GARFIELD-NGS: Genomic vARiants FIltering by dEep Learning moDels in NGS

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Human Genetics

Background:

• Exome sequencing is crucial for identifying disease-causing genetic variants.
• A high rate of false positive variant calls complicates interpretation and diagnosis.
• Existing filtering methods struggle to accurately differentiate true from false variants.

Purpose of the Study:

• To introduce GARFIELD-NGS, a novel deep learning tool for filtering genetic variants.
• To enhance the accuracy of variant identification in next-generation sequencing (NGS) data.
• To improve the reliability of exome sequencing for research and clinical diagnostics.

Main Methods:

• Developed GARFIELD-NGS utilizing deep learning models.
• Trained and validated models on exome sequencing data from Illumina and ION platforms.
• Assessed performance for both Single Nucleotide Polymorphism (SNP) and Insertion/Deletion (INDEL) variants.

Main Results:

• GARFIELD-NGS demonstrated strong performance, achieving AUC values from 0.71 to 0.98.
• The tool significantly outperformed established hard filtering methods.
• It remains robust at low sequencing coverage (down to 30X) and is compatible with new Illumina chemistries.

Conclusions:

• GARFIELD-NGS effectively filters false positive variants from exome sequencing data.
• The tool integrates seamlessly into existing bioinformatics pipelines via standard VCF files.
• It offers adjustable sensitivity and specificity for diverse research and diagnostic needs.