Abstract
Loss of stem cell self-renewal may underpin aging. Here, we combined single cell profiling, deep-learning, mathematical modelling and in vivo functional studies to explore how hematopoietic stem cell (HSC) division patterns evolve with age. We trained an artificial neural network (ANN) to accurately identify cell types in the hematopoietic hierarchy and predict their age from their gene expression patterns. We then used this ANN to compare daughter cell identities immediately after HSC divisions, and found that HSC self-renewal declines sharply with age. Furthermore, while HSC cell divisions are deterministic and intrinsically-regulated in young and old age, they become stochastic and niche-sensitive in mid-life. Analysis of evolving division patterns indicated that the self-renewal ability an individual HSC depends upon number times it has divided previously. We propose a model of HSC proliferation that accurately predicts the accumulation of HSCs with age, and reconciles the stochastic and instructive views of fate commitment.