Using Machine Learning to Predict the Development of Diabetes and Potential Biomarkers Linked to Cardiac Risk

Abstract

BackgroundDiabetes mellitus is a chronic, debilitating disease that continues to affect a greater percentage of people each year. Among its systemic comorbidities, diabetics are two to four times more likely to develop cardiovascular diseases. While glycosylated hemoglobin (HbA1c) remains the primary diagnostic for diabetes mellitus onset, predicting health outcomes through a single measure is difficult, with disparities existing between ethnic and demographic groups. The purpose of this study was to use machine learning algorithms to develop personalized prognostics in predicting the development and progression of diabetes mellitus in the heart.MethodsRight atrial appendages from 48 patients, 29 non‐diabetic and 19 type 2 diabetic, were procured from the WVU Ruby Memorial Hospital. Machine learning was applied to physiological (sex, BMI, HbA1c, comorbidities), biochemical (total methylation and mitochondrial bioenergetics functional parameters), and sequencing data (mitochondrial genomic DNA single nucleotide polymorphisms (SNPs) and nuclear genomic DNA CpG island methylation) for each patient. Supervised (including HbA1c) and unsupervised (without HbA1c) learning evaluated the efficiency of testing and training sets through classification and regression trees, logistic regression, support vector machines, and k nearest neighbors.ResultsNuclear 5‐methylcytosine content and s‐adenosyl methionine activity matched diagnostic accuracy of HbA1c (~91% testing) for all patients. Mitochondrial DNA SNPs found in the D‐Loop region (SNP‐152C, ‐16126C, and ‐16362C) were significantly associated with diabetes incidence. The CpG island for the mitochondrial transcription factor A (TFAM) revealed specific CpG sites (chr10:58384915 and 58385262) that showed high accuracy (~89% testing) in predicting diabetes incidence. Unsupervised learning, both binary (diabetic or non‐diabetic) and multi‐classification (diabetic, pre‐diabetic, or non‐diabetic), confirmed these results.ConclusionsUsing machine learning, we were able to predict health outcomes through the combination of multiple physiological, biochemical, and sequencing approaches, assessing the development and progression of diabetes. Linking mitochondrial and genomic/epigenomic data, we detail how learning algorithms can be applied to forwarding personalized medicine and developing novel, more advanced prognostics in the heart.Support or Funding InformationR01 HL‐128485 (JMH), AHA‐17PRE33660333 (QAH), WV‐INBRE support by NIH Grant P20GM103434, WVU Flow Cytometry & Single Cell Core supported by MBRCC CoBRE Grant GM103488 and Fortessa S10 Grant OD016165, and the Community Foundation for the Ohio Valley Whipkey Trust. We would like to acknowledge the WVU Genomics Core Facility, Morgantown WV for support provided to help make this work possible.This abstract is from the Experimental Biology 2019 Meeting. There is no full text article associated with this abstract published in The FASEB Journal.

Mitochondrial epigenetics in aging and cardiovascular diseases

“…Application of machine learning to identify novel cardiac biomarkers reveals that total nuclear methylation and methylation in a specific CpG island of TFAM were the best diagnostic measures related to diabetes progression (118). Also this study support the theory that epigenetics and mitoepigenetics are two processes interconnected.…”

Physical frailty identification using machine learning to explore the 5-item FRAIL scale, Cardiovascular Health Study index, and Study of Osteoporotic Fractures index

“…For patients with severe COVID-19 intubation, Fleuren et al applied the SHAP and found predictors of extubation failure, including ventilatory settings, inflammatory parameters, neurological status, and BMI (51). Hathaway et al (52) conducted supervised learning through SHAP by identifying the most relevant and novel cardiac biomarkers for forecasting diabetes mellitus development, and discovered that this approach may be a potential guideline for investigating disease pathogenesis and discovering novel biomarkers in the future. For predicting infant autopsy outcome, Booth et al used three models for model training, including decision tree, RF, and gradient boosting.…”

“…Fifth, because the dataset is inherently predictive, when the sample size is small, models may face challenges. One of these challenges is the high sensitivity to outliers, which may overly emphasize anomalies in the samples, leading the ML model to believe that these outliers have a greater impact (52). Due to limitations in the dataset, the model may overfit to the training data, especially when using derived models like classification trees.…”