Poster
Building Machine Learning Aging Models on High-Content Imaging Data from Primary Human PBMCs
May 13, 2024
•
10 minutes
Cell-based models of aging could enable high throughput screening of potential aging interventions. Machine learning-based analysis of high-content cellular imaging is a scalable, cost-effective method for morphological profiling and phenotypic screening.
Here, we utilize Cell Painting, a set of morphological dyes that stain for different organelles, to analyze the morphological features of primary human PBMCs derived from 224 donors of ages 19-79. We utilize our custom ML-based software to build aging models between young (<35 yrs) and old (>60 yrs) PBMC samples.
We then analyze the cell types and features that contribute to the accuracy of the aging models. We find that (1) we can differentiate age groups, (2) lymphocytes carry more aging information than monocytes, and (3) field level images carry a similar amount of signal as individual cells.
Here, we showed that we can successfully build aging models on primary human PBMCs. We highlight tradeoffs in performance and interpretability for different modeling strategy (e.g. field-level vs cell-level).
Finally, we identify age-related trends in lymphoid cell morphology. Future studies will further elaborate functional T cell types and metabolic states identifiable by imaging that correlate with immunosenescence for improved in vitro aging models.
Authors: Daniel Chen, Hope O’Donnell, Francesco Rubbo
Presenter: Daniel Chen