prashanthi-ravichandran / sc_eqtl

Single cell eQTL calling

Single cell eQTL

• First run Process_per_cell_type.ipynb which reads in the raw data from CLUESImmVar_nonorm.V6.h5ad and splits the samples based on the provided cell-types and retains only those samples that belong to individuals with Lupus. Then the following QC steps are performed

  • Remove cells with fewer than 200 genes expressed
  • Remove genes expressed in fewer than 3 cells
  • Remove cells with % mito reads greater than 0.05
  • Remove cells with more than 2500 genes expressed
  • We then subset to autosomal, protein-coding genes, and recalculate the QC metrics, the eQTL sepecific filters are then performed which include
  • Remove cells with fewer than 400 expressed genes
  • Remove genes expressed in fewer than 5% of all cells This script also computes the log1p normalized and scaled single cell data

• Run processing_pseudobulk.R to compute

  • pseudobulk by summing counts from all cells belonging to a specific individual
  • compute expression PCs
  • compute genotype PCs
  • create covariates.txt file for downstream eQTL calling

• Run call_eQTLs.R for each combination of cell type and chromosome. This uses matrix-eQTL to call cis-eQTLs by testing all SNPs that are within a 1Mb distance from the gene TSS and have a MAF > 0.05. Further, for each gene, we select the gene-SNP pair that has the lowest p-value after Bonferroni correction within each gene and then perform BH correction across genes to obtain significant eGene-eSNP pairs at FDR of 0.05.