Evaluation of clinically significant miRNAs level by machine learning approaches utilizing total transcriptome data

Analysis of the mechanisms underlying the occurrence and progression of cancer represents a key objective in contemporary clinical bioinformatics and molecular biology. Utilizing omics data, particularly transcriptomes, enables a detailed characterization of expression patterns and post-transcriptional regulation across various RNA types relative to the entire transcriptome. Here, we assembled a dataset comprising transcriptomic data from approximately 16.000 patients encompassing over 160 types of cancer. We employed state-of-the-art gradient boosting algorithms to discern intricate correlations in the expression levels of four clinically significant microRNAs, specifically hsa-mir-21, hsa-let-7a-1, hsa-let-7b, and hsa-let-7i, with the expression levels of the remaining 60.660 unique RNAs. Our analysis revealed a dependence of the expression levels of the studied microRNAs on the concentrations of several small nucleolar RNAs and regulatory long non-coding RNAs. Notably, the roles of these RNAs in the development of specific cancer types had been previously established through experimental evidence. Subsequent evaluation of the created database will facilitate the identification of a broader spectrum of overarching dependencies related to changes in the expression levels of various RNA classes in diverse cancers. In future it will make possible discovery of unique alterations specific to certain types of malignant transformations.