Combining lipidomics and machine learning to identify lipid biomarkers for nonsyndromic cleft lip with palate.

Abstract

Nonsyndromic cleft lip with palate (nsCLP) is a common birth defect disease. Current diagnostic methods comprise fetal ultrasound images, which are mainly limited by fetal position and technician skills. We aimed to identify reliable maternal serum lipid biomarkers to diagnose nsCLP. Eight-feature selection methods were used to assess the dysregulated lipids from untargeted lipidomics in a discovery cohort. The robust rank aggregation algorithm was applied on these selected lipids. The data were subsequently processed using 7 classification models to retrieve a panel of 35 candidate lipid biomarkers. Potential lipid biomarkers were evaluated using targeted lipidomics in a validation cohort. Seven classification models and multivariate analyses were constructed to identify the lipid biomarkers for nsCLP. The diagnostic model achieved high performance with 3 lipids in determining nsCLP. A panel of 3 lipid biomarkers showed great potential for nsCLP diagnosis. FA (20:4) and LPC (18:0) were also significantly downregulated in early serum samples from the nsCLP group in the additional validation cohort. We demonstrate the applicability and robustness of a machine-learning algorithm to analyze lipidomic data for efficient and reliable biomarker screening.