Prioritizing genomic variants through neuro-symbolic, knowledge-enhanced learning.

Area of Science:

• Genomics
• Computational Biology
• Medical Genetics

Background:

• Whole-exome and genome sequencing are crucial for rare disease diagnosis but often leave patients undiagnosed.
• Interpreting genomic variants requires understanding gene function, expression, and physiological impacts.
• Existing phenotype-based methods are limited by reliance on known gene-phenotype associations and inconsistent phenotype data.

Purpose of the Study:

• To develop a novel computational method for prioritizing variants implicated in genetic diseases.
• To overcome limitations of current phenotype-based approaches by integrating broader biological knowledge.
• To improve the diagnostic yield of genomic sequencing in rare disease patients.

Main Methods:

• Developed Embedding-based Phenotype Variant Predictor (EmbedPVP), a computational tool.
• Integrated genomic information with clinical phenotypes using neuro-symbolic, knowledge-enhanced machine learning.
• Leveraged background knowledge from human and model organisms regarding molecular mechanisms of abnormal phenotypes.

Main Results:

• EmbedPVP effectively prioritizes variants involved in genetic diseases.
• The method successfully combines genomic data and clinical phenotypes.
• Demonstrated efficacy on synthetic and real-world genomic datasets.

Conclusions:

• EmbedPVP offers a powerful new approach for variant prioritization in rare disease diagnosis.
• The method enhances the interpretation of genomic data by incorporating extensive biological knowledge.
• EmbedPVP has the potential to increase diagnostic rates for previously undiagnosed rare disease patients.