Data-driven consideration of genetic disorders for global genomic newborn screening programs

Area of Science:

• Genomics
• Bioinformatics
• Public Health

Background:

• Newborn sequencing (NBSeq) is expanding the scope of genetic disorder screening.
• Significant variability exists in gene selection across NBSeq programs globally.
• A systematic approach is needed to prioritize genes for NBSeq.

Purpose of the Study:

• To identify predictors of gene inclusion in NBSeq programs.
• To develop a machine learning model for ranking genes for NBSeq.
• To provide a data-driven method for gene prioritization in newborn screening.

Main Methods:

• Compiled a dataset of 25 characteristics for 4390 genes across 27 NBSeq programs.
• Utilized regression analysis to determine key predictors of gene inclusion.
• Developed a boosted trees machine learning model to rank genes based on public health relevance.

Main Results:

• Gene selection varied greatly, with only 74 genes (1.7%) common to over 80% of programs.
• Inclusion was strongly associated with the US Recommended Uniform Screening Panel, natural history evidence, and treatment efficacy.
• The machine learning model achieved high accuracy (AUC = 0.915, R² = 84%) in predicting gene inclusion.

Conclusions:

• A machine learning model offers a ranked gene list for NBSeq initiatives.
• This model can adapt to new evidence and regional requirements.
• It facilitates more consistent and informed gene selection for newborn screening programs.