Reverse Aging
Epigenetic Reprogramming & Reverse Aging · Aging Biology
Restore a youthful epigenome with partial reprogramming — turn biological age down without losing cell identity.
📋 The problem
Aging is, in part, a loss of epigenetic information — cells forget which genes to express. Transient partial reprogramming with Yamanaka factors (OSK) can reset the methylation landscape toward a youthful state and reverse biological-age clocks in vivo, without erasing cell identity.
🧗 Why it's a grand challenge
The window between rejuvenation and dedifferentiation (or tumorigenesis) is narrow, effects differ by tissue, and 'biological age' itself is measured by clocks we don't fully trust.
🧮 Governing model
dA/dt = k_dmg − k_rep · u(t); Age_bio = Clock(M_methylation)Information-theoretic aging: biological age A(t) accumulates as epigenetic noise; transient OSK reprogramming u(t) drives dA/dt = k_dmg − k_rep·u(t), read out by a DNA-methylation clock.
Current best: Partial OSK reprogramming (Lu et al., optic-nerve / Sinclair lab)
🧭 Possible approaches
- Learn a causal map: reprogramming dose/duration → methylation state → biological age
- Closed-loop control of OSK exposure that maximizes age reversal under identity-preservation constraints
- Multi-tissue methylation foundation models that separate rejuvenation from dedifferentiation
🎯 Build the benchmark
Given single-cell methylation before/after transient OSK across tissues, predict restored biological age within ~2 years while flagging any loss of cell identity.
Metric: years_reversed — biological years restored
Datasets to start from: Multi-tissue DNAm reprogramming atlas, Horvath pan-tissue clock corpus, OSK transient-induction time course
🤖 Build an AI agent to solve it
An agent that designs reprogramming protocols (factors, dose, schedule) per cell type and predicts the age-reversal vs safety trade-off.
Once a benchmark exists, an AI4Science agent can iterate solutions against it — every verified solution earns PWM.
This is a frontier framing page — an open problem, not yet benchmarked or verified, unlike PWM's mature computational-imaging benchmarks.