Temporal AI model predicts drivers of cell state trajectories across human aging

Area of Science:

• Computational biology
• Artificial intelligence in medicine
• Aging research

Background:

• Current AI models for cell states are limited to single time points.
• Understanding long-term cellular dynamics, especially in aging and disease, remains a challenge.
• Predicting cellular responses over extended periods is crucial for aging research.

Purpose of the Study:

• To develop a temporal AI model capable of predicting cell state trajectories over long durations.
• To leverage large-scale gene expression data for modeling the human lifespan.
• To accelerate the discovery of interventions for age-related diseases.

Main Methods:

• Developed MaxToki, a temporal AI model trained on nearly 1 trillion gene tokens.
• Utilized cell state trajectories across the human lifespan for training.
• Employed in-context learning for generalization to unseen trajectories.

Main Results:

• MaxToki successfully generated cell states across long human aging timelines.
• The model generalized effectively to novel, unseen cellular trajectories.
• Experimentally validated novel age-modulating targets predicted by MaxToki.
• Identified targets that influence age-related gene programs and in vivo functional decline.

Conclusions:

• MaxToki demonstrates a powerful approach for temporal AI modeling in biology.
• This AI strategy accelerates the discovery of therapeutic interventions for aging.
• MaxToki can guide the programming of cellular trajectories for therapeutic benefit.