A scalable tool for analyzing genomic variants of humans using knowledge graphs and graph machine learning

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-throughput genome sequencing advances enable large-scale genomic data analysis in clinical practice and research.
• Genomic variant analysis is crucial for understanding complex disease risk factors, including cancer and COVID-19.
• Knowledge graphs (KGs) and graph machine learning (GML) offer powerful methods for modeling and analyzing complex genomic datasets.

Purpose of the Study:

• To present VariantKG, a scalable tool for analyzing human genomic variants using KGs and GML.
• To demonstrate the application of VariantKG using publicly available genome sequencing data from COVID-19 patients.
• To provide an intuitive platform for KG construction, model inference, and interpretation of genomic variants.

Main Methods:

• VariantKG extracts and annotates variant-level genetic information, converting it into a KG using Resource Description Framework (RDF).
• The KG is enhanced with patient metadata and stored in a scalable graph database for efficient querying.
• GML tasks, including node classification, are performed using the Deep Graph Library (DGL) on subsets of the KG.

Main Results:

• KG construction using 1,508 genome sequences resulted in 4 billion RDF statements.
• Node classification tasks were evaluated on a subset of 500 sequences using GML techniques like GraphSAGE, GCN, and Graph Transformer.
• VariantKG demonstrated efficient KG construction, scalable data handling, and effective GML inference.

Conclusions:

• VariantKG offers a scalable and accessible solution for analyzing human genomic variants through KGs and GML.
• The tool facilitates a deeper understanding of genetic factors contributing to complex diseases.
• VariantKG lowers the barrier to entry for researchers and clinicians working with large-scale genomic data and GML.