iCellR: Combined Coverage Correction and Principal Component Alignment for Batch Alignment in Single-Cell Sequencing Analysis

Abstract

Under-sampling RNA molecules and low-coverage sequencing in some single cell sequencing technologies introduce zero counts (also known as drop-outs) into the expression matrices. This issue may complicate the processes of dimensionality reduction and clustering, often forcing distinct cell types to falsely resemble one another, while eliminating subtle, but important differences. Considering the wide range in drop-out rates from different sequencing technologies, it can also affect the analysis at the time of batch/sample alignment and other downstream analyses. Therefore, generating an additional harmonized gene expression matrix is important. To address this, we introduce two separate batch alignment methods: Combined Coverage Correction Alignment (CCCA) and Combined Principal Component Alignment (CPCA). The first method uses a coverage correction approach (analogous to imputation) in a combined or joint fashion between multiple samples for batch alignment, while also correcting for drop-outs in a harmonious way. The second method (CPCA) skips the coverage correction step and uses k nearest neighbors (KNN) for aligning the PCs from the nearest neighboring cells in multiple samples. Our results of nine scRNA-seq PBMC samples from different batches and technologies shows the effectiveness of both these methods. All of our algorithms are implemented in R, deposited into CRAN, and available in the iCellR package.