Abstract: TH-PO186
Transfer Learning Predicts Network Biology for Fibroblast Growth Factor 23 (FGF-23) and Cardiac Disease
Session Information
- CKD-MBD: Clinical
October 24, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Bone and Mineral Metabolism
- 502 Bone and Mineral Metabolism: Clinical
Authors
- Perwad, Farzana, University of California San Francisco, San Francisco, California, United States
- Akwo, Elvis Abang, Vanderbilt University Medical Center, Nashville, Tennessee, United States
- Robinson-Cohen, Cassianne, Vanderbilt University Medical Center, Nashville, Tennessee, United States
Background
Multi-trait analysis of Genome-wide association (MTAG) is a novel method to explore overlapping genetic architecture between traits. We investigated genetic traits common to mineral metabolism to identify novel genetic associations for fibroblast growth factor 23 in the context of cardiac disease. We utilized transfer learning, a novel machine learning tool to enable context-specific predictions in a setting with limited data in network biology.
Methods
We applied MTAG to genetic variants common to 5 genetically correlated mineral metabolism markers (phosphorus, FGF23, calcium, vitamin D and PTH) in European-ancestry subjects from UKBioBank (n=366,484), and CHARGE consortium (n= 45,779). We applied a context-aware, attention-based deep learning model, Geneformer, pre-trained on a large-scale corpus of ~ 30 million single-cell transcriptomes to enable context-specific predictions. We also queried for genes functionally related to uremia using gene reference into function database.
Results
MTAG identified independent genome-wide significant SNPs for all traits, including novel and known loci for FGF23. We identified statistical overlap of GWAS variants with Geneformer-derived genetic targets obtained from modeling single nuclear cell RNA seq from human heart transplant recipients and from deceased organ donors with healthy hearts (Figure 1). In silico gene perturbations revealed genes that, when activated or deleted, were protective against hypertrophic or dilated cardiomyopathy (Table 1). Further, we identified uremic solutes, glyoxal, erythritol and calcitonin gene-related peptide associated with both MTAG variants and pathophysiology of cardiac disease.
Conclusion
Novel genetic traits for FGF23 were identified with MTAG and associated with cardiac disease. Geneformer revealed networks that inform unique biological processes and cellular functions that could be targeted to develop therapeutics for human heart failure.
Funding
- NIDDK Support