Type 2 Diabetes Mellitus and its comorbidity, Alzheimer's disease: Identifying critical microRNA using machine learning

MicroRNAs (miRNAs) are critical regulators of gene expression in healthy and diseased states, and numerous studies have established their tremendous potential as a tool for improving the diagnosis of Type 2 Diabetes Mellitus (T2D) and its comorbidities. In this regard, we computationally identify novel top-ranked hub miRNAs that might be involved in T2D. We accomplish this via two strategies: 1) by ranking miRNAs based on the number of T2D differentially expressed genes (DEGs) they target, and 2) using only the common DEGs between T2D and its comorbidity, Alzheimer's disease (AD) to predict and rank miRNA. Then classifier models are built using the DEGs targeted by each miRNA as features. Here, we show the T2D DEGs targeted by hsa-mir-1-3p, hsa-mir-16-5p, hsa-mir-124-3p, hsa-mir-34a-5p, hsa-let-7b-5p, hsa-mir-155-5p, hsa-mir-107, hsa-mir-27a-3p, hsa-mir-129-2-3p, and hsa-mir-146a-5p are capable of distinguishing T2D samples from the controls, which serves as a measure of confidence in the miRNAs' potential role in T2D progression. Moreover, for the second strategy, we show other critical miRNAs can be made apparent through the disease's comorbidities, and in this case, overall, the hsa-mir-103a-3p models work well for all the datasets, especially in T2D, while the hsa-mir-124-3p models achieved the best scores for the AD datasets. To the best of our knowledge, this is the first study that used predicted miRNAs to determine the features that can separate the diseased samples (T2D or AD) from the normal ones, instead of using conventional non-biology-based feature selection methods.

Automated summary

MicroRNAs (miRNAs) play critical regulatory roles in gene expression in both healthy and diseased states, and have tremendous potential as a tool for improving the diagnosis of Type 2 Diabetes Mellitus (T2D) and its comorbidities. In this study, novel hub miRNAs potentially involved in T2D were computationally identified using two strategies: ranking miRNAs based on the number of T2D differentially expressed genes (DEGs) they target, and predicting and ranking miRNA using the common DEGs between T2D and Alzheimer's disease (AD). Classifier models were built using the DEGs targeted by each miRNA as features. The results showed that specific T2D DEGs targeted by certain miRNAs were capable of distinguishing T2D samples from controls, indicating their potential role in T2D progression.