In Gynecological endocrinology : the official journal of the International Society of Gynecological Endocrinology
OBJECTIVE : To explore gestational diabetes mellitus (GDM) diagnostic markers and establish the predictive model of GDM.
METHODS : We downloaded the DNA methylation data of GSE70453 and GSE102177 from the Gene Expression Omnibus database. Epigenome-wide association study (EWAS) was performed to analyze the relationship between cytosine-phosphate-guanine (CpG) methylation and GDM. And then the logistic regression models were constructed, with the β-values of CpG sites as predictor variable and the GDM occurrence as binary outcome variable. Data from GSE70453 served as training sets and data from GSE102177 served as verification sets.
RESULTS : The EWAS and overlap analysis identified nine-shared significant CpGs in the two DNA methylation data sets. Remarkably, these nine CpGs were differently methylated in GDM samples compared to their matched normal specimens, among which five fully methylated CpGs were finally selected. Importantly, we established a binary logistic regression model based on the above five CpGs, in which cg11169102, cg21179618 and cg21620107 were critical. Hence, we further built a logistic regression model by using the three CpGs and found that the area under the curve was 0.8209. The validation of the model by using the verification sets indicated the area under the curve was 0.8519.
CONCLUSIONS : We identified potential CpG biomarkers for the diagnosis of gestational diabetes mellitus patients through using EWAS and Logistic regression models in combination.
Liu Yan, Wang Zhenglu, Zhao Lin
Gestational diabetes mellitus, cytosine-phosphate-guanine methylation, diagnosis, epigenome-wide association study, logistic regression, machine learning