Deep Learning Predicts Hematopoietic Stem Cell Aging From 3D Chromatin Images

The functional decline of the hematopoietic system during aging affects organismal function and contributes to reduced healthspan. Quantifying hematopoietic aging holds great scientific and clinical relevance. Alterations in chromatin architecture are a well-established hallmark of aging that encode rich and informative signatures of the aging process, yet they remain largely unexplored as quantitative markers. Here, we present an interpretable deep learning approach based on convolutional neural networks, Chrom…

By the Truvace deskOctober 16, 20262 min read1 sourcePeer-reviewedTRV-2026-1512· cite & verify

Deep Learning Predicts Hematopoietic Stem Cell Aging From 3D Chromatin Images — Scanning Electron Microscope image of contents of pendant, showing unprocessed flax fibres by The British Museum, Ian Richardson, 2012-06-14 16:21:56. CC BY-SA 4.0 · https://creativecommons.org/licenses/by-sa/4.0

In brief

Researchers built ChromAgeNet, a convolutional neural network, and trained it on 3D microscope images of DAPI-stained nuclei from murine hematopoietic stem cells. The model learned to tell young cells from aged ones and scored better than classical machine learning models using handcrafted chromatin features.

Quantifying blood stem cell aging has scientific and clinical relevance, and chromatin architecture had been largely unexplored as a quantitative marker. The work is a proof of concept in mouse cells, and its use for screening rejuvenation drugs was only preliminarily evaluated.

Main points

The model is an interpretable convolutional neural network named ChromAgeNet.
It was trained on 3D images of DAPI-stained murine hematopoietic stem cell nuclei.
Reported performance was an AUROC of 0.77 ± 0.03 for young versus aged cells.
Explainable AI identified chromatin entropy, peripheral heterochromatin, and chromatin condensates as predictive markers.
Aged cells treated with epigenetic drugs were used as a proof-of-concept screen for rejuvenation.

The gain

A convolutional neural network trained on 3D images of stained mouse stem cell nuclei separated young from aged cells better than classical models and flagged interpretable chromatin markers of aging.

The rundown

The study starts from the observation that functional decline of the hematopoietic system during aging affects organismal function and contributes to reduced healthspan. Chromatin architecture changes are a known hallmark of aging, but they had rarely been used as quantitative markers.

ChromAgeNet learns spatial features of chromatin from images instead of relying on handcrafted features. Explainable AI techniques were then applied to identify which chromatin properties drive predictions, namely entropy, peripheral heterochromatin, and condensates.

As a proof of concept, the authors assessed the model as a phenotypic screening tool for aged stem cells treated with epigenetic drugs to detect rejuvenation. They propose it could support high-throughput drug screening for rejuvenation therapies.