Deep learning predicts blood stem cell aging from 3D DNA structure images

An AUROC of 0.77 ± 0.03 demonstrates the ability to distinguish between young and aged hematopoietic stem cells using a deep learning approach.

• Alterations in chromatin architecture are associated with the ageing process in hematopoietic stem cells.
• A deep learning model, ChromAgeNet, learns spatial features of chromatin from 3D images of cell nuclei.
• This model outperforms traditional machine learning methods based on handcrafted chromatin features.
• Key predictive markers identified include chromatin entropy, peripheral heterochromatin, and chromatin condensates.
• The model may serve as a tool for screening aged stem cells treated with epigenetic drugs to detect rejuvenation.

Simplified