Feature importance analysis in guide strand identification of microRNAs

► In this study, we constructed random forest models based on local sequence-structure feature to identify miRNA in four species. ► We achieved the accuracies of 86.51% for Homo sapiens, 81.66% for Ornithorhynchus anatinus, 82.33% for Mus musculus and 85.71% for Schmidtea mediterranea, respectively. ► We implemented the importance analysis of all feature elements by using the conditional feature importance strategy. ► We concluded that the GC base pair is crucial to the selection of guide strand.