Biologically relevant neural network architectures for support vector machines

Complement factors and alpha-fetoprotein as biomarkers for noninvasive prenatal diagnosis of neural tube defects

Neural tube defects (NTDs) are serious congenital malformations. In this study, we aimed to identify more specific and sensitive maternal serum biomarkers for noninvasive NTD screenings. We collected serum from 37 pregnant women carrying fetuses with NTDs and 38 pregnant women carrying normal fetuses. Isobaric tags for relative and absolute quantitation were conducted for differential proteomic analysis, and an enzyme-linked immunosorbent assay was used to validate the results. We then used a support vector machine (SVM) classifier to establish a disease prediction model for NTD diagnosis. We identified 113 differentially expressed proteins; of these, 23 were either up- or downregulated 1.5-fold or more, including five complement proteins (C1QA, C1S, C1R, C9, and C3); C3 and C9 were downregulated significantly in NTD groups. The accuracy rate of the SVM model of the complement factors (including C1QA, C1S, and C3) was 62.5%, with 60% sensitivity and 67% specificity, while the accuracy rate of the SVM model of alpha-fetoprotein (AFP, an established biomarker for NTDs) was 62.5%, with 75% sensitivity and 50% specificity. Combination of the complement factor and AFP data resulted in the SVM model accuracy of 75%, and receiver operating characteristic curve analysis showed 75% sensitivity and 75% specificity. These data suggest that a disease prediction model based on combined complement factor and AFP data could serve as a more accurate method of noninvasive prenatal NTD diagnosis.