Biological aging trajectories
Time-aware deep learning and multi-omic aging scores
People age at different speeds, and clinical visits happen at irregular times. This project builds models that follow each person’s aging path over time.
- PoA-HyTrax / SS-HyTrax: self-supervised, time-aware transformers using masked trajectory modeling and delta-time tokens to model non-linear biological aging across irregular visits
- Multi-omic fusion: combining longitudinal epigenetic and proteomic data with clinical phenotypes into sensitive aging scores
- Validation: testing whether these scores detect slowed aging in randomized trials (CALERIE, FAST Initiative) and harmonized multi-cohort biobank data
- Equity: checking calibration and bias across racial, socioeconomic, and underrepresented groups
Supported by an NIA Biomarker Network pilot project (2026–2027), with Dr. Daniel Belsky at the Columbia Aging Center.