Statistical Evaluation and Selection of Depth Normalization in Small RNA Sequencing

About this Class

Deep sequencing has emerged as the primary tool for transcriptome profiling in cancer research. Like other high-throughput profiling technologies, sequencing is susceptible to systematic non-biological artifacts stemming from inconsistent experimental handling. A critical initial step in sequencing data analysis is to “normalize” sequencing depth to enhance data comparability across samples. While numerous normalization methods have been proposed, there is no systematically superior method, and different methods may yield divergent analysis results. This underscores the urgent need for a realistic and objective performance evaluation, particularly in the context of small RNA sequencing, along with a statistically principled approach to guide the method selection for a given dataset.

To address these needs, we have developed (1) benchmark data and computational tools for the comprehensive evaluation of depth normalization methods in microRNA sequencing and (2) a data-driven and biology-motivated approach for the objective selection of a suitable method tailored to the dataset at hand. We assessed the performance of the latter approach using our benchmark data and applied it to data in the Cancer Genome Atlas.

The evaluation tools and selection approach are implemented in R packages named PRECISION.seq and DANA, both of which are freely available for download on GitHub.