A Statistical Framework to Predict Functional Non-Coding Regions in the Human Genome Through Integrated Analysis of Annotation Data

Qiongshi Lu , Yiming Hu , Jiehuan Sun , Yuwei Cheng , Kei-Hoi Cheung , Hongyu Zhao
doi: http://dx.doi.org/10.1101/018093

Identifying functional regions in the human genome is a major goal in human genetics. Great efforts have been made to functionally annotate the human genome either through computational predictions, such as genomic conservation, or high-throughput experiments, such as the ENCODE project. These efforts have resulted in a rich collection of functional annotation data of diverse types that need to be jointly analyzed for integrated interpretation and annotation. Here we present GenoCanyon, a whole-genome annotation method that performs unsupervised statistical learning using 22 computational and experimental annotations thereby inferring the functional potential of each position in the human genome. With GenoCanyon, we are able to predict many of the known functional regions. The ability of predicting functional regions as well as its generalizable statistical framework makes GenoCanyon a unique and powerful tool for whole-genome annotation. The GenoCanyon web server is available at http://genocanyon.med.yale.edu

Natural selection defines the cellular complexity

Han Chen , Xionglei He
doi: http://dx.doi.org/10.1101/018069

Current biology is perplexed by the lack of a theoretical framework for understanding the organization principles of the molecular system within a cell. Here we first studied growth rate, one of the seemingly most complex cellular traits, using functional data of yeast single-gene deletion mutants. We observed nearly one thousand expression informative genes (EIGs) whose expression levels are linearly correlated to the trait within an unprecedentedly large functional space. A simple model considering six EIG-formed protein modules revealed a variety of novel mechanistic insights, and also explained ~50% of the variance of cell growth rates measured by Bar-seq technique for over 400 yeast mutants (Pearson’s R = 0.69), a performance comparable to the microarray-based (R = 0.77) or colony-size-based (R = 0.66) experimental approach. We then applied the same strategy to 501 morphological traits of the yeast and achieved successes in most fitness-coupled traits each with hundreds of trait-specific EIGs. Surprisingly, there is no any EIG found for most fitness-uncoupled traits, indicating that they are controlled by super-complex epistases that allow no simple expression-trait correlation. Thus, EIGs are recruited exclusively by natural selection, which builds a rather simple functional architecture for fitness-coupled traits, and the endless complexity of a cell lies primarily in its fitness-uncoupled features.

Interrogating conserved elements of diseases using Boolean combinations of orthologous phenotypes

John O Woods , Matthew Z Tien , Edward M Marcotte
doi: http://dx.doi.org/10.1101/017947

Conserved genetic programs often predate the homologous structures and phenotypes to which they give rise; eyes, for example, have evolved several dozen times, but their development seems to involve a common set of conserved genes. Recently, the concept of orthologous phenotypes (or phenologs) offered a quantitative way to describe this property. Phenologs are phenotypes or diseases from separate species who share an unexpectedly large set of their associated gene orthologs. It has been shown that the phenotype pairs which make up a phenolog are mutually predictive in terms of the genes involved. Recently, we demonstrated the ranking of gene–phenotype association predictions using multiple phenologs from an array of species. In this work, we demonstrate a computational method which provides a more targeted view of the conserved pathways which give rise to diseases. Our approach involves the generation of synthetic pseudo-phenotypes made up of Boolean combinations (union, intersection, and difference) of the gene sets for phenotypes from our database. We search for diseases that overlap significantly with these Boolean phenotypes, and find a number of highly predictive combinations. While set unions produce less specific predictions (as expected), intersection and difference-based combinations appear to offer insights into extremely specific aspects of target diseases. For example, breast cancer is predicted by zebrafish methylmercury response minus metal ion response, with predictions MT-COI, JUN, SOD2, GADD45B, and BAX all involved in the pro-apoptotic response to reactive oxygen species, thought to be a key player in cancer. We also demonstrate predictions from Arabidopsis Boolean phenotypes for increased brown adipose tissue in mouse (salt stress response’s intersection with sucrose stimulus response); and for human myopathy (red light response minus water deprivation response). We demonstrate the ranking of predictions for human holoprosencephaly from the set intersections between each pair of a variety of closely-related zebrafish phenotypes. Our results suggest that Boolean phenolog combinations may provide a more informed insight into the conserved pathways underlying diseases than either regular phenologs or the naïve Bayes approach.

Bayesian Modeling of Epigenetic Variation in Multiple Human Cell Types

Yu Zhang , Feng Yue , Ross C. Hardison
doi: http://dx.doi.org/10.1101/018028

With high-throughput sequencing data generated for multiple epigenetic features in many cell types, a chief challenge is to explain the dynamics in multiple epigenomes that lead to differential regulation and phenotypes. We introduce a Bayesian framework for jointly annotating multiple epigenomes and detecting differential regulation among multiple cell types. Our method, IDEAS (integrative and discriminative epigenome annotation system), achieves superior power by modeling both position and cell type specific epigenetic activities. Using ENCODE data sets in 6 cell types, we identified epigenetic variation strongly associated with differential gene expression. The detected regions are significantly enriched in disease genetic variants with much stronger enrichment scores than achievable by existing methods, and the enriched phenotypes are highly relevant to the corresponding cell types. IDEAS is a powerful tool for integrative epigenome annotation and detection of variation, which could be of important utility in elucidating the interplay between genetics, gene regulation and diseases.