Process genetic data

MODULE STATUS

Developer: Developer's team

Scripts status: Ready

Prerequisite scripts: Script 0, Script 01a

Data upload method: automated upload to sftp bristol

Issue report to: Haotian Tang haotian.tang@bristol.ac.uk

Overview: Processing genetic data

Once the imputed genetic data is in the correct format and in the correct location, the following tasks need to be performed:

• Liftover to hg38 if hg19
• Extract relevant IDs
• HWE and MAF QC
• Genotype missingness QC
• Remove sexcheck outliers if chr X is available
• Re-code SNP IDs (chr:position_A1_A2 (ascii sorted order))
• Set the maximum allele length to be 70
• Remove duplicate SNPs
• Remove poorly imputed SNPs
• Construct kinship matrices
• Create a pedigree matrix for 'related' samples
• Remove any cryptic relatedness for 'unrelated' data
• Calculate principal components
• Detect outliers in SNP loadings
• Check PC outliers and generate PCA plots
• Create a global PCA plot using PCA loadings
• Check allele frequencies between genetic data and ancestry-specific reference panel.

To do this, run the following script:

./01b-process_genetic_data.sh -c /path/to/config/file

Note that for 'unrelated' samples, we are generating PCs by removing long LD tracts, extracting HapMap3 SNPs, LD pruning, and then calculating PCs on the data after cryptic relateds have been removed. For 'related' samples, we are using a general method implemented in the Rpackage 'GENESIS' for estimating PCs that account for any relatedness. This involves estimating a kinship matrix that is robust to population structure and relatedness (Manichaikil et al. (2010)), estimating principal components on an unrelated subset, and then projecting principal components onto the related subset based on estimated kinships (Conomos et al. (2015)).

After you have run this script, please check:

cd results/01

IMPORTANT PLEASE CHECK YOUR RESULTS:

• The pcaplot.pdf is a PCA plot on the pruned genetic data. The blue dashed lines define the SD threshold. You can check the number of genetic outliers that are recorded by:

grep outlier ./results/01/logs_b/log.txt

If you think the outlier threshold should be adjusted, you can change the sd threshold pca_sd in the config file. If you change the threshold, you need to rerun 01b-process_genetic_data.sh.

• If you find that your PCA plot contains a V-shaped or poorly clustered distribution of samples when plotting PC1 vs. PC2. Please check the global PCA plot, which uses PCA loadings from global biobank.

• The easyQC_topmed.rep shows the number of SNPs that are flipped because they were on the wrong strand as compared to one of the TopMed reference panels ("AlleleChange"). Please find an example here: (https://github.com/genetics-of-dna-methylation-consortium/godmc_phase2/wiki/images//easyQC_hrc.rep).

• The easyQC_topmed.rep file also shows you the number of SNPs that are going to be removed. eg, allele mismatches ("AlleleMismatch"). It also gives a list of SNPs that have a discrepant allele frequency (>0.2) as compared to TopMED ("AFCHECK.numOutlier").

• The easyQC.multi.AFCHECK.png plot shows a comparison between allele frequencies of your study as compared to TopMED imputed 1000G samples. Only SNPs with allele frequency differences greater than 0.2 from the reference are plotted. Please read the link for an explanation of this plot here. Please find two examples below. Currently, SNVs highlighted in red are NOT removed from the analysis. If all allele frequency differences between your data and the reference panel are ≤ 0.2, the plot will be empty.

  easyQC.multi.AFCHECK.png.

• Please make sure that there is no data.smiss.gz file in your results folder.

Second-round global PCA projection (for cohorts that have completed QC stage 01)

After cohorts submit their first-round QC results (see Upload to Bristol SFTP), the DEEP coordinating team rebuilds the reference PCA using SNPs shared across cohorts and the reference. When requested by the coordinating team, cohorts should follow the Second-round global PCA projection instructions using their cleaned genotype data from this step.