PRECISION.seq: An R Package for Benchmarking Depth Normalization in microRNA Sequencing.

Jian Zou,Li-Xuan Qin,Yannick Düren

doi:10.3389/fgene.2021.823431

Abstract

We present a new R package PRECISION.seq for assessing the performance of depth normalization in microRNA sequencing data. It provides a pair of microRNA sequencing data sets for the same set of tumor samples, additional pairs of data sets simulated by re-sampling under various patterns of differential expression, and a collection of numerical and graphical tools for assessing the performance of normalization methods. Users can easily assess their chosen normalization method and compare its performance to nine methods already included in the package. PRECISION.seq enables an objective and systematic evaluation of normalization methods in microRNA sequencing using realistically distributed and robustly benchmarked data under a wide range of differential expression patterns. To our best knowledge, this is the first such tool available. The data sets and source code of the R package can be found at https://github.com/LXQin/PRECISION.seq.

Highlights

Depth normalization is a critical preprocessing step for accurate and reproducible analysis of transcriptomic sequencing data (Bullard et al, 2010)
JZ, YD, and LXQ contributed to conception and design of the study
JZ and LXQ wrote the first draft of the manuscript

Summary

INTRODUCTION

Depth normalization is a critical preprocessing step for accurate and reproducible analysis of transcriptomic sequencing data (Bullard et al, 2010). Methods for depth normalization have been newly proposed or repurposed from normalization methods previously developed for microarray data (Dillies et al, 2013) Their performances have been evaluated primarily for RNA sequencing data and a thorough assessment is still in need for microRNAs (miRNAs), a class of small RNAs regulating gene expression and closely linked to carcinogenesis, which tend to be expressed in a tissue-specific manner with a small number of markers abundantly expressed (Dillies et al, 2013; Maza et al, 2013). P90 shows mediocre FDR and poor FNR, generally comparable to Total Count and Quantile Normalization and slightly worse than Upper Quartile and Median (Figure 3)

SUMMARY

ETHICS STATEMENT