In vjcitn/TxRegInfra: Metadata management for multiomic specification of transcriptional regulatory networks

suppressPackageStartupMessages({
library(TxRegInfra)
library(GenomicFiles)
library(TFutils)
})

Introduction

TxRegQuery addresses exploration of transcriptional regulatory networks by integrating data on eQTL, digital genomic footprinting (DGF), DnaseI hypersensitivity binding data (DHS), and transcription factor binding site (TFBS) data. Owing to the volume of emerging tissue-specific data, special data modalities are used.

Managing bed file content with mongodb

Querying the txregnet database

We have a long-running server that will respond to queries. We focus on r CRANpkg("mongolite") as the interface.

The connection

suppressPackageStartupMessages({
library(TxRegInfra)
library(mongolite)
library(Gviz)
library(EnsDb.Hsapiens.v75)
library(BiocParallel)
register(SerialParam())
})
con1 = mongo(url=URL_txregInAWS(), db="txregnet")
con1

We will write methods that work with the 'fields' of this object.

There is not much explicit reflectance in the mongolite API. The following is improvised and may be fragile:

parent.env(con1)$orig

Queries and aggregation

If the mongo utility is available as a system command, we can get a list of collections in the database as follows.

if (verifyHasMongoCmd()) {
  head(c1 <- listAllCollections(url=URL_txregInAWS(), db="txregnet"))
  }

Otherwise, as long as r CRANpkg("mongolite") is installed, as long as we know the collection names of interest, we can use them as noted throughout this vignette.

We can get a record from a given collection:

mongo(url=URL_txregInAWS(), db="txregnet", 
   collection="Adipose_Subcutaneous_allpairs_v7_eQTL")$find(limit=1)

Queries can be composed using JSON. We have a tool to generate queries that employ the mongodb aggregation method. Here we demonstrate this by computing, for each chromosome, the count and minimum values of the footprint statistic on CD14 cells.

m1 = mongo(url = URL_txregInAWS(), db = "txregnet",  collection="CD14_DS17215_hg19_FP")
newagg = makeAggregator( by="chr", vbl="stat", op="$min", opname="min")

The JSON layout of this aggregating query is

[
  {
    "$group": {
      "_id": ["$chr"],
      "count": {
        "$sum": [1]
      },
      "min": {
        "$min": ["$stat"]
      }
    }
  }
] 

Invocation returns a data frame:

head(m1$aggregate(newagg))

An integrative container

We need to bind the metadata and information about the mongodb.

Sample metadata

The following turns a very ad hoc filtering of the collection names into a DataFrame.

# cd = makeColData() # works when mongo does
cd = TxRegInfra::basicColData
head(cd,2)

Extended RaggedExperiment

rme0 = RaggedMongoExpt(con1, colData=cd)
rme1 = rme0[, which(cd$type=="FP")]

A key method in development is subsetting the archive by genomic coordinates.

si = GenomeInfoDb::Seqinfo(genome="hg19")["chr17"] # to fix query genome
myg = GRanges("chr17", IRanges(38.07e6,38.09e6), seqinfo=si)
s1 = sbov(rme1, myg, simplify=FALSE)
s1
dim(sa <- sparseAssay(s1, 3))  # compact gives segfault
sa[953:956,c("fLung_DS14724_hg19_FP", "fMuscle_arm_DS17765_hg19_FP")]

Visualizing coincidence

ormm = txmodels("ORMDL3", plot=FALSE, name="ORMDL3")
sar = strsplit(rownames(sa), ":|-")
an = as.numeric
gr = GRanges(seqnames(ormm)[1], IRanges(an(sapply(sar,"[", 2)), an(sapply(sar,"[", 3))))
gr1 = gr
gr1$score = 1-sa[,1]
gr2 = gr
gr2$score = 1-sa[,2]
sc1 = DataTrack(gr1, name="Lung FP")
sc2 = DataTrack(gr2, name="Musc/Arm FP")
plotTracks(list(GenomeAxisTrack(), sc1, sc2, ormm), showId=TRUE)

Higher-level work with sbov

Building annotated GRanges for a selected target interval

We begin with three 'single-concept' assays with relevance to lung genomics. The v7 GTEx lung eQTL data, an encode DnaseI narrowPeak report on lung fibroblasts, and a digital genomic footprint report for fetal lung.

lname_eqtl = "Lung_allpairs_v7_eQTL"
lname_dhs = "ENCFF001SSA_hg19_HS" # see dnmeta, fibroblast of lung
lname_fp = "fLung_DS14724_hg19_FP"
si17 = GenomeInfoDb::Seqinfo(genome="hg19")["chr17"]
si17n = si17
GenomeInfoDb::seqlevelsStyle(si17n) = "NCBI"
s1 = sbov(rme0[,lname_eqtl], GRanges("17", IRanges(38.06e6, 38.15e6),
    seqinfo=si17n))
s2 = sbov(rme0[,lname_dhs], GRanges("chr17", IRanges(38.06e6, 38.15e6),
   seqinfo=si17))
s3 = sbov(rme0[,lname_fp], GRanges("chr17", IRanges(38.06e6, 38.15e6),
   seqinfo=si17))

Now we have annotated GRanges for each assay. The eQTL data in part are:

names(mcols(s1))
head(s1[, c("gene_id", "variant_id", "maf", "pval_nominal")])

The names of genes and variants used here are cumbersome -- symbols and rsids are preferable.

addsyms = function(x, EnsDb=EnsDb.Hsapiens.v75::EnsDb.Hsapiens.v75) {
  ensids = gsub("\\..*", "", x$gene_id) # remove post period
  gns = genes(EnsDb)
  x$symbol = gns[ensids]$symbol
  x
}
s1 = addsyms(s1)

Note that it is possible to retrieve rsids for the SNPs by address. But this is a slow operation involving a huge SNPlocs package that we do not want to work with directly for this vignette.

> snpsByOverlaps(SNPlocs.Hsapiens.dbSNP144.GRCh37, s1b)
UnstitchedGPos object with 265 positions and 2 metadata columns:
        seqnames       pos strand |   RefSNP_id alleles_as_ambig
           <Rle> <integer>  <Rle> | <character>      <character>
    [1]       17  38061054      * |  rs36049276                R
    [2]       17  38061439      * |   rs4795399                Y
    [3]       17  38062196      * |   rs2305480                R
    [4]       17  38062217      * |   rs2305479                Y
    [5]       17  38062503      * |  rs35104165                Y
    ...      ...       ...    ... .         ...              ...
  [261]       17  38149258      * |  rs58212353                K
  [262]       17  38149350      * |   rs8073254                V
  [263]       17  38149411      * |  rs34648856                R
  [264]       17  38149724      * |   rs3785549                Y
  [265]       17  38149727      * |   rs3785550                H
  -------
  seqinfo: 25 sequences (1 circular) from GRCh37.p13 genome

A bipartite graph for eQTL-gene relationships

The object s1 computed above is available as demo_eQTL_granges. We convert it to a graph via

library(graph)
g1 = sbov_to_graphNEL(demo_eQTL_granges)
g1

Nodes are SNPs and genes, edges are present when the resource (in this case the GTEx lung study) declares an association (in this case, an FDR for SNP-gene association not exceeding 0.10.) The r Biocpkg("graph") library includes functions for creation of incidence matrices from graphs, and vice versa.

Connecting eQTL-SNPs via DHS and DGF

Given the GRanges representations for sbov results, we can use overlap computations to conveniently identify relationships between eQTL SNPs, genes, and hypersensitivity or footprint regions.

We use sbov_output_HS as a persistent instance of s2 computed above.

seqlevelsStyle(demo_eQTL_granges) = "UCSC"
fo1 = findOverlaps(demo_eQTL_granges, sbov_output_HS)
fo1 
eq_by_hs = split(demo_eQTL_granges[queryHits(fo1)],
   subjectHits(fo1))
eq_by_hs

This shows that there are two DHS sites that overlap with SNPs showing eQTL associations with various genes.

For the footprint data, we have:

fo2 = findOverlaps(demo_eQTL_granges, sbov_output_FP)
fo2 
eq_by_fp = split(demo_eQTL_granges[queryHits(fo2)],
   subjectHits(fo2))
eq_by_fp

Relationships to FIMO-based TFBS

We have a small number of cloud-resident FIMO search results through the r Biocpkg("TFutils") package.

library(TFutils)
data(demo_fimo_granges)
seqlevelsStyle(demo_eQTL_granges) = "UCSC"
lapply(demo_fimo_granges, lapply, function(x) 
   subsetByOverlaps(demo_eQTL_granges, x))

vjcitn/TxRegInfra documentation built on July 6, 2019, 10:47 p.m.