Classification of microRNA precursors using reduced features of dinucleotide repeats in cattle (Bos Taurus)

The latest major advances in transcriptomics technologies, especially next-generation sequencing technologies and advanced bioinformatics tools, allows deeper exploration of messenger RNAs (mRNAs) and non-coding RNAs (ncRNAs), including miRNAs. These technologies have offered important chance for a deeper study of miRNA association in farm animal diseases, as well as livestock productivity and welfare. Since the discovery of lin‑4 and let‑7, many microRNAs have been identified in farm animal species and deposited in miRNA databases. miRNA can be used as biomarkers in the context of farm animal disease diagnostics, prediction, and therapeutic purposes, for the management of livestock diseases. By the sequencing of Bos Taurus (cattle) genome, we have an opportunity to discover novel miRNAs in this species. However, the experimental determination of miRNA sequence and structure is both expensive and time-consuming, therefore, computational and machine learning-based approaches have been adopted to predict novel microRNAs in the Bos Taurus (cattle) genome.

Finding an accurate method for Identification of miRNA molecules can help for understanding of regulatory processes. Currently, computational methods based on learning algorithms have been extensively applied for miRNA prediction.Inspired by the work of predecessors, we proposed an improved computational model based on random forest (RF) for identifying real miRNA precursor sequences (pre-miRNAs). First, the occurrence frequencies of the dinucleotide of pre-miRNAs genes, and the percentage of G+C content were calculated. The observed dinucleotide composition was calculated as the structural features of the sequence composition for each miRNA gene. A total of cattle (Bos Taurus) dinucleotide compositions with their genomic G+C contents for 1064 genes encoding miRNA and non-miRNA sequences were calculated. In the next step two classification models based on machine learning approach were trained to identify real and pseudo bovine pre-miRNAs. One set of 17 optimized features related to sequence structures were used to train the models. These models were trained and validated with 10-fold cross validation method.

Our goal was to investigate the predictive performance of RNA features in distinguishing pre-miRNAs from pseudo hairpins. Our model achieved 99% precision, and 97.9% MCC using Bos Taurus datasets.

Computational methods of Artificial intelligence can detect novel potential miRNAs in the bovine genome, some of which to have previously undetected in this genome. As a result, it seems necessary to use computational methods to identify these regulatory RNAs in livestock for breeding purposes. Our discoveries support that dinucleotide features will be beneficial to achieve the highest accuracies for miRNA sequences prediction.