🧬 Our 2nd preprint today,
MethylGPT: A foundation model for the DNA methylome!
Our generative pretained transformer-based model:
- Trained on 226,555 samples (154,063 after QC & deduplication)
- 49,156 curated CpG sites
- 7.6B training tokens
- Spans 20+ tissue types from 5,281 datasets
Key achievements:
- Robust methylation prediction (Pearson R=0.929)
- Maintains performance with up to 70% missing data
- Superior age prediction accuracy vs existing methods
- Disease risk prediction across 60 conditions
The model captures both genomic and biological context without external supervision.
Analysis reveals distinct methylation signatures between young & old samples, with developmental vs aging pathway enrichment.
Big thanks to the team:
@HAOTIANCUI1@jinyeop_song@VadimGladyshev@BoWang87@mahdi_moqri@jpoganik@alec_eames
@MarioniGroup
@gladyshev_lab@BrighamWomens@harvardmed
Read full text here: https://t.co/OEPYfIeT36
🧵 1/7 Co-authored by several Consortium members, and leveraging our open-source biomarkers of aging analysis tooling Biolearn, "multi-omic aging genes" were identified as potential biomarkers of aging and targets for epigenetic editing to facilitate cellular rejuvenation.
Now a U-M BME BSE alum, Alec Eames led a study on using artificial intelligence (AI) to study cell quiescence (cells that are sleeping or dormant). Eames conducted his undergrad research under Associate Professor Sriram Chandrasekaran. Read more: https://t.co/bY0OBg9dOY