Pitfalls and opportunities for applying latent variables in single-cell eQTL analyses

This repository contains the analysis code pipeline to generate PEER factors from pseudo-bulk data and perform eQTL association analysis as part of the manuscript "Pitfalls and opportunities for applying latent variables in single-cell eQTL analyses"

Scripts are listed by the order in the methods section of the manuscript:

1. Extract the whole OneK1K dataset from .RDS and subgroup into 14 cell types
2. Generate the pseudo-bulk mean matrix
3. Generate PEER factors (PFs) with 13 QC options
   • Extra information of runtime and nr of iterations
   • Make new covariate files
4. Run sensitivity test by MatrixeQTL
   • Merge results
   • Summarize and nr of eQTL and eGenes
5. Down-sampling analysis
6. Principal component analysis (PCA)
   • Generate PCs by PCAForQTL
   • Run eQTL sensitivity test adjusting PCs 0-50 (similar to PFs)
7. Main figures and supplementary figures
   • Include the analysis for identifying the optimal number of latent variables by comparing automatic elbow detection method and our local greedy method.

All code is also available on Github: https://github.com/powellgenomicslab/PEER_factors or https://github.com/anglixue/PEER_factors