Fig. 1
From: Multiple freeze-thaw cycles lead to a loss of consistency in poly(A)-enriched RNA sequencing
3′ Bias is Exacerbated in Frozen, Poly(A)-enriched Samples Across Multiple Studies. a Demonstration for determining median coverage percentile (red vertical line). When coverage is unbiased, reads (yellow) are distributed throughout the entire body of the transcript (green). In the absence of read bias and observing coverage as a function of the nucleotide percentile, we see that cumulative coverage along the transcript reaches 50% half-way through the gene body, at the 50th percentile nucleotide. In contrast, given a 3′ read bias, there is a shift in the distribution of reads and cumulative coverage reaches 50% at, for example, the 60th percentile nucleotide. This can be seen by the "rightward" shift in median coverage percentile towards the 3′ end of the transcript. In the right panel, gene coverage (y-axis) at the ith nucleotide percentile from 5′ to 3′ (x-axis) displayed for the unbiased and 3' biased transcripts. b Median coverage percentile was calculated for 237 blood tissue samples spanning 10 RNA-Seq datasets downloaded from SRA. Samples are stratified by sample handling (unfrozen or frozen) and library preparation (poly(A)-enrichment or ribosomal depletion). Read coverage distributions were compared using a two-sided, two-sample t-test with a Benjamini-Hochberg correction (* FDR ≤ 0.05, ** FDR ≤ 0.01, *** FDR ≤ 1e-3, **** FDR ≤ 1e-4). c Comparison of 5′ to 3′ bias ratio (y-axis) of samples from the TCGA and UNC tissue repositories (x-axis) between extraction methods (two-sample t-test). Querying human RNA samples listed in GEO from 2008 to 2018, and stratifying by those annotated as “frozen”, we observe (d) the number of samples prepared with poly(A)-enrichment or ribosomal depletion (x-axis), (e) the proportion of samples extracted using either method, and (f) the change in the number of samples over time