Improving newborn screening accuracy through genome sequencing, targeted metabolomics, and machine learning.

backgroundNewborn screening (NBS) enables early detection of metabolic disorders, but current tandem mass spectrometry (MS/MS) methods often lead to false positives and require confirmatory testing, causing diagnostic delays. We evaluated whether integrating genome sequencing, expanded metabolite profiling, and artificial intelligence/machine learning (AI/ML) could improve the accuracy of NBS.

methodsWe analyzed dried blood spots (DBS) from 119 screen-positive cases identified by the California NBS program across four disorders: GA-I, PA/MMA, OTCD, and VLCADD. Genome sequencing was performed to identify variants in condition-related genes using ACMG guidelines, and an AI/ML classifier trained on previously generated metabolomic data was applied to differentiate true and false positives.

resultsGenome sequencing confirmed 89% (31/35) of true positives based on the presence of two reportable variants. Among 84 false positives, 74% (62) had no variant, while 26% (22) carried a pathogenic/likely pathogenic variant or rare VUS in a condition-related gene. For VLCADD, half of false positives (15/29) were ACADVL variant carriers (P = 4.66 × 10⁻⁷). VLCADD biomarker levels were highest in patients, intermediate in carriers, and lowest in non-carriers, indicating that ACADVL variants elevate biomarker levels and increase false-positive rates. Metabolomics with AI/ML detected all true positives (100% sensitivity), while genome sequencing reduced false positives by 98.8%.

conclusionTargeted metabolomics with AI/ML showed high sensitivity for identifying true positives, though its ability to reduce false positives varied by condition. Genome sequencing effectively reduced false positives but lacked sufficient sensitivity as a standalone test. The elevated false-positive rate among pathogenic variant carriers underscores the potential value of parental or prenatal carrier screening to improve NBS accuracy. Integrating genomic and metabolomic data may enhance NBS precision and enable earlier diagnosis and intervention for rare diseases.