Experimental design, preprocessing, normalization and differential expression analysis of small RNA sequencing experiments

Deep sequencing technologies have revolutionized the field of genomics since their inception in 2000, when Lynx Therapeutics' Massively Parallel Signature Sequencing (MPSS; Lynx Therapeutics, Hayward, CA, USA) was described as a way to quantify messenger RNA (mRNA) populations [1]. MPSS allowed the parallel sequencing of 17- or 20-nucleotide (nt) signatures from hundreds of thousands of cloned RNA, but it has been made obsolete by newer systems enabling longer sequence reads with fewer biases. Next-generation sequencing has since been adapted to the study of a wide range of nucleic acid populations, including mRNA (RNA-seq) [2], small RNA (sRNA) [3], microRNA (miRNA)-directed mRNA cleavage sites (called parallel analysis of RNA ends (PARE), genome-wide mapping of uncapped transcripts (GMUCT) or degradome sequencing) [4-6], double-stranded RNA (dsRNA) [7,8], actively transcribing RNA (NET-seq) [9], translated mRNA [10], transcription factor DNA binding sites and histone modification sites (chromatin immunoprecipitation (ChIP)-seq) [11], methylated DNA (BS-seq) [12] and genomic DNA (DNA-seq) [13-15]. These applications vary with regard to the templates used, but they rely on the same sequencing technologies.Prior to high-throughput sequencing, DNA microarrays were the predominant method of genome-wide transcriptional analysis. Microarrays have been used to quantify the levels of both known and unknown mRNA, alternative splicing products, translated mRNA and miRNA, as well as to detect miRNA cleavage sites, transcription factor binding sites, single-nucleotide polymorphisms and deletions. Now, however, high-throughput sequencing is often favored over microarrays for such experiments because sequencing avoids several problems encountered in microarray experiments. First, unlike microarrays, sequencing approaches do not require knowledge of the genome a priori, enabling any organism to be easily studied. Second, sequencing is not dependent on hybridization. Microarray da