The impact of read length on quantification of differentially expressed genes and splice junction detection

Abstract

Background: The initial next-generation sequencing technologies produced reads of 25 or 36 bp, and only from a single-end of the library sequence. Currently, it is possible to reliably produce 300 bp paired-end sequences for RNA expression analysis. While read lengths have consistently increased, people have assumed that longer reads are more informative and that paired-end reads produce better results than single-end reads. We used paired-end 101 bp reads and trimmed them to simulate different read lengths, and also separated the pairs to produce single-end reads. For each read length and paired status, we evaluated differential expression levels between two standard samples and compared the results to those obtained by qPCR.

Results: We found that, with the exception of 25 bp reads, there is little difference for the detection of differential expression regardless of the read length. Once single-end reads are at a length of 50 bp, the results do not change substantially for any level up to, and including, 100 bp paired-end. However, splice junction detection significantly improves as the read length increases with 100 bp paired-end showing the best performance. We performed the same analysis on two ENCODE samples and found consistent results confirming that our conclusions have broad application.

Conclusions: A researcher could save substantial resources by using 50 bp single-end reads for differential expression analysis instead of using longer reads. However, splicing detection is unquestionably improved by paired-end and longer reads. Therefore, an appropriate read length should be used based on the final goal of the study.

Fig. 1

Mapping statistics of all samples and read lengths. a Each sample and read length is plotted with its respective percentage of uniquely mapped, multi-mapped and unmapped reads. The 25 bp reads have the lowest percent of uniquely mapped reads across all samples. Single-end reads also have higher percentage of multi-mapped reads and slightly lower percentage of uniquely mapped reads. b The number of splice junctions detected for each sample is plotted. The 25 bp samples detected the least number of junctions and single-end reads detected significantly fewer junctions overall than paired-end reads. The error bars represent the highest and lowest number of splice junctions detected across replicates

Fig. 2

Determination of differentially expressed genes according to read length and differential expression method. a Single-end read samples. The number of orphan genes (read-length-specific genes) in the overlap of the top 200 genes sorted by -Log2-based fold change (-Log2FC; down-regulated), +Log2FC (up-regulated) and p value. b Paired-end read samples. The number of orphan genes (read-length-specific genes) in the overlap of the top 200 genes sorted by -Log2FC, +Log2FC and p value. c Single-end read samples. The plot shows the agreement for the top 200 differentially expressed genes by different read length. d Paired-end read samples. The plot shows the agreement for the top 200 differentially expressed genes by different read length

Fig. 3

Comparison of previously reported qPCR results with our DEG results. a Pearson correlation between Log2FC of genes according to various differential expression methods and qPCR. Single-end 25 bp reads have the worst correlation when using DESeq and EdgeR. b Root mean square deviation (RMSD) between Log2FC and qPCR. Single-end 25 bp reads give results farthest from the true values. c Common genes between the top 200 genes identified by various differential expression methods and qPCR sorted by +Log2FC. d Same as (c), except sorted by –Log2FC. The overlap of common genes improves as read length increases. However, the gain is not significant for reads >50 bp for paired-end and >75 bp for single-end reads

Fig. 4

Splice junction agreement and inter-replicate reproducibility. a Number of known and novel junctions that were orphans (read-length-specific junctions) according to the read length in a specific sample. b Percentage of known junctions that were common when paired-end and single-end samples of the same read length were intersected. Error bars represent the range of all the replicates

Fig. 5

Common splice junctions detected with different read lengths. a Percentage of splice junctions detected with all four read lengths. b Percentage of splice junctions detected with all read lengths except 25 bp