67: M-REGLE: Multimodal AI improves genetic prediction of cardiovascular traits

Zhou Y et al., The American Journal of Human Genetics - This episode explores M-REGLE, a multimodal deep‑learning pipeline that jointly learns representations from ECG and PPG waveforms to boost GWAS discovery and polygenic risk prediction for cardiovascular traits, including atrial fibrillation, and validates results across multiple biobanks. Key terms: M-REGLE, multimodal learning, ECG, PPG, GWAS. Study Highlights:The authors developed M-REGLE, an early‑fusion convolutional variational autoencoder that jointly encodes complementary electrophysiological waveforms (12‑lead ECG, ECG lead I, and PPG) into low‑dimensional embeddings, then ran GWAS on orthogonalized PCs and combined chi‑squared statistics. Compared to unimodal representation learning (U‑REGLE), M‑REGLE reduced reconstruction error, discovered more genome‑wide significant hits and loci (e.g., ~19.3% more loci on 12‑lead ECG and ~13.0% more on ECG lead I + PPG), and yielded higher expected chi‑squared statistics. Polygenic risk scores built from M‑REGLE hits significantly improved prediction for several cardiovascular phenotypes, notably atrial fibrillation, and findings were validated in Indiana Biobank, EPIC‑Norfolk, and the British Women’s Heart and Health Study. The method generalized to adding a third modality (spirograms) and outperformed PCA and CAE baselines in power and enrichment for cardiovascular terms. Conclusion:Joint multimodal representation learning with M‑REGLE leverages complementary ECG and PPG signals to increase GWAS power and improve polygenic risk prediction for cardiac traits, demonstrating a practical route to use wearable waveform data for genetic discovery. Music:Enjoy the music based on this article at the end of the episode. Article title:Applying multimodal AI to physiological waveforms improves genetic prediction of cardiovascular traits First author:Zhou Y Journal:The American Journal of Human Genetics DOI:10.1016/j.ajhg.2025.05.015 Reference:Zhou Y., Khasentino J., Yun T., et al. Applying multimodal AI to physiological waveforms improves genetic prediction of cardiovascular traits. The American Journal of Human Genetics. 2025;112:1562–1579. https://doi.org/10.1016/j.ajhg.2025.05.015 License:This episode is based on an open-access article published under the Creative Commons Attribution 4.0 International License (CC BY 4.0) – https://creativecommons.org/licenses/by/4.0/ Support:Base by Base – Stripe donations: https://donate.stripe.com/7sY4gz71B2sN3RWac5gEg00 Official website https://basebybase.com On PaperCast Base by Base you'll discover the latest in genomics, functional genomics, structural genomics, and proteomics. Episode link: https://basebybase.com/episodes/m-regle-multimodal-ai-ecg-ppg-gwas QC:This episode was checked against the original article PDF and publication metadata for the episode release published on 2025-07-06. QC Scope:- article metadata and core scientific claims from the narration- excludes analogies, intro/outro, and music- transcript coverage: Audited sections cover the M-REGLE architecture (early fusion VAE), comparison with unimodal REGLE, reconstruction improvements, GWAS power gains, PRS performance for atrial fibrillation, cross-biobank validation, and extension to spirograms.- transcript topics: M-REGLE concept and early fusion; VAE latent embeddings and joint multimodal representation; Canonical correlation between ECG lead I and PPG; GWAS power gains and loci discovery (12-lead ECG and lead I + PPG); Polygenic risk scores and atrial fibrillation prediction; Cross-biobank validation (UK Biobank, Indiana Biobank, EPIC-Norfolk, BWHHS) QC Summary:- factual score: 10/10- metadata score: 10/10- supported core claims: 4- claims flagged for review: 0- metadata che...