Aligning the Aligners: Comparison of RNA Sequencing Data Alignment and Gene Expression Quantification Tools for Clinical Breast Cancer Research

Abstract

The rapid expansion of transcriptomics and affordability of next-generation sequencing (NGS) technologies generate rocketing amounts of gene expression data across biology and medicine, including cancer research. Concomitantly, many bioinformatics tools were developed to streamline gene expression and quantification. We tested the concordance of NGS RNA sequencing (RNA-seq) analysis outcomes between two predominant programs for read alignment, HISAT2, and STAR, and two most popular programs for quantifying gene expression in NGS experiments, edgeR and DESeq2, using RNA-seq data from breast cancer progression series, which include histologically confirmed normal, early neoplasia, ductal carcinoma in situ and infiltrating ductal carcinoma samples microdissected from formalin fixed, paraffin embedded (FFPE) breast tissue blocks. We identified significant differences in aligners’ performance: HISAT2 was prone to misalign reads to retrogene genomic loci, STAR generated more precise alignments, especially for early neoplasia samples. edgeR and DESeq2 produced similar lists of differentially expressed genes, with edgeR producing more conservative, though shorter, lists of genes. Gene Ontology (GO) enrichment analysis revealed no skewness in significant GO terms identified among differentially expressed genes by edgeR versus DESeq2. As transcriptomics of FFPE samples becomes a vanguard of precision medicine, choice of bioinformatics tools becomes critical for clinical research. Our results indicate that STAR and edgeR are well-suited tools for differential gene expression analysis from FFPE samples.

The Transcriptomic Toolbox: Resources for Interpreting Large Gene Expression Data within a Precision Medicine Context for Metabolic Disease Atherosclerosis

“…The most common software platforms to align RNA sequence to a reference genome are TopHat [68], HiSAT [69], and STAR [70]. These platforms differ with respect to speed, memory usage, and their algorithms for handling pseudogenes [48], base and splice junction alignment precision, with HiSAT and STAR optimized to process large datasets (>10 8 reads), whereas TopHat is designed for smaller datasets (<2 × 10 7 reads).…”

“…In eukaryotic organisms, if only protein coding genes are of interest, poly(A) selection yields greater accuracy of transcript quantification [47]. These issues are particularly critical for clinical samples from patients, which are routinely processed as formalin-fixed, paraffin-embedded (FFPE) samples, which adversely impact the quality of RNA and subsequent alignment to pseudogenes [48]. Fortuitously, side-by-side comparison of FFPE and flash-frozen samples shows a great degree of concordance (e.g., r 2 in the range of 0.90–0.97 in recent studies [49,50]), proving RNAseq is a viable tool for gene quantification in clinical settings.…”

XPRESSyourself: Enhancing, Standardizing, and Automating Ribosome Profiling Computational Analyses Yields Improved Insight into Data

“…The first step of RNA-Seq alignment is curating an organism reference to which the alignment software will map sequence reads. XPRESSpipe uses STAR [53] for mapping reads as it has been shown consistently to be the best performing RNA-Seq read aligner for the majority of cases [54,55]. The appropriate reference files are automatically curated by providing the appropriate GTF file saved as transcripts.gtf and the directory path to the genomic FASTA file(s).…”

“…Reads are aligned to the reference genome with STAR, which, despite being more memory-intensive, is one of the fastest and most accurate sequence alignment options currently available [53][54][55]. XPRESSpipe is capable of performing a single-pass, splice-aware, GTF-guided alignment or a two-pass alignment of reads wherein novel splice junctions are determined and built into the genome index, followed by alignment of reads using the updated index.…”

“…Compared to the count data made available with the original manuscript, when XPRESSpipe-processed samples were thresholded as in the original published count data, samples showed generally comparable read counts per gene between the two analytical regimes (Spearman ρ values 0.937-0.951) ( Figure 2B shows representative plots; Figure S3B shows all comparisons). This is in spite of the fact that the methods section of the original publication employed software that was current at the time but is now outdated, such as TopHat2 [77], which has a documented higher false-positive alignment rate, generally lower recall, and lower precision at correctly aligning multi-mapping reads compared to STAR [53][54][55]. Many of the genes over-represented in the original count data as compared to data processed by XPRESSpipe appear to be due to the over-estimation of pseudogenes or other gene paralogs.…”