Abstract
The popularity of microRNA expression analyses is reflected by the existence of thousands of sRNA-seq studies in which matched total RNA-seq data are often unavailable. The lack of paired sequencing experiments limits the analysis of microRNA-gene regulatory networks. Here, we explore whether protein-coding gene expression can be quantified directly from transcript fragments present in sRNA-seq experiments. We analyze studies containing matched total RNA and small RNA from four human tissues and recover transcript fragments from the sRNA-seq data sets. We find that the expression levels of protein-coding gene transcripts derived from sRNA-seq data sets are comparable to those from total RNA-seq experiments (R (2) ranging from 0.33 to 0.76). Analyses across multiple tissues and species show similar correlations, indicating that the approach is applicable across organisms. We confirm that transcript half-life and the expression of housekeeping or highly abundant genes do not bias the results. Analysis of the expression of both microRNAs and coding genes from the same sRNA-seq experiments demonstrates that known microRNA-target interactions are, as expected, inversely correlated with the expression profiles of these microRNA-mRNA pairs. For a dual mRNA/miRNA profile, we recommend sequencing the ≥25 nucleotide fraction at 5 million or more reads. To confirm the utility of this approach, we apply our method to breast cancer sRNA-seq data sets lacking total RNA-seq data and achieve 75% recall and 64% accuracy comparing inferred coding gene expression with qPCR-validated targets. Our findings demonstrate that quantifying mRNA fragments from sRNA-seq experiments provides a reliable approach to investigate microRNA-mRNA interactions when total RNA-seq is unavailable.