Avaliação molecular da variante no gene MTR na susceptibilidade à diabetes mellitus: uma abordagem integrada com machine learning

Abstract

Type 2 diabetes mellitus (T2DM) is the most prevalent form of diabetes worldwide and represents a major public health challenge. In Brazil, its prevalence reaches 7.7%, imposing substantial economic and social burdens and reinforcing the need to elucidate the genetic and metabolic mechanisms underlying susceptibility to the disease. One-carbon metabolism plays a central role in regulating homocysteine levels, and variants in the MTR gene can destabilize this pathway, leading to hyperhomocysteinemia and increased risk for T2DM and its complications. To investigate this hypothesis, a population-based case–control study was conducted with more than 219 individuals matched by age and sex, with emphasis on the MTR rs1805087 variant. Genotyping data were analyzed under different inheritance models using chi-square tests and odds ratios (95% CI), and machine learning algorithms were applied to evaluate predictive performance through accuracy, precision, sensitivity, and ROC/AUC metrics. Although no significant association was observed between the rs1805087 variant and T2DM, a borderline result emerged under the dominant model. On the other hand, in analyses restricted to cases, the rs1805087 variant showed a significant association with smoking in the codominant (p = 0.009) and recessive (p = 0.041) models, revealing gene–environment interactions. In the machine learning approach, the K‑Nearest Neighbors (KNN) algorithm demonstrated the most stable performance between training and testing, outperforming more complex models such as Random Forest, which showed evidence of overfitting. Moreover, the variant emerged as one of the most influential variables in the predictive model. Taken together, the findings suggest that the MTR rs1805087 variant may contribute to T2DM risk through gene–environment interactions, particularly with smoking, and highlight the potential of integrated approaches combining molecular genetics and machine learning to refine risk assessment in multifactorial diseases.