Normalization Methods on Single-Cell RNA-Seq Data and Metagenomics Data

Abstract

DNA and RNA sequencing uncover the genomes and transcriptomes of cells, permitting greater understanding of the biological processes that fuel them and their relationship to one another. Advances in sequencing technology have expanded such studies to include both single-cell RNA sequencing, which analyzes cells individually instead of in bulk, and metagenomics, which describes the microbial composition of ecosystems found both in nature and in the human body. Normalization is required to accurately assess the genetic content of cells, adjusting for uneven sample sizes, stochasticity due to low input material, and dropout from uneven RNA amplification, all of which obscure the ground truth of a sample. This dissertation presents two novel normalization methods: one for single-cell RNA sequencing and one for metagenomics sequencing. It opens with a survey of normalization methods on single-cell RNA-seq data to provide ample background on common practices among existing approaches. Weighted Between Groups Normalization (WeBe) is proposed to normalize single-cell RNA sequencing data by utilizing external spike-in RNAs to establish relationships both within and between cell conditions/groups/types; 2-Stage Scaling Normalization (2SS) is designed for metagenomics sequencing data, first normalizing within conditions before identifying a set of stable features across conditions, which are used for across conditions normalization. Simulation studies and real data analysis demonstrate the effectiveness of each new method.