Abstract: Objective  The aim of this study was to identify characteristic genetic biomarkers that can differentiate between early-onset and late-onset preeclampsia patients and to investigate the differentiating ability of these genes. Methods  Data sets for early-onset and late-onset preeclampsia(GSE74341, GSE190639 and GSE4707 data sets) were obtained from the GEO database. The GSE74341 and GSE190639 data sets were used as the experimental group, and the GSE4707 data set was used as the verification group to screen and verify the differentially expressed genes during early-onset preeclampsia and late-onset preeclampsia. Two machine learning methods, namely  least absolute shrinkage and selection operator （LASSO） and support vector machines-recursive feature elimination （SVM-RFE）, were employed to select characteristic genes. The discriminative ability of these genes was evaluated using receiver operating characteristic (ROC) curves. Results  Compared to early-onset preeclampsia, we identified seven significantly upregulated genes and three significantly downregulated genes in late-onset preeclampsia. By utilizing the two machine learning methods and analyzing gene expression differences in the validation group, one characteristic gene (MME) was selected. The area under the ROC curve (AUC) for the experimental group and validation group was 0.975 (95% CI: 0.921-1.000) and 1 (95% CI:1.000–1.000), respectively. Conclusions  Our findings suggest that MME may serve as a potential characteristic gene for distinguishing between early-onset and late-onset preeclampsia.

Key words: early-onset and late-onset preeclampsia, bioinformatics analysis, machine learning, characteristic genetic markers