Comparing prioritized cell types from GWAS using single nucleus-RNA or -ATAC sequencing (in Alzheimer individuals and healthy controls)

Minor Research Project - Master Bioinformatics and Systems Biology, Vrije Universiteit Amsterdam

The main aim of the project is develop a pipeline to link scRNAseq and scATACseq data to GWAS summary statistics.

The input are two count matrices: Gene Expression matrix for scRNAseq and Peak intensity matrix for scATACseq from Morabito et al.2021. Features in the scRNAseq datasets are genes, in scATACseq are cis-candidate regulatory elements (cCREs). The Alzheimer Disease GWAS used is from Wightman et al. 2021.

preprocessing_rna.py and preprocessing_atac.py filter the intial datasets, divides Cases (AD) and Controls and averages the count matrix by grouping cells belonging to the same cell type.

Next, we used the EWCE package to create a specificity matrix (one for each sc technology dataset) for each feature in each cell type. Script: specificity.R.

specificity_analysis.py creates the Feature Sets for each cell type. In this scripts two methods are utilized to generate the Sets:

1. Top 10% most specific features: after filtering out features with zero value as specificity, we took the top 10% of most specific features for each cell type.
2. One-cell-type specific feature: we plotted the distribution of specificity scores for each feature to determine a threshold that could confidently identify features that are specific for one cell type only. For both scRNAseq and scATACseq features, a threshold of 0.52 was chosen to generate the set of specifically expressed/open features for each cell type.

LDSC is used to estimate the enrichment of cell types from GWAS summary statistics (partition heritability method). ldsc_analysis.sh includes:

• creation of snp set from baseline model (v2.2)
• annotation (feature to SNP)
• compute LD scores
• run LD regression (cell-type specific)