Data Concordance

This page summarizes the methods and results of the concordance checks between the following data sets:

• Groningen Immunochip data
• Groningen Sequence data
• BGI Sequence data

Methods & Tools

File Types

All data sets were either generated or converted to VCF files aligned on the build Hg19 of the Human Reference Genome:

• See GoNL_Immunochip_Data_Preparation about how the Immunochip data was processed
• See ​GATK LiftOverVCF about how to liftover a VCF file from one reference to another

Concordance calculation using ​VCFTools

To calculate the concordance between the different files, ​VCFTools was used. More specifically: <pre>vcftools --vcf /data/lfrancioli/immunochip/hg19/GvNL.hg19.final.vcf --indv ${sample} --diff /data/lfrancioli/results/pilot/${sample}.human_g1k_v37.immuno.vcf --diff-site-discordance --diff-indv-discordance --diff-discordance-matrix</pre> This computes the concordance per file, site and individual as well as a discordance matrix. This was applied on a sample level so only the file, site and discordance matrix where actually used.

Concordance aggregation using home-made scripts

The output of the VCFTools being per sample, they are useful for single individual QC but not for population level QC. A few scripts were developed in order to easily aggregate the data over a selection of samples files.

vcftools-diff_site-concordance.pl

As its name suggests, this script runs over the individual .diff.site files produced by VCFTools and aggregate their information. The following features are available:

• Per-major report

SNP filtering Report as plain text or tab-delimited

• SNP-major report

SNP filtering Addition of MAF from a plink frq files Addition of SNP ID from a plink bim file Output of shared SNPs only

vcftools-discordance-matrix.py

This script aggregates the discordance matrix files produced by vcftools into one.

Reporting using R scripts

For reporting purpose, R scripts were created. These scripts all take files created using vcftools-diff_site-condordance.pl or vcftools-discordance-matrix.py as input. The following scripts are available:

• plot_shared_loci.R

Plots the shared/unique loci in the two datasets per individual as a barplot Usage: Rscript plot_shared_loci.R <concordance_file> <out_plot.jpg> [name_dataset1] [name_dataset2]

• plot_geno_concordance.R

Plots the genotype concordance between two datasets per individual as a barplot Usage: Rscript plot_geno_concordance.R <concordance_file> <out_plot.jpg> [plot_title]

• plot_discordance_matrix.R

Plots the genotype discordance by "discordance type" (0/0 -> 0/1, 0/0 -> 1/1, 0/1 -> 0/0, etc.) Usage: Rscript plot_discordance_matrix.R <discordance_matrix_file> <out_plot.jpg> [dataset1_name] [dataset2_name] [show_concordant_data=FALSE] [<concordance_file>] Note: * The last optional argument is a concordance file over the same data to plot as 'unknown' all loci that were not captured by the concordance matrix since the alleles were not exact matches (e.g. if one of the allele was monomorphic in one set).

Results - GoNL Pilot

Groningen / BGI

Datasets:

• Groningen

Produced using Groningen pipeline on hg19 SNPs were only filtered for quality > Q10 (to avoid extreme numbers in VCF). It is considered an 'unfiltered' set and is expected to contain many false positives, however sensitivity should be excellent as any SNP not reported at this point will not be detected further in the pipeline.

• BGI

Produced using BGI pipeline on b36, then lifted over to hg19 SNPs filtered using standard BGI filter setup

Loci Concordance

Below is a chart showing the shared and unique SNPs in the two datasets regardless of their genotypes. As expected, the vast majority of the SNPs are shared between the datasets, a relatively high number of SNPs are only found in Groningen (amongst them a majority of unfiltered false positives) and a small number of SNPs unique to the BGI dataset (to be investigated).

[[File:bgi_groningen_loci_concordance.jpg|center]]

After investigation, the three least concordant individuals encountered a problem while processing one of their lanes, thus leading to 2/3 of the normal coverage. The figures should be updated when the lanes have been processed and these individuals corrected.

Genotype Concordance

The following chart shows the genotype concordance on the shared SNPs between BGI and Groningen datasets.

[[File:bgi_groningen_concordance.jpg|center]]

Note: The chart above does not take sex chromosomes into account as an artifact introduced by the way the Y-chrom was mapped by BGI was showing all males as completely discordant over the sex chromosomes.

Groningen / Immunochip

Datasets:

• Groningen

Produced using Groningen pipeline on hg19 SNPs were only filtered for quality > Q10 (to avoid extreme numbers in VCF). It is considered an 'unfiltered' set and is expected to contain many false positives, however sensitivity should be excellent as any SNP not reported at this point will not be detected further in the pipeline. Homozygous reference loci corresponding to the Immunochip were added to the dataset as well

• Immunochip

~165K loci after QC (both SNPs and homozygous reference) * SNP HWE p-val > 1e-3 * SNP callrate > 99% Exported from Genome Studio, QC'ed and lifted over from hg18 to hg19

Genotype Concordance

The following chart shows the genotype concordance on the 165K Immunochip loci left after QC.

[[File:groningen_immunochip_concordance.jpg|center]]

The 5 least concordant individuals can be explained as follow:

• A3b, A7b samples are contaminated
• A8a,A8c,R5A encountered a problem while processing one of their lanes, thus leading to 2/3 of the normal coverage. The figures should be updated when the lanes have been processed and these individuals corrected.

Moreover, if we exclude the individuals above and filter for sites that are commonly reported by the BGI and Groningen pipelines, the concordance reaches 99.029% on average.

The graph below shows a preliminary analysis of the "types" of discordance observed. An important caveat has to be taken into account: VCFTools only reports sites where the alleles perfectly match. This means that all monomorphic sites in one dataset that are polymorphic in the other will not appear. This was especially problematic since we compared each sequenced sample separately against the whole Immunochip dataset. As a result almost all homozygous reference sites in the sequence data were not reported by VCFTools. All the discordant sites that did not have perfectly matching alleles are reported below as 'unknown' as it has yet to be investigated what discordance "type" they belong to.

[[File:groningen_immunochip_discordance_matrix.jpg|center]]

BGI / Immunochip

Datasets:

• BGI

Produced using BGI pipeline on b36, then lifted over to hg19 SNPs filtered using standard BGI filter setup. Note that no homozygous reference locus is reported.

• Immunochip

~165K loci after QC (both SNPs and homozygous reference) * SNP HWE p-val > 1e-3 * SNP callrate > 99% Exported from Genome Studio, QC'ed and lifted over from hg18 to hg19

Genotype Concordance

The following chart shows the concordance between the 2 datasets over ~47K shared loci.

[[File:bgi_immunochip_concordance.jpg|center]]

Note that the 2 least concordant samples are explained as being contaminated.