Polishing copy number variant calls on exome sequencing data via deep learning

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Copy number variants (CNVs) are crucial in genetic diseases.
• Whole-genome sequencing (WGS) offers high accuracy for CNV detection.
• Whole-exome sequencing (WES) is cost-effective but less accurate for CNVs due to capture biases.

Purpose of the Study:

• To develop a deep learning model to improve CNV detection accuracy on WES data.
• To correct inaccuracies in existing WES-based CNV callers.
• To enhance the reliability of germline CNV detection using ubiquitous WES data.

Main Methods:

• Developed DECoNT, a novel deep learning model.
• Trained DECoNT using matched WES and WGS data from the 1000 Genomes Project.
• Evaluated DECoNT's performance across different sequencing technologies, capture kits, and CNV callers.

Main Results:

• DECoNT significantly enhances CNV detection accuracy on WES data.
• Achieved a threefold increase in duplication call precision and a twofold increase in deletion call precision.
• Demonstrated consistent performance improvement irrespective of sequencing or capture methods.

Conclusions:

• DECoNT acts as a universal polishing tool for exome CNV calls.
• The model substantially improves the reliability of germline CNV detection from WES data.
• Facilitates more accurate genetic disease association studies using cost-efficient WES data.