Author(s)

Min Chen, Bikun Song, Yuke Wang

Corresponding Author

Min Chen

Abstract

The accuracy of non-invasive prenatal testing (NIPT) is highly affected by the concentration of free DNA in the fetus, and the male fetus needs to rely on the concentration of Y chromosome to ensure the reliability of the test. In this paper, statistical regression and cluster optimization methods are used to systematically solve the problems of Y chromosome concentration modeling and optimal detection time recommendation. In the first step, the Spearman rank correlation coefficient was used to analyze the relationship between Y chromosome concentration and gestational age, BMI and other indicators, and multiple linear and nonlinear regression models were established. The results showed that gestational age was significantly positively correlated with Y chromosome concentration, and BMI was significantly negatively correlated, and the goodness of fit R² of the nonlinear model reached 0.095, which was better than the linear model and passed the significance test. In the second step, a data-driven hierarchical model based on BMI is constructed, and K-means clustering is used to balance the risk of detection failure and delay by weighted risk function, and the optimal detection time point is optimized. The results showed that the high BMI group (group 1 and group 2) needed 16 weeks to meet the standard, and the rest group took 13 weeks. In the third step, multidimensional features such as age and X chromosome concentration were fused, and the weights (X chromosome 0.42, BMI 0.31, age 0.27) were determined by recursive feature elimination and entropy weight method, and five groups were determined by clustering with K-means. The model is verified to be reliable by stability analysis.

Keywords

NIPT; multiple linear regression; random forest; Cox returns; risk function; Cluster analysis