RNA-seq data science: From raw data to effective interpretation

Abstract

RNA sequencing (RNA-seq) has become an exemplary technology in modern biology and clinical science. Its immense popularity is due in large part to the continuous efforts of the bioinformatics community to develop accurate and scalable computational tools to analyze the enormous amounts of transcriptomic data that it produces. RNA-seq analysis enables genes and their corresponding transcripts to be probed for a variety of purposes, such as detecting novel exons or whole transcripts, assessing expression of genes and alternative transcripts, and studying alternative splicing structure. It can be a challenge, however, to obtain meaningful biological signals from raw RNA-seq data because of the enormous scale of the data as well as the inherent limitations of different sequencing technologies, such as amplification bias or biases of library preparation. The need to overcome these technical challenges has pushed the rapid development of novel computational tools, which have evolved and diversified in accordance with technological advancements, leading to the current myriad of RNA-seq tools. These tools, combined with the diverse computational skill sets of biomedical researchers, help to unlock the full potential of RNA-seq. The purpose of this review is to explain basic concepts in the computational analysis of RNA-seq data and define discipline-specific jargon.

Keywords: RNA sequencing; bioinformatics; differential gene expression; high throughput sequencing; read alignment; transcriptome quantification.

FIGURE 1

Overview of RNA-seq. RNA-seq is a process of creating short sequencing reads from RNA molecules. The steps consist of first converting the RNA (A) into cDNA (B), then (optionally) amplifying the cDNA by PCR (C), and finally fragmenting the cDNA into short pieces (known as fragments). After the sequencing library (D) is prepared, the fragments are used as input for next-generation sequencing (E). The resulting sequence reads contained in FASTQ files are then aligned to a reference sequence (F). Modern high-throughput sequencing machines can generate up to 150 million reads per run. The reference sequence, shown as a pink line, is known. The goal of the alignment is to find the locus in the reference sequence with the greatest match to each read. Reads are shown to align to the specific positions/locations and these mapped locations are recorded.

FIGURE 2

Alternative splicing and RNA-seq technologies. The flow of genetic information begins with DNA, which consists of introns and exons. DNA is transcribed into pre-mRNA and then further processed into mature mRNA by splicing out the introns and leaving the exons glued together. The mRNA is then translated into a protein. Transcripts with different arrangements of exons can be formed in a process called alternative splicing or exon skipping. An RNA-seq read is a short sequence sampled from a transcript. Reads are generated using sequencing technologies such as (A) the Illumina platform, which produces short reads, and the (B) Nanopore and PacBio platforms, which produce long reads. The figure depicts two scenarios in which uniquely mapped reads are aligned to a reference transcriptome (C) and a reference genome (D), respectively. A few of the reads are multicolored, indicating that when aligned, they span across an exon-exon junction. Some of the shorter reads (single-colored) are aligned only to a single exon and do not span across the junction. TSS, transcription start site; TES, transcription end site.

FIGURE 3

Measuring allele-specific expression with RNA-seq. RNA-seq can be used to generate allele-specific expression (ASE) data for genes with a heterozygous single-nucleotide polymorphism in the transcribed region (aseSNP). The aseSNP enables sequencing reads to be mapped to the haplotype from which they originate. Imbalance in ASE data is a functional indicator of a cis-regulatory difference between the two haplotypes that is driven by heterozygous regulatory variants. Data from multiple aseSNPs can be aggregated to improve ASE data quality. The non-coding regulatory variant depicted here has two alleles inducing higher (H) and lower (L) expression of the target gene.