Noninvasive fetal genotyping using deep neural networks.

Area of Science:

• Genetics and Genomics
• Molecular Diagnostics
• Bioinformatics

Background:

• Circulating cell-free DNA (cfDNA) analysis is crucial for liquid biopsy and noninvasive prenatal testing (NIPT) for chromosomal abnormalities.
• Current noninvasive prenatal testing for monogenic diseases (NIPT-M) using cfDNA is limited, despite advances in fragmentomics and deep learning (DL) for individual genotyping.

Purpose of the Study:

• To develop the first DL-based framework for cfDNA-based fetal genotyping to enable genome-wide NIPT-M.
• To create an efficient DL model for ultra-deep whole genome sequencing (WGS) data, integrating diverse information levels.

Main Methods:

• Developed a novel DL framework for cfDNA-based genotyping.
• Utilized ultra-deep WGS data, incorporating DNA nucleotides, fragments, mutation regions, samples, and familial traits.
• Compared performance against traditional statistical and machine-learning methods.

Main Results:

• The DL framework surpassed existing methodologies in cfDNA-based genotyping.
• Successfully detected three deleterious mutations, enabling NIPT-M as early as the 7th week of gestation.
• Demonstrated improved integration of multi-level information for enhanced accuracy.

Conclusions:

• The proposed DL approach significantly advances the clinical feasibility of genome-wide NIPT-M for all mutation types.
• This innovation empowers families and healthcare providers with earlier, informed decisions during pregnancy.
• Reduces anxiety and uncertainty associated with prenatal genetic screening.