Changes between Version 5 and Version 6 of DataConcordance


Ignore:
Timestamp:
Apr 21, 2011 4:09:39 PM (13 years ago)
Author:
laurent
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • DataConcordance

    v5 v6  
    1 = Compare Sequence Data With Genotype Chip =
    2 Status: Alpha Authors: Patrick Deelen, Morris Swertz
     1= Data Concordance =
    32
    4 Tool to compare sequence data with data from genotyping chips.
    5 == Features ==
    6  * Overview of concordance per sample
    7  * Matrix of all sequence samples vs all genotype samples
    8    * This is used to detected sample mix-ups
    9  * Performance per SNP
     3This page summarizes the methods and results of the concordance checks between the following data sets:
     4* Groningen Immunochip data
     5* Groningen Sequence data
     6* BGI Sequence data
    107
    11 == Input ==
    12 The genotype data input is currently PED/MAP as generated by [http://pngu.mgh.harvard.edu/~purcell/plink/ plink].
     8= Methods & Tools =
    139
    14 Both QCed as non QCed data is supported. The following sequence file formates are supported
     10== File Types ==
     11All data sets were either generated or converted to VCF files aligned on the build Hg19 of the Human Reference Genome:
     12* See [[GoNL_Immunochip_Data_Preparation]] about how the Immunochip data was processed
     13* See [https://www.broad.harvard.edu/gsa/wiki/index.php/LiftOverVCF.pl GATK LiftOverVCF] about how to liftover a VCF file from one reference to another
    1514
    16  * VCF
    17  * Q20 (as provided by BGI)
    18  * CNS
     15== Concordance calculation using [http://vcftools.sourceforge.net/ VCFTools] ==
     16To calculate the concordance between the different files, [http://vcftools.sourceforge.net/ VCFTools] was used. More specifically:
     17<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>
     18This 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.
    1919
    20 == Future ==
    21  * Plots to give more insight in SNP performance
     20== Concordance aggregation using home-made scripts ==
     21The 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.
     22
     23=== vcftools-diff_site-concordance.pl ===
     24As its name suggests, this script runs over the individual .diff.site files produced by VCFTools and aggregate their information. The following features are available:
     25* Per-major report
     26** SNP filtering
     27** Report as plain text or tab-delimited
     28* SNP-major report
     29** SNP filtering
     30** Addition of MAF from a plink frq files
     31** Addition of SNP ID from a plink bim file
     32** Output of shared SNPs only
     33
     34=== vcftools-discordance-matrix.py ===
     35This script aggregates the discordance matrix files produced by vcftools into one.
     36
     37== Reporting using R scripts ==
     38For 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.
     39The following scripts are available:
     40* plot_shared_loci.R
     41** Plots the shared/unique loci in the two datasets per individual as a barplot
     42** Usage: Rscript plot_shared_loci.R <concordance_file> <out_plot.jpg> [name_dataset1] [name_dataset2]
     43* plot_geno_concordance.R
     44** Plots the genotype concordance between two datasets per individual as a barplot
     45** Usage: Rscript plot_geno_concordance.R <concordance_file> <out_plot.jpg> [plot_title]
     46* plot_discordance_matrix.R
     47** Plots the genotype discordance by "discordance type" (0/0 -> 0/1, 0/0 -> 1/1, 0/1 -> 0/0, etc.)
     48** Usage: Rscript plot_discordance_matrix.R <discordance_matrix_file> <out_plot.jpg> [dataset1_name] [dataset2_name] [show_concordant_data=FALSE] [<concordance_file>]
     49** Note:
     50*** 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).
     51
     52= Results - GoNL Pilot =
     53== Groningen / BGI ==
     54Datasets:
     55* Groningen
     56** Produced using Groningen pipeline on hg19
     57** 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.
     58* BGI
     59** Produced using BGI pipeline on b36, then lifted over to hg19
     60** SNPs filtered using standard BGI filter setup
     61
     62=== Loci Concordance ===
     63Below 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).
     64
     65[[File:bgi_groningen_loci_concordance.jpg|center]]
     66
     67After 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.
     68
     69=== Genotype Concordance ===
     70The following chart shows the genotype concordance on the shared SNPs between BGI and Groningen datasets.
     71
     72[[File:bgi_groningen_concordance.jpg|center]]
     73
     74Note: 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.
     75
     76== Groningen / Immunochip ==
     77Datasets:
     78* Groningen
     79** Produced using Groningen pipeline on hg19
     80** 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.
     81** Homozygous reference loci corresponding to the Immunochip were added to the dataset as well
     82* Immunochip
     83** ~165K loci after QC (both SNPs and homozygous reference)
     84*** SNP HWE p-val > 1e-3
     85*** SNP callrate > 99%
     86** Exported from Genome Studio, QC'ed and lifted over from hg18 to hg19
     87
     88=== Genotype Concordance ===
     89The following chart shows the genotype concordance on the 165K Immunochip loci left after QC.
     90
     91[[File:groningen_immunochip_concordance.jpg|center]]
     92
     93The 5 least concordant individuals can be explained as follow:
     94* A3b, A7b samples are contaminated
     95* 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.
     96Moreover, 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.
     97
     98
     99The 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.
     100
     101[[File:groningen_immunochip_discordance_matrix.jpg|center]]
     102
     103== BGI / Immunochip ==
     104Datasets:
     105* BGI
     106** Produced using BGI pipeline on b36, then lifted over to hg19
     107** SNPs filtered using standard BGI filter setup. Note that no homozygous reference locus is reported.
     108* Immunochip
     109** ~165K loci after QC (both SNPs and homozygous reference)
     110*** SNP HWE p-val > 1e-3
     111*** SNP callrate > 99%
     112** Exported from Genome Studio, QC'ed and lifted over from hg18 to hg19
     113
     114=== Genotype Concordance ===
     115The following chart shows the concordance between the 2 datasets over ~47K shared loci.
     116
     117[[File:bgi_immunochip_concordance.jpg|center]]
     118
     119Note that the 2 least concordant samples are explained as being contaminated.