Towards a machine learning framework for predicting MicroRNAs involved in Candidiasis

Infections caused by Candida species present significant clinical challenges, leading to high morbidity and mortality rates. Emerging research suggests that microRNAs (miRNAs) may serve as valuable biomarkers for diagnosing and treating candidiasis, although the field is still developing. The scarcity of reported miRNAs associated with candidiasis stems from the complexities and costs of screening numerous potential biomarkers, making in silico prediction essential for identifying candidates for experimental validation. This study aims to enhance the understanding of miRNAs related to candidiasis using bioinformatics approaches, focusing on two primary objectives: analyzing existing experimental data and creating a novel bioinformatics tool using Machine Learning (ML) to predict candidiasis-related miRNAs. For this end, a dataset with 80 experimental miRNAs related to candidiasis was constructed through a systematic literature search and using bioinformatics databases such as miRBase, miRPathDB and miRDB. Notably, 63 of these miRNAs exhibited high cellular activity, with five (miR-17-3p, miR-222-3p, miR-133a, miR-132-5p, and miR-100) showing ideal characteristics for diagnostic biomarkers, highlighting their potential application. To support further identification of biomarkers, a ML-based tool is being developed, able of predicting miRNAs related to different types of candidiasis, also providing valuable insights as cellular localization and specificity. Our prediction hypothesis posits that if a gene is a target of candidiasis and also a target of a miRNA, that miRNA may be associated to candidiasis, suggesting that a higher number of shared targets correlates with a higher prediction score. As so, ML models, including ensemble and hybrid models, have being tested, and the results showed about 70% of accuracy correlating the predicted candidiasis-associated miRNAs with the ones obtained through literature. Further developments include optimizing the models, as well as experiment others able to classify the candidiasis type, obtaining a comprehensive analysis of these relations. Upon successful completion, this work is poised to make significant contributions to the fields of bioinformatics and biomedicine, ultimately contributing to mitigate the devast consequences of candidiasis.