Comparisons of isomiR patterns and classification performance using the rank-based MANOVA and 10-fold cross-validation

• Rank-based MANOVA compared the isomiR patterns between normal and tumor samples.
• Comparison of classification performance on three miRNA features was also conducted.
• The three features were “annotated sequence”, “miRNA total”, “multiple DE isomiRs”.
• The last performed best in predictive accuracy, with highest AUC and lowest AIC.

Next generation sequencing technology has identified a series of miRNA variants (named “isomiRs”), which might be associated with cancer progression. We provide a new strategy to reanalyze the miR-seq datasets through a view of the isomiR spectrum. Firstly, differentially expressed (DE) isomiRs were detected with the DESeq algorithm based on negative binomial distribution. Secondly, the rank-based MANOVA was adopted to compare the isomiR patterns between normal and tumor tissues. Moreover, a comprehensive survey on classification performance of three features was conducted, including the logistic regression, k-nearest neighbors and Random Forest. Finally, functional enrichment analysis was performed with the putative targets of specific isomiRs to elucidate their biological functions. Furthermore, the methods were applied to the downloaded miR-seq datasets of breast invasive carcinoma from TCGA. We found that the expression levels of multiple isomiRs derived from the same miRNA locus showed significant inconsistency between normal and tumor samples. In most cases, logistic regression with multiple DE isomiRs was superior to the others, with highest AUC and lowest AIC. Similarly, DE isomiRs performed best in the average accuracy of standard classifiers. Integrated targets were significantly enriched in some cancer-related pathways, including MAPK signaling pathway, and focal adhesion. Collectively, we could recommend the rank-based MANOVA for comparing different isomiR patterns, and further investigation on isomiRs needs to be considered in miRNA sequencing research.