Large scale genotype- and phenotype-driven machine learning in Von Hippel-Lindau disease

Area of Science:

• Genetics
• Oncology
• Bioinformatics

Background:

• Von Hippel-Lindau (VHL) disease is an inherited cancer syndrome caused by VHL gene mutations, leading to tumors in various organs.
• Collecting standardized VHL patient data is challenging due to disease rarity and patient diversity.
• Existing knowledge on VHL genotype-phenotype correlations is fragmented.

Purpose of the Study:

• To create a standardized, open-access knowledgebase of VHL germline genotype-phenotype data.
• To identify and confirm genotype-phenotype correlations and disease patterns in VHL.
• To facilitate the interpretation of VHL variants' pathogenicity.

Main Methods:

• Screened over 4100 articles published up to October 2019 for VHL germline genotype-phenotype data.
• Standardized patient data using Human Genome Variation Society nomenclature and Human Phenotype Ontology terms.
• Applied statistical methods and spectral clustering for unsupervised learning to analyze genotype-phenotype relationships.

Main Results:

• Compiled data from 427 papers, including 634 unique VHL variants, 2882 patients, and 1991 families.
• Identified trends showing earlier onset of pheochromocytoma/paraganglioma and retinal angiomas.
• Revealed phenotype co-occurrences and genotype-phenotype correlations, including mutation hotspots.

Conclusions:

• The curated VHL knowledgebase effectively aggregates and translates complex genetic and clinical information.
• The findings confirm known VHL disease associations and uncover novel patterns.
• This resource aids in understanding VHL variant pathogenicity and facilitates clinical interpretation.