Identify gestational diabetes mellitus by deep learning model from cell-free DNA at the early gestation stage

Abstract

Gestational diabetes mellitus (GDM) is a common complication of pregnancy, which has significant adverse effects on both the mother and fetus. The incidence of GDM is increasing globally, and early diagnosis is critical for timely treatment and reducing the risk of poor pregnancy outcomes. GDM is usually diagnosed and detected after 24 weeks of gestation, while complications due to GDM can occur much earlier. Copy number variations (CNVs) can be a possible biomarker for GDM diagnosis and screening in the early gestation stage. In this study, we proposed a machine-learning method to screen GDM in the early stage of gestation using cell-free DNA (cfDNA) sequencing data from maternal plasma. Five thousand and eighty-five patients from north regions of Mainland China, including 1942 GDM, were recruited. A non-overlapping sliding window method was applied for CNV coverage screening on low-coverage (~0.2×) sequencing data. The CNV coverage was fed to a convolutional neural network with attention architecture for the binary classification. The model achieved a classification accuracy of 88.14%, precision of 84.07%, recall of 93.04%, F1-score of 88.33% and AUC of 96.49%. The model identified 2190 genes associated with GDM, including DEFA1, DEFA3 and DEFB1. The enriched gene ontology (GO) terms and KEGG pathways showed that many identified genes are associated with diabetes-related pathways. Our study demonstrates the feasibility of using cfDNA sequencing data and machine-learning methods for early diagnosis of GDM, which may aid in early intervention and prevention of adverse pregnancy outcomes.

Keywords: cell-free DNA; copy number variations; deep learning; gestational diabetes mellitus.

Figure 1

An overview of the GDM classification model. (A) The pipeline for processing sequencing data to generate count matrix for deep learning model. (B) An example for the count matrix constructure. (C) The deep learning architecture of our model.

Figure 2

Clinical information of the collected samples. (A) The age distribution of the normal group and GDM group. (B) The weight distribution of the normal group and GDM group. (C) The sample time distribution of the normal and GDM group. (D) PCA scatter plot of normal samples and GDM samples.

Figure 3

Copy number diversity on different window size along the whole chromosome.

Figure 4

Model learning performance on different window size. (A–D) Learning accuracy curve for window size 50, 20, 10 and 5 k, respectively. (E–H) Learning loss curve for window size 50, 20, 10 and 5 k, respectively.

Figure 5

Model performance on test dataset. (A–D) ROC curve for window size 50, 20, 10 and 5 k, respectively. The black line represents the ROC curve generated by the test dataset, while the gray lines represent the ROCs obtained from cross-validation. (E) The AUC distribution of 10 k window size on 10 000 bootstrapping experiments. (F) The AUCs of variant window sizes on bootstrapping experiments.

Figure 6

The results of GO term and KEGG pathway enrichment of annotated genes. (A) Enriched GO terms with adjusted P-value smaller than 0.05. (B) Enriched KEGG pathways with adjusted P-value smaller than 0.05.