Tissue Specificity

A large number of datasets originate outside UCSC and contribute to the Genome Browser’s core idea of hosting as many high-quality resources as possible. In most cases, UCSC does not perform significant postprocessing or computation on the data, limiting intervention to converting file formats or other parsing tasks and quality-assurance review. Examples of this type of track include probe sets for commercially available microarrays, human copy-number variation (CNV) data from the Database of Genomic Variants (DGV) [25 (link)]; human dismorphology data from DECIPHER [26 (link)]; expression data for mouse and human from the GNF Expression Atlas [27 (link)]; and segmental duplication data for human, mouse, rat, dog and chicken [28 (link)].
The ENCODE project, for which the UCSC Genome Browser is the Data Coordination Center [29 (link)], presents a large number of functional annotations: including DNAse hypersensitivity sites, indicating open chromatin; histone marks, implicated in gene regulation; and gene expression levels from whole-genome RNA-seq experiments. These data, which are available on the human and mouse assemblies hg19 and mm9, are mapped across multiple cell lines. The resulting tracks represent tissue specificity and developmental mileposts (e.g. embryonic stem cells) for these elements. They can be displayed along with any other tracks on the same assembly, such as GenBank mRNAs or multispecies conservation.
A complete list of tracks available for any assembly can be found by visiting the Gateway page for any genome assembly (http://genome.ucsc.edu/cgi-bin/hgGateway) and clicking the button, ‘configure tracks and display’ or by simply inspecting the track controls beneath the main Browser graphic. The Track Search feature provides keyword lookup.
Examples of data tracks that do undergo further processing or filtering at UCSC include dbSNP [30 (link)] and OMIM (Online Mendelian Inheritance in Man) [31 (link)]. In these tracks, data from the providers are subdivided into categories to make them more useful to our users. For example, dbSNP data are presented in their entirety in one track, but three other tracks offer subsets: Common Single Nucleotide Polymorphisms (SNPs) (those with minor allele frequency >1%), Flagged SNPs (those identified in dbSNP as ‘clinical’—may be associated with disease, but use with caution!) and Multiple SNPs (those mapping to more than one genomic location).
Similarly, the OMIM data set has been filtered by UCSC to create three separate tracks, including one track of Allelic Variant SNPs that have phenotypic associations annotated by OMIM. These filtered sets are transmitted to OMIM for redistribution to their licensees. As always, details of how the filtering was done are available by clicking into an item or via the track configuration page.
Users may read about the filtering options available when using tracks by clicking on the small button to the left of the track in the Genome Browser image, or on the label in the track control area below the image. This configuration page gives users an opportunity to set colors and filters to suit themselves.
For users who do not know exactly which data set contains the information they seek, each data track is accompanied by a description outlining the rationale for the production of the data, implementation details, interpretation guidelines and references to the literature. All of this information is indexed and may be searched by keyword via the Track Search button beneath the Browser graphic. The result is a list of all tracks that have the search term in the documentation and a link to the track description.

We used GTEx v6 human gene expression data (Carithers et al. 2015 (link)) from control donors to infer transcriptional regulatory networks for healthy tissues. We downloaded gene-level raw counts for 18,737 samples from the Expression Atlas (Petryszak et al. 2014 (link)). One hundred forty-four samples with >30% of zero raw counts were discarded. Also, we removed genes with an average log counts per million (CPM) lower than zero. Next, we normalized the data using the TMM method implemented in the edgeR R package (version 3.14.0) (Robinson et al. 2010 (link)). Then, we used the voom function in the limma package (version 3.28.21) (Ritchie et al. 2015 (link)) to obtain fitted log₂ CPM. To account for potential sample batch effects, we downloaded the annotation file GTEx_Data_V6_Annotations_SampleAttributesDS.txt from the gtexportal.org and extracted the isolation batch field “SMNABTCH” and corrected it, keeping the histological type as a covariate, using ComBat function from sva R package (version 3.20.0) (Leek et al. 2012 (link)). We also assessed regulons based on noncorrected data (Supplemental Fig. S10). Samples from cell lines were discarded. Finally, replicates (i.e., samples from the same tissue and donor) were averaged. These data cover 9407 samples from 30 tissues (histological types) (Supplemental Fig. S1B). Next, we used the ARACNE software (version 1.4) (Margolin et al. 2006 (link)) to reverse engineer tissue-specific networks. For each tissue with at least 15 donors, we first precalculated the ARACNE threshold with a fixed seed with the --calculateThreshold parameter. Second, we ran 100 reproducible bootstraps with a controlled seed and, with the --consolidate parameter, derived the tissue-specific network. Finally, we used the aracne2regulon function in VIPER R package (version 1.12.0) (Alvarez et al. 2016 (link)) to infer the sign of each TF–target interaction (i.e., MoR; activation or inhibition). We also downloaded cancer regulatory networks from the aracne.networks R package (version 0.99.7) covering 14 cancer types from TCGA (https://bioconductor.org/packages/release/data/experiment/html/aracne.networks.html).
Finally, we also aggregated the data from all the tissue-specific/cancer-specific regulons to infer four healthy and pancancer consensus regulons by selecting TF–target signed interactions appearing in at least two, three, five, and 10 GTEx tissues or TCGA cancer types, respectively.