DRIVE-KG: Enhancing variant-phenotype association discovery in understudied complex diseases using heterogeneous...

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Women's Health Research

Background:

• Endometriosis is a prevalent women's health condition affecting 10% of reproductive-aged women.
• Limited genetic characterization of endometriosis exists, with current GWAS explaining only 11% of its heritability.
• Multi-omics data integration is challenging due to sparsity and high dimensionality.

Purpose of the Study:

• To present DRIVE-KG, a novel heterogeneous knowledge graph for integrating multi-omics data.
• To identify novel single nucleotide polymorphism (SNP)-endometriosis associations.
• To improve clinical prediction of endometriosis status using graph-based machine learning.

Main Methods:

• Constructed a heterogeneous graph (DRIVE-KG) integrating data from dbSNP, NCBI Human Gene, Omics Pred, GTEx, and Open Targets.
• Utilized a link prediction model to identify SNP-phenotype associations.
• Employed a graph convolutional network (GCN) for patient-level endometriosis classification using data from 1,441 participants.

Main Results:

• Uncovered 66 high-confidence, previously unreported SNP-endometriosis associations.
• Identified links between novel variants and obesity, lipid metabolism, and depressive disorders, aligning with emerging endometriosis hypotheses.
• Achieved an AUPRC of 0.738 for endometriosis classification using GCN, outperforming a genetic risk score (0.679).

Conclusions:

• Heterogeneous multi-omics data integration via DRIVE-KG is valuable for discovery and clinical prediction in understudied diseases.
• DRIVE-KG facilitates the identification of novel genetic associations for endometriosis.
• Graph-based approaches offer improved clinical prediction for endometriosis compared to traditional methods.