MicroRNAs are present in biofluids and are dysregulated in many diseases, suggesting they could be a valuable source of non-invasive biomarkers. However, lack of reliable detection methods and substantial variation in microRNA expression between people have limited this potential, particularly in biofluid samples which typically have low overall microRNA yields and are thus vulnerable to bias related to sample collection, purification and storage. To overcome these limitations, we proposed to expand our search beyond the approximately 2,000 known microRNAs in humans by considering disease-specific microRNA length isoforms (isomiRs).
In this study, we analysed small-RNAÂ sequencing data from publicly available datasets of influenza A and COVID-19 infections, among others. Using bioinformatic analyses and supervised machine learning, we evaluated the ability of pairs of isomiRs derived from the same microRNA locus to classify samples based on disease state. Our findings demonstrate that the ratio of expression of a single pair of isomiRs can successfully distinguish influenza A-infected monocytes from uninfected with over 90% accuracy. Furthermore, a distinct isoform pair can predict COVID-19 infection status in patients with >80% accuracy, across samples from two independent datasets. Critically, our use of pairs of isoforms derived from the same locus presents a new approach to microRNA normalisation by measuring post-transcriptional regulation of microRNA processing into separate isoforms, rather than tracking transcriptional variation. Our data suggest that this novel approach can offer greater resilience to differences in baseline microRNA expression and provides a potential solution to the normalisation issues that have hindered the development of microRNA biomarkers to date.