VarPPUD: Variant post prioritization developed for undiagnosed genetic disorders

Area of Science:

• Genomics
• Computational Biology
• Medical Genetics

Background:

• Rare and ultra-rare genetic conditions affect 1 in 17 people globally, posing diagnostic challenges.
• Current *in silico* variant pathogenicity predictions struggle to distinguish etiological variants in complex undiagnosed cases.
• Pinpointing disease-causing variants often requires extensive manual effort, including clinical workups and experimental assays.

Approach:

• Introducing VarPPUD, a novel tool trained on variants from Undiagnosed Diseases Network (UDN) cases.
• VarPPUD utilizes gene-, amino acid-, and nucleotide-level features for variant pathogenicity prediction.
• The tool was validated on challenging UDN cases and synthetic, GAN-generated variants.

Key Points:

• VarPPUD achieved 79.3% cross-validated accuracy and 77.5% precision on difficult UDN cases.
• This represents an 18.6% and 23.4% improvement over nine traditional prediction methods, respectively.
• VarPPUD's feature importance can be analyzed to understand disease-causing variant characteristics.

Conclusions:

• VarPPUD effectively discerns pathogenic variants from other deleterious variants in challenging cases.
• The tool offers a significant advancement over existing pathogenicity prediction approaches for rare diseases.
• Understanding feature contributions aids in characterizing novel disease-causing variants.