Aprendizado de máquina multirrótulo para predição de doenças associadas a RNAs longos não codificantes

Abstract

Long non-coding RNAs (lncRNAs) are RNA molecules longer than 200 nucleotides that are not translated into proteins. These RNAs play crucial roles in processes such as cell cycle regulation, epigenetics, cell differentiation, and both post-transcriptional and transcriptional regulation. Abnormal expression of lncRNAs has been increasingly associated with various human diseases, including cardiovascular, neurological, and oncological conditions. While different biotechnological approaches have been employed to identify and detect lncRNAs, these methods face significant challenges related to cost, operational complexity, and experimental procedure. Consequently, the development of artificial intelligence-based models for disease prediction using lncRNA data presents itself as a promising alternative. This study focuses on the implementation of machine learning algorithms for predicting diseases associated with lncRNAs, addressing it as a multi-label classification problem, given that a single lncRNA can be linked to multiple diseases. Various methods, including problem transformation approaches and algorithm adaptation approaches, were tested. The classifiers' performance was assessed using evaluation metrics for multi-label problems, including Precision, Recall, F-measure, and Hamming Loss. Furthermore, for comparative analysis, statistical tests were performed, specifically the Friedman and Nemenyi tests. Based on the analysis of the results, hypotheses regarding the performance of the algorithms were formulated. However, despite the observed differences, the methods showed an overall unsatisfactory performance, which we attribute to the high sparsity presented in the dataset. Thus, for future work, the investigation of more robust methods to address this issue was proposed, aiming to enhance prediction performance and improve the understanding of biological interactions.