In UPSCb/RnaSeqTutorial: RNA-Seq Tutorial

Extra - Working with Called Variants

    options(digits=2)
    library(EBI2015)
    library(VariantAnnotation)
    library(ggplot2)
    library(SNPlocs.Hsapiens.dbSNP.20101109)
    library(TxDb.Hsapiens.UCSC.hg19.knownGene)
    library(BSgenome.Hsapiens.UCSC.hg19)
    library(SIFT.Hsapiens.dbSNP132)

Annotation of Variants

A major product of DNASeq experiments are catalogs of called variants (e.g., SNPs, indels). We will use the VariantAnnotation package to explore this type of data. Sample data included in the package are a subset of chromosome 22 from the 1000 Genomes project. Variant Call Format (VCF; full description) text files contain meta-information lines, a header line with column names, data lines with information about a position in the genome, and optional genotype information on samples for each position.

Variant call format (VCF) files

Data are read from a VCF file and variants identified according to region such as coding, intron, intergenic, spliceSite etc. Amino acid coding changes are computed for the non-synonymous variants. SIFT and PolyPhen databases provide predictions of how severely the coding changes affect protein function.

Data exploration

The objective of this exercise is to compare the quality of called SNPs that are located in dbSNP, versus those that are novel.

Locate the sample data in the file system. Explore the metadata (information about the content of the file) using scanVcfHeader. Discover the ‘info’ fields VT (variant type), and RSQ (genotype imputation quality).

Input the sample data using readVcf. You’ll need to specify the genome build (genome=“hg19”) on which the variants are annotated. Take a peak at the rowData to see the genomic locations of each variant.

dbSNP uses abbreviations such as ch22 to represent chromosome 22, whereas the VCF file uses 22. Use rowData and renameSeqlevels to extract the row data of the variants, and rename the chromosomes.

The SNPlocs.Hsapiens.dbSNP.20101109 contains information about SNPs in a particular build of dbSNP. Load the package, use the dbSNPFilter function to create a filter, and query the row data of the VCF file for membership.

Create a data frame containing the dbSNP membership status and imputation quality of each SNP. Create a density plot to illustrate the results.

Explore the header:

    library(VariantAnnotation)
    fl <- system.file("extdata", "chr22.vcf.gz", package="VariantAnnotation")
    (hdr <- scanVcfHeader(fl))
    info(hdr)[c("VT", "RSQ"),]

Input the data and peak at their locations:

    (vcf <- readVcf(fl, "hg19"))
    head(rowRanges(vcf), 3)

Discover whether SNPs are located in dbSNP:

    library(SNPlocs.Hsapiens.dbSNP.20101109)
    rd <- rowRanges(vcf)
    seqlevels(rd) <- "ch22"
    ch22snps <- getSNPlocs("ch22")
    inDbSNP <- sub("rs", "", names(rd)) %in% ch22snps$RefSNP_id
    table(inDbSNP)

Create a data frame summarizing SNP quality and dbSNP membership:

    metrics <- data.frame(inDbSNP=inDbSNP, RSQ=info(vcf)$RSQ)

Finally, visualize the data, e.g., using ggplot2.

    library(ggplot2)
    ggplot(metrics, aes(RSQ, fill=inDbSNP)) +
       geom_density(alpha=0.5) +
       scale_x_continuous(name="MaCH / Thunder Imputation Quality") +
       scale_y_continuous(name="Density") +
       theme(legend.position="top")

Coding consequences

Locating variants in and around genes

Variant location with respect to genes can be identified with the locateVariants function. Regions are specified in the region argument and can be one of the following constructors: CodingVariants(), IntronVariants(), FiveUTRVariants(), ThreeUTRVariants(), IntergenicVariants(), SpliceSiteVariants(), or AllVariants(). Location definitions are shown in the table below.

| Location | Details | |------------------------------|--------------------------------------------------------------| | coding | Within a coding region | | fiveUTR | Within a 5’ untranslated region | | threeUTR | Within a 3’ untranslated region | | intron | Within an intron region | | intergenic | Not within a transcript associated with a gene | | spliceSite | Overlaps any of the first or last 2 nucleotides of an intron |

Load the TxDb.Hsapiens.UCSC.hg19.knownGene annotation package, and read in the chr22.vcf.gz example file from the VariantAnnotation package.

Remembering to re-name sequence levels, use the locateVariants function to identify coding variants.

Summarize aspects of your data, e.g., did any coding variants match more than one gene? How many coding variants are there per gene ID?

Here we open the known genes data base, and read in the VCF file.

    library(TxDb.Hsapiens.UCSC.hg19.knownGene)
    txdb <- TxDb.Hsapiens.UCSC.hg19.knownGene 

    fl <- system.file("extdata", "chr22.vcf.gz", package="VariantAnnotation")
    vcf <- readVcf(fl, "hg19")
    vcf <- renameSeqlevels(vcf, c("22"="chr22"))

The next lines locate coding variants.

    rd <- rowData(vcf)
    loc <- locateVariants(rd, txdb, CodingVariants())
    head(loc, 3)

To answer gene-centric questions data can be summarized by gene regardless of transcript.

    ## Did any coding variants match more than one gene?
    splt <- split(loc$GENEID, loc$QUERYID)
    table(sapply(splt, function(x) length(unique(x)) > 1))

    ## Summarize the number of coding variants by gene ID
    splt <- split(loc$QUERYID, loc$GENEID)
    head(sapply(splt, function(x) length(unique(x))), 3)

Amino acid coding changes

predictCoding computes amino acid coding changes for non-synonymous variants. Only ranges in query that overlap with a coding region in subject are considered. Reference sequences are retrieved from either a or fasta file specified in seqSource. Variant sequences are constructed by substituting, inserting or deleting values in the column into the reference sequence. Amino acid codes are computed for the variant codon sequence when the length is a multiple of 3.

The query argument to predictCoding can be a or . When a is supplied the varAllele argument must be specified. In the case of a object, the alternate alleles are taken from alt(<VCF>) and the varAllele argument is not specified.

The result is a modified query containing only variants that fall within coding regions. Each row represents a variant-transcript match so more than one row per original variant is possible.

    library(BSgenome.Hsapiens.UCSC.hg19)
    coding <- predictCoding(vcf, txdb, seqSource=Hsapiens)
    coding[5:9]

Using variant rs114264124 as an example, we see varAllele A has been substituted into the refCodon CGG to produce varCodon CAG. The refCodon is the sequence of codons necessary to make the variant allele substitution and therefore often includes more nucleotides than indicated in the range (i.e. the range is 50302962, 50302962, width of 1). Notice it is the second position in the refCodon that has been substituted. This position in the codon, the position of substitution, corresponds to genomic position 50302962. This genomic position maps to position 698 in coding region-based coordinates and to triplet 233 in the protein. This is a non-synonymous coding variant where the amino acid has changed from R (Arg) to Q (Gln).

When the resulting varCodon is not a multiple of 3 it cannot be translated. The consequence is considered a frameshift and will be missing.

    coding[coding$CONSEQUENCE == "frameshift"]

SIFT and PolyPhen databases

From predictCoding we identified the amino acid coding changes for the non-synonymous variants. For this subset we can retrieve predictions of how damaging these coding changes may be. SIFT (Sorting Intolerant From Tolerant) and PolyPhen (Polymorphism Phenotyping) are methods that predict the impact of amino acid substitution on a human protein. The SIFT method uses sequence homology and the physical properties of amino acids to make predictions about protein function. PolyPhen uses sequence-based features and structural information characterizing the substitution to make predictions about the structure and function of the protein.

Collated predictions for specific dbSNP builds are available as downloads from the SIFT and PolyPhen web sites. These results have been packaged into SIFT.Hsapiens.dbSNP132.db and PolyPhen.Hapiens.dbSNP131.db and are designed to be searched by rsid. Variants that are in dbSNP can be searched with these database packages. When working with novel variants, SIFT and PolyPhen must be called directly. See references for home pages.

Identify the non-synonymous variants and obtain the rsids.

    nms <- names(coding)
    idx <- coding$CONSEQUENCE == "nonsynonymous"
    nonsyn <- coding[idx]
    names(nonsyn) <- nms[idx]
    rsids <- unique(names(nonsyn)[grep("rs", names(nonsyn), fixed=TRUE)])

Detailed descriptions of the database columns can be found with ?SIFTDbColumns and ?PolyPhenDbColumns. Variants in these databases often contain more than one row per variant. The variant may have been reported by multiple sources and therefore the source will differ as well as some of the other variables.

    library(SIFT.Hsapiens.dbSNP132)

    ## rsids in the package 
    head(keys(SIFT.Hsapiens.dbSNP132), 3)
    ## list available columns
    columns(SIFT.Hsapiens.dbSNP132)
    ## select a subset of columns
    ## a warning is thrown when a key is not found in the database
    subst <- c("RSID", "PREDICTION", "SCORE", "AACHANGE", "PROTEINID")
    sift <- select(SIFT.Hsapiens.dbSNP132, keys=rsids, cols=subst)
    head(sift, 3)

PolyPhen provides predictions using two different training datasets and has considerable information about 3D protein structure. See ?PolyPhenDbColumns or the PolyPhen web site listed in the references for more details.

UPSCb/RnaSeqTutorial documentation built on Nov. 24, 2020, 12:40 a.m.