Text mining-based identification of promising miRNA biomarkers for diabetes mellitus.

Abstract

MicroRNAs (miRNAs) are small, non-coding RNAs that play a critical role in diabetes development. While individual studies investigating the mechanisms of miRNA in diabetes provide valuable insights, their narrow focus limits their ability to provide a comprehensive understanding of miRNAs' role in diabetes pathogenesis and complications. To reduce potential bias from individual studies, we employed a text mining-based approach to identify the role of miRNAs in diabetes and their potential as biomarker candidates. Abstracts of publications were tokenized, and biomedical terms were extracted for topic modeling. Four machine learning algorithms, including Naïve Bayes, Decision Tree, Random Forest, and Support Vector Machines (SVM), were employed for diabetes classification. Feature importance was assessed to construct miRNA-diabetes networks. Our analysis identified 13 distinct topics of miRNA studies in the context of diabetes, and miRNAs exhibited a topic-specific pattern. SVM achieved a promising prediction for diabetes with an accuracy score greater than 60%. Notably, miR-146 emerged as one of the critical biomarkers for diabetes prediction, targeting multiple genes and signal pathways implicated in diabetic inflammation and neuropathy. This comprehensive approach yields generalizable insights into the network miRNAs-diabetes network and supports miRNAs' potential as a biomarker for diabetes.