of NMIBC with distinct GRIs and signatures. When applied by expert pathologists, interobserver variability in histological grading was observed in 17.5%. In the intermediate group (GRI 0.13 to 0.19), pathologists disagreed in 37.5% whether BC was LG or HG. HG NMIBC clustered with MIBC. LG NMIBC in the intermediate GRI group included either very bulky tumors or extremely rare metastatic LG BC (n = 4). HG disease was associated with a shift in BMP signaling and a germ stem cell-like phenotype. Multiple components of the centromere complex and APOBEC3B were upregulated in HG BC. FGFR3::TACC3 fusion events were observed in LG NMIBC only (11.5%). Conclusions: Whole transcriptomic sequencing data delineated three molecular classes of NMIBC.

The problem of multilabel classification when the labels are related through a hierarchical categorization scheme occurs in many application domains such as computational biology. For example, this problem arises naturally when trying to automatically assign gene function using a controlled vocabularies like Gene Ontology. However, most existing approaches for predicting gene functions solve independent classification problems to predict genes that are involved in a given function category, independently of the rest. Here, we propose two simple methods for incorporating information about the hierarchical nature of the categorization scheme. In the first method, we use information about a gene's previous annotation to set an initial prior on its label. In a second approach, we extend a graph-based semi-supervised learning algorithm for predicting gene function in a hierarchy. We show that we can efficiently solve this problem by solving a linear system of equations. We compare these approaches with a previous label reconciliation-based approach. Results show that using the hierarchy information directly, compared to using reconciliation methods, improves gene function prediction., Appears in Proceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence (UAI2009)

Background Epigenetic modifications, transcription factor (TF) availability and differences in chromatin folding influence how the genome is interpreted by the transcriptional machinery responsible for gene expression. Enhancers buried in non-coding regions are found to be associated with significant differences in histone marks between different cell types. In contrast, gene promoters show more uniform modifications across cell types. Here we used histone modification and chromatin-associated protein ChIP-Seq data sets in mouse embryonic stem (ES) cells as well as genomic features to identify functional enhancer regions. Using co-bound sites of OCT4, SOX2 and NANOG (co-OSN, validated enhancers) and co-bound sites of MYC and MYCN (limited enhancer activity) as enhancer positive and negative training sets, we performed multinomial logistic regression with LASSO regularization to identify key features. Results Cross validations reveal that a combination of p300, H3K4me1, MED12 and NIPBL features to be top signatures of co-OSN regions. Using a model from 10 signatures, 83% of top 1277 putative 1 kb enhancer regions (probability greater than or equal to 0.8) overlapped with at least one TF peak from 7 mouse ES cell ChIP-Seq data sets. These putative enhancers are associated with increased gene expression of neighbouring genes and significantly enriched in multiple TF bound loci in agreement with combinatorial models of TF binding. Furthermore, we identified several motifs of known TFs significantly enriched in putative enhancer regions compared to random promoter regions and background. Comparison with an active H3K27ac mark in various cell types confirmed cell type-specificity of these enhancers. Conclusions The top enhancer signatures we identified (p300, H3K4me1, MED12 and NIPBL) will allow for the identification of cell type-specific enhancer regions in diverse cell types.

The cellular composition of heterogeneous samples can be predicted using an expression deconvolution algorithm to decompose their gene expression profiles based on pre-defined, reference gene expression profiles of the constituent populations in these samples. However, the expression profiles of the actual constituent populations are often perturbed from those of the reference profiles due to gene expression changes in cells associated with microenvironmental or developmental effects. Existing deconvolution algorithms do not account for these changes and give incorrect results when benchmarked against those measured by well-established flow cytometry, even after batch correction was applied. We introduce PERT, a new probabilistic expression deconvolution method that detects and accounts for a shared, multiplicative perturbation in the reference profiles when performing expression deconvolution. We applied PERT and three other state-of-the-art expression deconvolution methods to predict cell frequencies within heterogeneous human blood samples that were collected under several conditions (uncultured mono-nucleated and lineage-depleted cells, and culture-derived lineage-depleted cells). Only PERT's predicted proportions of the constituent populations matched those assigned by flow cytometry. Genes associated with cell cycle processes were highly enriched among those with the largest predicted expression changes between the cultured and uncultured conditions. We anticipate that PERT will be widely applicable to expression deconvolution strategies that use profiles from reference populations that vary from the corresponding constituent populations in cellular state but not cellular phenotypic identity., Author Summary The cellular composition of heterogeneous samples can be predicted from reference gene expression profiles that represent the homogeneous, constituent populations of the heterogeneous samples. However, existing methods fail when the reference profiles are not representative of the constituent populations. We developed PERT, a new probabilistic expression deconvolution method, to address this limitation. PERT was used to deconvolve the cellular composition of variably sourced and treated heterogeneous human blood samples. Our results indicate that even after batch correction is applied, cells presenting the same cell surface antigens display different transcriptional programs when they are uncultured versus culture-derived. Given gene expression profiles of culture-derived heterogeneous samples and profiles of uncultured reference populations, PERT was able to accurately recover proportions of the constituent populations composing the heterogeneous samples. We anticipate that PERT will be widely applicable to expression deconvolution strategies that use profiles from reference populations that vary from the corresponding constituent populations in cellular state but not cellular phenotypic identity.

Genome Wide Association (GWA) studies resulted in discovery of genetic variants underlying several complex diseases including Chron's disease and age-related macular degeneration (AMD). Still geneticists find that in majority of studies the size of the effect even if it is significant tends to be very small. There are several factors contributing to this problem such as rare variants, complex relationships among SNPs (epistatic effect), and heterogeneity of the phenotype. In this work we focus on addressing phenotypic heterogeneity. We introduce the problem of identifying, from GWAS data, separate genotypic markers from overlapping mixtures of clinically indistinguishable phenotypes. We propose a generative model for this scenario and derive an expectation-maximization (EM) procedure to fit the model to data, as well as a novel screening procedure designed to identify skew specific to certain phenotypic regimes. We present results on several simulated datasets as well as preliminary findings in applying the model to type 2 diabetes dataset.

A large number of genome-scale networks, including protein-protein and genetic interaction networks, are now available for several organisms. In parallel, many studies have focused on analyzing, characterizing, and modeling these networks. Beyond investigating the topological characteristics such as degree distribution, clustering coefficient, and average shortest-path distance, another area of particular interest is the prediction of nodes (genes) with a given characteristic (labels) - for example prediction of genes that cause a particular phenotype or have a given function. In this chapter, we describe methods and algorithms for predicting node labels from network-based datasets with an emphasis on label propagation algorithms (LPAs) and their relation to local neighborhood methods.

A large number of genome-scale networks, including protein-protein and genetic interaction networks, are now available for several organisms. In parallel, many studies have focused on analyzing, characterizing, and modeling these networks. Beyond investigating the topological characteristics such as degree distribution, clustering coefficient, and average shortest-path distance, another area of particular interest is the prediction of nodes (genes) with a given characteristic (labels) - for example prediction of genes that cause a particular phenotype or have a given function. In this chapter, we describe methods and algorithms for predicting node labels from network-based datasets with an emphasis on label propagation algorithms (LPAs) and their relation to local neighborhood methods.

Current microRNA target prediction techniques provide long lists of putative miRNA-target interactions, many of which are false positives. The goal of this paper is to identify functional targets in these lists based on biological evidence obtained from the expression profiles of the host genes of intronic miRNAs and those of their targets. We propose a scoring strategy for each interaction based on the combinatorial effect of miRNAs. In particular, the change in expression level of a target gene is expressed in terms of a linear combination of the host gene data which are used as surrogates for expression data of the intronic miRNAs. The parameters of this linear model give an estimate of the contribution of each intronic miRNA in down-regulating the target gene. The experimental results show that our prediction technique is able to detect several functional interactions. In addition, the analysis of mRNA microarrays after intronic miRNA transfection confirms that significantly down-regulated genes are among targets detected by our technique.

Summary: Cytoscape Web is a web-based network visualization tool–modeled after Cytoscape–which is open source, interactive, customizable and easily integrated into web sites. Multiple file exchange formats can be used to load data into Cytoscape Web, including GraphML, XGMML and SIF. Availability and Implementation: Cytoscape Web is implemented in Flex/ActionScript with a JavaScript API and is freely available at http://cytoscapeweb.cytoscape.org/ Contact: gary.bader@utoronto.ca Supplementary information: Supplementary data are available at Bioinformatics online.

Making Connections Genetic interaction profiles highlight cross-connections between bioprocesses, providing a global view of cellular pleiotropy, and enable the prediction of genetic network hubs. Costanzo et al. (p. 425 ) performed a pairwise fitness screen covering approximately one-third of all potential genetic interactions in yeast, examining 5.4 million gene-gene pairs and generating quantitative profiles for ∼75% of the genome. Of the pairwise interactions tested, about 3% of the genes investigated interact under the conditions tested. On the basis of these data, a reference map for the yeast genetic network was created.

Transcriptional Regulation Gets More Complicated Sequence preferences of DNA binding proteins are a primary mechanism by which cells interpret the genome. A central goal in genome biology is to identify regulatory sequences in the genome; however, few proteins' DNA binding specificities have been characterized comprehensively. Badis et al. (p. 1720 , published online 14 May) studied 104 known and predicted transcription factors (TFs), spanning 22 structural classes, in the mouse genome. While traditional models of TF binding sites are based on a single collection of highly similar DNA sequences, binding profiles were represented better by multiple motifs. Roughly half of the TFs recognized distinct primary and secondary motifs that are different from each other. At least some of these interaction modes appeared to be attributable to biophysically distinct protein conformations, adding to the complexity of transcriptional regulation.

Metazoan genomes encode hundreds of RNA-binding proteins (RBPs) but RNA-binding preferences for relatively few RBPs have been well defined. Current techniques for determining RNA targets, including in vitro selection and RNA co-immunoprecipitation, require significant time and labor investment. Here we introduce RNAcompete, a method for the systematic analysis of RNA binding specificities that uses a single binding reaction to determine the relative preferences of RBPs for short RNAs that contain a complete range of k-mers in structured and unstructured RNA contexts. We tested RNAcompete by analyzing nine diverse RBPs (HuR, Vts1, FUSIP1, PTB, U1A, SF2/ASF, SLM2, RBM4 and YB1). RNAcompete identified expected and previously unknown RNA binding preferences. Using in vitro and in vivo binding data, we demonstrate that preferences for individual 7-mers identified by RNAcompete are a more accurate representation of binding activity than are conventional motif models. We anticipate that RNAcompete will be a valuable tool for the study of RNA-protein interactions.

Large-scale proteomic approaches have been used to study signaling pathways. However, identification of biologically relevant hits from a single screen remains challenging due to limitations inherent in each individual approach. To overcome these limitations, we implemented an integrated, multi-dimensional approach and used it to identify Wnt pathway modulators. The LUMIER protein-protein interaction mapping method was used in conjunction with two functional screens that examined the effect of overexpression and siRNA-mediated gene knockdown on Wnt signaling. Meta-analysis of the three data sets yielded a combined pathway score (CPS) for each tested component, a value reflecting the likelihood that an individual protein is a Wnt pathway regulator. We characterized the role of two proteins with high CPSs, Ube2m and Nkd1. We show that Ube2m interacts with and modulates beta-catenin stability, and that the antagonistic effect of Nkd1 on Wnt signaling requires interaction with Axin, itself a negative pathway regulator. Thus, integrated physical and functional mapping in mammalian cells can identify signaling components with high confidence and provides unanticipated insights into pathway regulators.

Motivation: Recognition of specific DNA sequences is a central mechanism by which transcription factors (TFs) control gene expression. Many TF-binding preferences, however, are unknown or poorly characterized, in part due to the difficulty associated with determining their specificity experimentally, and an incomplete understanding of the mechanisms governing sequence specificity. New techniques that estimate the affinity of TFs to all possible k-mers provide a new opportunity to study DNA–protein interaction mechanisms, and may facilitate inference of binding preferences for members of a given TF family when such information is available for other family members. Results: We employed a new dataset consisting of the relative preferences of mouse homeodomains for all eight-base DNA sequences in order to ask how well we can predict the binding profiles of homeodomains when only their protein sequences are given. We evaluated a panel of standard statistical inference techniques, as well as variations of the protein features considered. Nearest neighbour among functionally important residues emerged among the most effective methods. Our results underscore the complexity of TF–DNA recognition, and suggest a rational approach for future analyses of TF families. Contact: t.hughes@utorotno.ca Supplementary information: Supplementary data are available at Bioinformatics online., Canadian Institutes of Health Research, Ontario Research Fund, National Institutes of Health (U.S.), National Human Genome Research Institute (U.S.)

Background: Regulation of gene expression at the post-transcriptional level, including the control of splicing, mRNA transcript stability, localization, and translation, is widespread in eukaryotes. An explosion of recent findings underscores both the predominance and complexity of PTR, including the discovery of ubiquitous but previously unknown regulatory mechanisms, such as microRNAs[1] (miRNAs); the observation that RNA recognition motif (RRM) and Kelch homology (KH) RNA-binding domains (RBDs) are among the most numerous protein domains in metazoan genomes, including the human genome[2, 3]; and the finding that, in several cases, RNA-binding proteins (RBPs) bind and presumably co-regulate sets of functionally-related transcripts[4]. An overall hypothesis has emerged that mRNA-containing ribonucleoprotein complexes (mRNPs) are the functional equivalent of “post-transcriptional operons”[5]. There are, however, relatively few examples of such “operons”; many putative PTR cis- and trans-regulatory factors were discovered by computational analysis of genome sequences, and have not been subjected to individual directed experimentation. Nonetheless, the sheer number of miRNAs (>500 in humans) and apparent sequence-specific RBPs (i.e. those carrying RRM, KH, and other RBDs) suggests that PTR mechanisms are as abundant as transcriptional regulatory mechanisms. The study of PTR regulation presents a number of experimental challenges. Two of the major modes of PTR (control of subcellular localization and translation) cannot be assayed by analysis of total RNA extracted from cells; rather, they must be analyzed either in situ or via extraction of intact macromolecular structures (e.g. organelles, Pbodies, polysomes). Another major mode, transcript degradation rate, must be separated from transcript synthesis rate in order to make accurate measurements. Coupled with these experimental challenges are computational challenges both in

Background: Several years after sequencing the human genome and the mouse genome, much remains to be discovered about the functions of most human and mouse genes. Computational prediction of gene function promises to help focus limited experimental resources on the most likely hypotheses. Several algorithms using diverse genomic data have been applied to this task in model organisms; however, the performance of such approaches in mammals has not yet been evaluated. Results: In this study, a standardized collection of mouse functional genomic data was assembled; nine bioinformatics teams used this data set to independently train classifiers and generate predictions of function, as defined by Gene Ontology (GO) terms, for 21,603 mouse genes; and the best performing submissions were combined in a single set of predictions. We identified strengths and weaknesses of current functional genomic data sets and compared the performance of function prediction algorithms. This analysis inferred functions for 76% of mouse genes, including 5,000 currently uncharacterized genes. At a recall rate of 20%, a unified set of predictions averaged 41% precision, with 26% of GO terms achieving a precision better than 90%. Conclusion: We performed a systematic evaluation of diverse, independently developed computational approaches for predicting gene function from heterogeneous data sources in mammals. The results show that currently available data for mammals allows predictions with both breadth and accuracy. Importantly, many highly novel predictions emerge for the 38% of mouse genes that remain uncharacterized.

We present a new model and learning algorithm, GenMiR3, which takes into account mRNA sequence features in addition to paired mRNA and miRNA expression profiles when scoring candidate miRNA-mRNA interactions. We evaluate three candidate sequence features for predicting miRNA targets by assessing the expression support for the predictions of each feature and the consistency of Gene Ontology Biological Process annotation of their target sets. We consider as sequence features the total energy of hybridization between the microRNA and target, conservation of the target site and the context score which is a composite of five individual sequence features. We demonstrate that only the total energy of hybridization is predictive of paired miRNA and mRNA expression data and Gene Ontology enrichment but this feature adds little to the total accuracy of GenMiR3 predictions using for expression features alone.

We have developed a new method that uses high-throughput reads that span multiple somatic point mutations to reconstruct multiple, genetically diverse subclonal populations from one or more heterogeneous tumor samples. Subclonal reconstruction algorithms attempt to infer the prevalence and genotype of multiple, genetically-related subclonal populations using the variant allele frequency (VAF) of somatic variants. To date, these algorithms exclusively use data on individual somatic mutations. This restriction greatly reduces their ability to fully resolve phylogenic ambiguities. In some cases, it is possible to determine the mutation status of >1 mutation in a single cell, for example, when single reads cover multiple single nucleotide variants (SNVs). This type of information can add considerable power to the phylogenetic reconstruction of the tumor subclonal population. We have developed the PhyloSpan algorithm which attempts to infer the states of multiple SNVs in single cells, and then exploits that information in subclonal reconstruction. Our algorithm starts with phasing somatic SNVs by looking for reads / read-pairs that cover both a somatic mutation and germline heterozygous single nucleotide polymorphism (SNP). These germline SNPs are often available through profiling of normal tissue. PhyloSpan then identifies SNVs that are on the same chromosome and close enough to be covered by a single read or paired reads. These pairs of mutations provide more phylogenetic certainty than can be found by looking at mutations independently. For example, if those SNVs are found in the same evolutionary branch, then we expect to see some reads containing both mutations. If however, the SNVs are an separate branches then no reads should show both SNVs. PhyloSpan integrates this phylogenetic information, along with information about the VAF of each somatic SNV in order to perform subclonal reconstruction. Incorporating these various types of information requires a rigorous statistical approach, and so we have developed a Bayesian non-parametric tree-based clustering algorithm. This algorithm not only infers the number of subclonal populations and their genotype but also provides a measure of uncertainty about this inference, enabling users to determine which parts of the subclonal reconstruction are certain and which parts remain ambiguous. While the number of SNVs a short-read length distance away from another SNV is small, a handful of such pairs are all that is needed to eliminate a substantial amount of ambiguity in subclonal reconstruction. Furthermore, long read technologies, such as PacBio, can be used to supplement short reads. Our approach generalizes to permit the integration of single-cell sequencing with bulk tumor sequencing. We will present results applying our algorithm to whole genome sequencing data showing the added value of considering multiple SNVs compared to independent SNVs. Citation Format: Amit G. Deshwar, Levi Boyles, Jeff Wintersinger, Paul C. Boutros, Yee Whye Teh, Quaid Morris. PhyloSpan: using multi-mutation reads to resolve subclonal architectures from heterogeneous tumor samples. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 4865. doi:10.1158/1538-7445.AM2015-4865

SummaryStem cell fate is governed by the integration of intrinsic and extrinsic positive and negative signals upon inherent transcriptional networks. To identify novel embryonic stem cell (ESC) regulators and assemble transcriptional networks controlling ESC fate, we performed temporal expression microarray analyses of ESCs after the initiation of commitment and integrated these data with known genome-wide transcription factor binding. Effects of forced under- or overexpression of predicted novel regulators, defined as differentially expressed genes with potential binding sites for known regulators of pluripotency, demonstrated greater than 90% correspondence with predicted function, as assessed by functional and high-content assays of self-renewal. We next assembled 43 theoretical transcriptional networks in ESCs, 82% (23 out of 28 tested) of which were supported by analysis of genome-wide expression in Oct4 knockdown cells. By using this integrative approach, we have formulated novel networks describing gene repression of key developmental regulators in undifferentiated ESCs and successfully predicted the outcomes of genetic manipulation of these networks.

Genome-wide microarray designs containing millions to hundreds of millions of probes are available for a variety of mammals, including mouse and human. These genome tiling arrays can potentially lead to significant advances in science and medicine, e.g., by indicating new genes and alternative primary and secondary transcripts. While bottom-up pattern matching techniques (e.g., hierarchical clustering) can be used to find gene structures in microarray data, we believe the many interacting hidden variables and complex noise patterns more naturally lead to an analysis based on generative models. We describe a generative model of tiling data and show how the sum-product algorithm can be used to infer hybridization noise, probe sensitivity, new transcripts, and alternative transcripts. The method, called GenRate, maximizes a global scoring function that enables multiple transcripts to compete for ownership of putative probes. We apply GenRate to a new exon tiling dataset from mouse chromosome 4 and show that it makes significantly more predictions than a previously described hierarchical clustering method at the same false positive rate. GenRate correctly predicts many known genes and also predicts new gene structures. As new problems arise, additional hidden variables can be incorporated into the model in a principled fashion, so we believe that GenRate will prove to be a useful tool in the new era of genome-wide tiling microarray analysis.

Motivation: We address the problem of multi-way clustering of microarray data using a generative model. Our algorithm, probabilistic sparse matrix factorization (PSMF), is a probabilistic extension of a previous hard-decision algorithm for this problem. PSMF allows for varying levels of sensor noise in the data, uncertainty in the hidden prototypes used to explain the data and uncertainty as to the prototypes selected to explain each data vector. Results: We present experimental results demonstrating that our method can better recover functionally-relevant clusterings in mRNA expression data than standard clustering techniques, including hierarchical agglomerative clustering, and we show that by computing probabilities instead of point estimates, our method avoids converging to poor solutions. Contact: delbert@psi.toronto.edu

Recently, researchers have made some progress in using microarrays to validate predicted exons in genome sequence and find new gene structures. However, current methods rely on separately making threshold-based decisions on intensity of expression, similarity of expression profiles, and arrangements of exons in the genome. We have taken a Bayesian approach and developed GenRate, a generative model that accounts for both genome-wide expression data taken from multiple conditions (e.g. tissues) and co-location and density of probes in DNA sequence data. GenRate balances probabilistic evidence derived from different sources and outputs scores (log-likelihoods) for each gene model, enabling the estimation of false-positive and false-negative rates. The model has a number of local minima that is exponential in the length of the DNA sequence data, so direct application of the EM learning algorithm produces poor results. We describe a novel way of parameterizing the model using examples from the data set, so that good solutions are found using an efficient algorithm. We apply GenRate to a subset of mouse genome-wide expression data that we have created, and discuss the statistical significance of the genes found by GenRate. Three of the highest-ranking gene structures found by GenRate, each containing thousands of bases from the genome, are confirmed using RT-PCR experiments.

A major objective in post-genome research is to fully understand the transcriptional control of each gene and the targets of each transcription factor. In yeast, large-scale experimental and computational approaches have been applied to identify co-regulated genes, cis regulatory elements, and transcription factor DNA binding sites in vivo. Methods for modeling and predicting system behavior, and for reconciling discrepancies among data types, are being explored. The results indicate that a complete and comprehensive yeast transcriptional network will ultimately be achieved.

Nearly 20% of yeast genes are required for viability, hindering genetic analysis with knockouts. We created promoter-shutoff strains for over two-thirds of all essential yeast genes and subjected them to morphological analysis, size profiling, drug sensitivity screening, and microarray expression profiling. We then used this compendium of data to ask which phenotypic features characterized different functional classes and used these to infer potential functions for uncharacterized genes. We identified genes involved in ribosome biogenesis (HAS1, URB1, and URB2), protein secretion (SEC39), mitochondrial import (MIM1), and tRNA charging (GSN1). In addition, apparent negative feedback transcriptional regulation of both ribosome biogenesis and the proteasome was observed. We furthermore show that these strains are compatible with automated genetic analysis. This study underscores the importance of analyzing mutant phenotypes and provides a resource to complement the yeast knockout collection.

In this issue, Zhao and Stormo introduce a new method for deriving position weight matrices (PWMs) from Protein Binding microarrays (BEEML-PBM). Using this method, they challenge a central claim of our 2009 paper 1, by “[concluding] that the widespread phenomenon of secondary binding preference identified by Badis et al. is not supported by the data and is likely due to suboptimal estimation of the PWM.” BEEML-PBM is simple, elegant and corrects for a pronounced positional effect of TF binding in the PBM assay; however we do not agree with their overall conclusion and believe that it is based on incomplete and biased analysis of our data. Zhao and Stormo’s conclusions are based on comparing the performance of BEEML-PBM PWMs and our methods on held-out data. However, they over-estimate the performance of their PWMs and under-estimate the performance of our methods. First, their claims of suboptimality of our PWMs are based on results from only one of the three motif finders that we employed, Seed-and-Wobble (SnW). SnW was not developed to predict probe intensities and does not attempt to produce a summary PWM that optimizes performance over all probes in predicting probe intensities. Instead, it was developed for the purpose of summarizing the 8-mer data, seeding with the highest scoring 8-mer, in a compact way for use in visual depiction as sequence logos. In contrast, another of the methods we employed, RankMotif++, is designed to produce summary PWMs and we have previously reported2 that it, like BEEML-PBM, better predicts probe intensities than SnW. So we suspect its performance would be much more competitive. In fact, RankMotif++ is very similar to the BEEML base method3; it fits a PWM model using a regression-like procedure to optimally predict PBM intensity data. RankMotif++ differs from BEEML primarily in that regresses on a partial preference ordering of probes inferred from their PBM intensities rather than on their actual intensities themselves. We acknowledge that comparisons with RankMotif++ PWMs would have been difficult because, although the source code for RankMotif++ has been available for three years, the PWMs we learned for Badis et al. were until recently only available as sequence logos. However, we made the motifs available to Zhao and Stormo when we were notified of this oversight and before the final submission of their paper. The motifs are available here: http://the_brain.bwh.harvard.edu/suppl105/. Second, we note that Zhao and Stormo use a positional effect model when training their PWMs but do not allow the methods that they are comparing against the same opportunity to correct this bias during training. We propose that this correction is a major cause of BEEML-PBM’s success and that both the multiple PWM methods and the 8-mer affinity estimates we employed would greatly benefit from a similar correction, thus restoring our reported gain in performance. For example, the 8-mer median intensities used in Figure 2A are not corrected for positional biases and this leads to the counter-intuitive claim that for the 15–20 (of 41) data points that lie above the diagonal, BEEML-PBM PWMs capture more than 100% of the replicate reproducibility. A more appropriate comparison would either employ PWMs uncorrected for positional bias (as we did in our original paper) or to compare against similarly corrected 8-mer median intensities. Zhao and Stormo do neither and as such, we believe that their prediction accuracy estimates are inflated. Finally, we note that explaining 90% of the reproducible binding signal is not the same as explaining 100%, and proteins that we and others have confirmed have multiple binding modes do not satisfy Zhao and Stormo’s 90% cut -off. For example, we reported that Jundm2 (Jdp2) binds two half -sites with variable spacing between them; this is clearly observed in the top-scoring 8-mers1. This mode of binding is common among other bZIP proteins. Furthermore, Zhao and Stormo do not consider the PBM data for Bcl6b, a C2H2 zinc finger for which we obtained two very different PWMs; these are also clearly observed in the top-scoring 8-mers, and moreover enrichment for motif matches can be observed in associated ChIP-chip data1. In general, variable spacing in long C2H2 zinc finger array seems to be common; for example, ChIP-seq for REST also supports use of partial vs. full sites, and different spacings4. Single summary PWMs cannot capture these binding modes, and it is important to do so, as C2H2 zinc fingers are the most common domain in metazoa, and long arrays of these domains are common in human and mouse genomes. We agree that simple and accurate representation of TF sequence specificity on the basis of PBM data is an important problem. We ourselves have been working on extensions to our algorithms to capture the PBM positional and orientation effects (which we have previously reported5). We also recently conducted a DREAM competition in which the goal was to predict PBM probe intensities using a two-array framework and evaluation criteria similar to those employed in1 and Zhao et al. A manuscript describing these new data, the competition, the methods of ~20 groups, their evaluation, and a web site that allows benchmarking any method to the DREAM results is in preparation (M. Weirauch, G. Stolovitsky et al.). We have now obtained the BEEML-PBM code, and we look forward

Smaug is an RNA-binding protein that induces the degradation and represses the translation of mRNAs in the early Drosophila embryo. Smaug has two identified direct target mRNAs that it differentially regulates: nanos and Hsp83. Smaug represses the translation of nanos mRNA but has only a modest effect on its stability, whereas it destabilizes Hsp83 mRNA but has no detectable effect on Hsp83 translation. Smaug is required to destabilize more than one thousand mRNAs in the early embryo, but whether these transcripts represent direct targets of Smaug is unclear and the extent of Smaug-mediated translational repression is unknown. To gain a panoramic view of Smaug function in the early embryo, we identified mRNAs that are bound to Smaug using RNA co-immunoprecipitation followed by hybridization to DNA microarrays. We also identified mRNAs that are translationally repressed by Smaug using polysome gradients and microarrays. Comparison of the bound mRNAs to those that are translationally repressed by Smaug and those that require Smaug for their degradation suggests that a large fraction of Smaug’s target mRNAs are both translationally repressed and degraded by Smaug. Smaug directly regulates components of the TRiC/CCT chaperonin, the proteasome regulatory particle and lipid droplets, as well as many metabolic enzymes, including several glycolytic enzymes. Smaug plays a direct and global role in regulating the translation and stability of a large fraction of the mRNAs in the early Drosophila embryo, and has unanticipated functions in control of protein folding and degradation, lipid droplet function and metabolism.

Summary: The GeneMANIA Cytoscape plugin brings fast gene function prediction capabilities to the desktop. GeneMANIA identifies the most related genes to a query gene set using a guilt-by-association approach. The plugin uses over 800 networks from six organisms and each related gene is traceable to the source network used to make the prediction. Users may add their own interaction networks and expression profile data to complement or override the default data. Availability and Implementation: The GeneMANIA Cytoscape plugin is implemented in Java and is freely available at http://www.genemania.org/plugin/. Contact: gary.bader@utoronto.ca; quaid.morris@utoronto.ca

Identifying microRNA signatures for the different types and subtypes of cancer can result in improved detection, characterization and understanding of cancer and move us towards more personalized treatment strategies. However, using microRNA's differential expression (tumour versus normal) to determine these signatures may lead to inaccurate predictions and low interpretability because of the noisy nature of miRNA expression data. We present a method for the selection of biologically active microRNAs using gene expression data and microRNA-to-gene interaction network. Our method is based on a linear regression with an elastic net regularization. Our simulations show that, with our method, the active miRNAs can be detected with high accuracy and our approach is robust to high levels of noise and missing information. Furthermore, our results on real datasets for glioblastoma and prostate cancer are confirmed by microRNA expression measurements. Our method leads to the selection of potentially functionally important microRNAs. The associations of some of our identified miRNAs with cancer mechanisms are already confirmed in other studies (hypoxia related hsa-mir-210 and apoptosis-related hsa-mir-296-5p). We have also identified additional miRNAs that were not previously studied in the context of cancer but are coherently predicted as active by our method and may warrant further investigation. The code is available in Matlab and R and can be downloaded on http://www.cs.toronto.edu/goldenberg/Anna_Goldenberg/Current_Research.html.

The functional module is fast becoming the operational unit of the postgenomics era. A new report in Nature by Gunsalus and colleagues describes, using a multiply supported network, functional modules within early C. elegans embryos and identifies several new components of known molecular machines (Gunsalus et al., 2005).

Background Popular miRNA target prediction techniques use sequence features to determine the functional miRNA target sites. These techniques commonly ignore the cellular conditions in which miRNAs interact with their targets in vivo. Gene expression data are rich resources that can complement sequence features to take into account the context dependency of miRNAs. Results We introduce BayMiR, a new computational method, that predicts the functionality of potential miRNA target sites using the activity level of the miRNAs inferred from genome-wide mRNA expression profiles. We also found that mRNA expression variation can be used as another predictor of functional miRNA targets. We benchmarked BayMiR, the expression variation, Cometa, and the TargetScan “context scores” on two tasks: predicting independently validated miRNA targets and predicting the decrease in mRNA abundance in miRNA overexpression assays. BayMiR performed better than all other methods in both benchmarks and, surprisingly, the variation index performed better than Cometa and some individual determinants of the TargetScan context scores. Furthermore, BayMiR predicted miRNA target sets are more consistently annotated with GO and KEGG terms than similar sized random subsets of genes with conserved miRNA seed regions. BayMiR gives higher scores to target sites residing near the poly(A) tail which strongly favors mRNA degradation using poly(A) shortening. Our work also suggests that modeling multiplicative interactions among miRNAs is important to predict endogenous mRNA targets. Conclusions We develop a new computational method for predicting the target mRNAs of miRNAs. BayMiR applies a large number of mRNA expression profiles and successfully identifies the mRNA targets and miRNA activities without using miRNA expression data. The BayMiR package is publicly available and can be readily applied to any mRNA expression data sets.

The properties (or labels) of nodes in networks can often be predicted based on their proximity and their connections to other labeled nodes. So-called “label propagation algorithms” predict the labels of unlabeled nodes by propagating information about local label density iteratively through the network. These algorithms are fast, simple and scale to large networks but nonetheless regularly perform better than slower and much more complex algorithms on benchmark problems. We show here, however, that these algorithms have an intrinsic limitation that prevents them from adapting to some common patterns of network node labeling; we introduce a new algorithm, 3Prop, that retains all their advantages but is much more adaptive. As we show, 3Prop performs very well on node labeling problems ill-suited to label propagation, including predicting gene function in protein and genetic interaction networks and gender in friendship networks, and also performs slightly better on problems already well-suited to label propagation such as labeling blogs and patents based on their citation networks. 3Prop gains its adaptability by assigning separate weights to label information from different steps of the propagation. Surprisingly, we found that for many networks, the third iteration of label propagation receives a negative weight. Availability The code is available from the authors by request.

Approximately half of known human miRNAs are located in the introns of protein coding genes. Some of these intronic miRNAs are only expressed when their host gene is and, as such, their steady state expression levels are highly correlated with those of the host gene's mRNA. Recently host gene expression levels have been used to predict the targets of intronic miRNAs by identifying other mRNAs that they have consistent negative correlation with. This is a potentially powerful approach because it allows a large number of expression profiling studies to be used but needs refinement because mRNAs can be targeted by multiple miRNAs and not all intronic miRNAs are co-expressed with their host genes. Here we introduce InMiR, a new computational method that uses a linear-Gaussian model to predict the targets of intronic miRNAs based on the expression profiles of their host genes across a large number of datasets. Our method recovers nearly twice as many true positives at the same fixed false positive rate as a comparable method that only considers correlations. Through an analysis of 140 Affymetrix datasets from Gene Expression Omnibus, we build a network of 19,926 interactions among 57 intronic miRNAs and 3,864 targets. InMiR can also predict which host genes have expression profiles that are good surrogates for those of their intronic miRNAs. Host genes that InMiR predicts are bad surrogates contain significantly more miRNA target sites in their 3′ UTRs and are significantly more likely to have predicted Pol II and Pol III promoters in their introns. We provide a dataset of 1,935 predicted mRNA targets for 22 intronic miRNAs. These prediction are supported both by sequence features and expression. By combining our results with previous reports, we distinguish three classes of intronic miRNAs: Those that are tightly regulated with their host gene; those that are likely to be expressed from the same promoter but whose host gene is highly regulated by miRNAs; and those likely to have independent promoters.

INTRODUCTION: Molecular-based diagnostics tools such as gene expression signatures for predicting patient outcome or response to treatment are cornerstones of personalized medicine. However, they have a reputation for poor validation performance on patient cohorts independent of those on which they were developed, a combined result of small patient cohorts and tumor heterogeneity. We quantitatively assess the impact of tumor heterogeneity on gene-expression based predictive models of patient outcome for non small cell lung cancer. EXPERIMENTS: We developed a fully automated pipeline to computationally purify tumor gene expression profiles using a tool called ISOLATE (Quon et al., 2009), identify a prognostic gene signature on a training cohort (Friedman et al., 2010), and validate the signature on independent cohorts. ISOLATE accepts heterogeneous tumor profiles and healthy tissue profiles, and outputs purified tumor profiles and percentage tumor content for each heterogeneous sample. We used 1 training cohort of 86 patients (Beer et al., 2002), and 4 independent validation cohorts totaling 443 patients (Shedden et al., 2008). The computational pipeline was run on the training and validation cohorts in two identical scenarios with the exception of either performing computational purification on all samples or not. RESULTS: We first validated the estimates of % tumor content of ISOLATE on 32 lung adenocarcinomas on which two pathologists independently estimated % content. For the 23 samples on which the two pathologists agreed strongly, ISOLATE estimates had 0.51 correlation with their average, compared with an overall correlation of 0.59 between the two pathologists. Our computational pipeline identified an 82-gene signature (‘unpurified-sig’) on the unpurified tumor profiles, and a 110-gene signature (‘ISOLATE-sig’) on the purified tumor profiles. We used these signatures to identify low and high risk patient groups within the validation cohorts of 443 patients. ISOLATE-sig achieved a very significant difference in survival between the identified low and high risk patients (Hazard Ratio (HR)=1.92, p=10⁁(−6)), compared to unpurified-sig (HR=1.53, p=10⁁(−3)). We separately examined all 227 Stage I patients, and found consistent results: ISOLATE-sig performance (HR=1.80, p=10⁁(−3)) compared very favorably to the original profiles (HR=1.45, p=10⁁(−2)). Here, only ISOLATE-sig achieves a hazard ratio whose 95% CI strictly exceeds 1. CONCLUSION: This is the first study to date that quantitatively assesses the effect of tumor heterogeneity on models for predicting patient outcome. We also present a fully automated computational pipeline for purifying, identifying, and validating gene expression based signatures. We showed purification lead to much better prognostic models of lung cancer, and anticipate similar results for other solid cancers. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 102nd Annual Meeting of the American Association for Cancer Research; 2011 Apr 2-6; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2011;71(8 Suppl):Abstract nr 4847. doi:10.1158/1538-7445.AM2011-4847

Metazoan genomes encode hundreds of RNA-binding proteins (RBPs). These proteins regulate post-transcriptional gene expression and have critical roles in numerous cellular processes including mRNA splicing, export, stability and translation. Despite their ubiquity and importance, the binding preferences for most RBPs are not well characterized. In vitro and in vivo studies, using affinity selection-based approaches, have successfully identified RNA sequence associated with specific RBPs; however, it is difficult to infer RBP sequence and structural preferences without specifically designed motif finding methods. In this study, we introduce a new motif-finding method, RNAcontext, designed to elucidate RBP-specific sequence and structural preferences with greater accuracy than existing approaches. We evaluated RNAcontext on recently published in vitro and in vivo RNA affinity selected data and demonstrate that RNAcontext identifies known binding preferences for several control proteins including HuR, PTB, and Vts1p and predicts new RNA structure preferences for SF2/ASF, RBM4, FUSIP1 and SLM2. The predicted preferences for SF2/ASF are consistent with its recently reported in vivo binding sites. RNAcontext is an accurate and efficient motif finding method ideally suited for using large-scale RNA-binding affinity datasets to determine the relative binding preferences of RBPs for a wide range of RNA sequences and structures., Author Summary Many disease-associated mutations do not change the protein sequence of genes; instead they change the instructions on how a gene's mRNA transcript should be processed. Translating these instructions allows us to better understand the connection between these mutations and disease. RNA-binding proteins (RBP) perform this translation by recognizing particular “phrases” that occupy short regions of the transcript. Recognition occurs by the binding of the RBP to the phrase. The set of phrases bound by a particular RBP is defined by the RNA base content of the binding site as well as the 3D configuration of these bases. Because it is impossible to assess RBP binding to every possible phrase, we have developed a mathematical model called RNAcontext that can be trained by measuring RBP binding strength on one set of phrases. Once trained, this model can then be used to accurately predict binding strength to any possible phrase. Compared to previously described methods, RNAcontext learns a more precise description of the 3D shapes of binding sites. This precision translates into more accurate generalization of RBP binding preferences to new phrases and allows us to make new discoveries about the binding preferences of well-studied RBPs.

Using a microarray that tiles all known yeast non-coding RNAs, we compared RNA from wild-type cells with RNA from mutants encoding known and putative RNA modifying enzymes. We show that at least five types of RNA modification (dihydrouridine, m1G, m2(2)G, m1A and m6(2)A) catalyzed by 10 different enzymes (Trm1p, Trm5, Trm10p, Dus1p-Dus4p, Dim1p, Gcd10p and Gcd14p) can be detected by virtue of differential hybridization to oligonucleotides on the array that are complementary to the modified sites. Using this approach, we identified a previously undetected m1A modification in GlnCTG tRNA, the formation of which is catalyzed by the Gcd10/Gcd14 complex. complex.

Large-scale quantitative analysis of transcriptional co-expression has been used to dissect regulatory networks and to predict the functions of new genes discovered by genome sequencing in model organisms such as yeast. Although the idea that tissue-specific expression is indicative of gene function in mammals is widely accepted, it has not been objectively tested nor compared with the related but distinct strategy of correlating gene co-expression as a means to predict gene function. We generated microarray expression data for nearly 40,000 known and predicted mRNAs in 55 mouse tissues, using custom-built oligonucleotide arrays. We show that quantitative transcriptional co-expression is a powerful predictor of gene function. Hundreds of functional categories, as defined by Gene Ontology 'Biological Processes', are associated with characteristic expression patterns across all tissues, including categories that bear no overt relationship to the tissue of origin. In contrast, simple tissue-specific restriction of expression is a poor predictor of which genes are in which functional categories. As an example, the highly conserved mouse gene PWP1 is widely expressed across different tissues but is co-expressed with many RNA-processing genes; we show that the uncharacterized yeast homolog of PWP1 is required for rRNA biogenesis. We conclude that 'functional genomics' strategies based on quantitative transcriptional co-expression will be as fruitful in mammals as they have been in simpler organisms, and that transcriptional control of mammalian physiology is more modular than is generally appreciated. Our data and analyses provide a public resource for mammalian functional genomics.

Mutational signatures are patterns of mutation types, many of which are linked to known mutagenic processes. Signature activity represents the proportion of mutations a signature generates. In cancer, cells may gain advantageous phenotypes through mutation accumulation, causing rapid growth of that subpopulation within the tumour. The presence of many subclones can make cancers harder to treat and have other clinical implications. Reconstructing changes in signature activities can give insight into the evolution of cells within a tumour. Recently, we introduced a new method, TrackSig, to detect changes in signature activities across time from single bulk tumour sample. By design, TrackSig is unable to identify mutation populations with different frequencies but little to no difference in signature activity. Here we present an extension of this method, TrackSigFreq, which enables trajectory reconstruction based on both observed density of mutation frequencies and changes in mutational signature activities. TrackSigFreq preserves the advantages of TrackSig, namely optimal and rapid mutation clustering through segmentation, while extending it so that it can identify distinct mutation populations that share similar signature activities.

The standard interpretation of importance-weighted autoencoders is that they maximize a tighter lower bound on the marginal likelihood than the standard evidence lower bound. We give an alternate interpretation of this procedure: that it optimizes the standard variational lower bound, but using a more complex distribution. We formally derive this result, present a tighter lower bound, and visualize the implicit importance-weighted distribution., ICLR 2017 Workshop