Capturing cell heterogeneity in representations of cell populations for image-based profiling using contrastive learning DOI Creative Commons
Robert van Dijk, John Arévalo, Mehrtash Babadi

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2023, Номер unknown

Опубликована: Ноя. 16, 2023

Image-based cell profiling is a powerful tool that compares perturbed populations by measuring thousands of single-cell features and summarizing them into profiles. Typically sample represented averaging across cells, but this fails to capture the heterogeneity within populations. We introduce CytoSummaryNet: Deep Sets-based approach improves mechanism action prediction 30-68% in mean average precision compared on public dataset. CytoSummaryNet uses self-supervised contrastive learning multiple-instance framework, providing an easier-to-apply method for aggregating feature data than previously published strategies. Interpretability analysis suggests model achieves improvement downweighting small mitotic cells or those with debris prioritizing large uncrowded cells. The requires only perturbation labels training, which are readily available all datasets. offers straightforward post-processing step profiles can significantly boost retrieval performance image-based

Язык: Английский

Cell Painting: a decade of discovery and innovation in cellular imaging DOI
Srijit Seal, Maria‐Anna Trapotsi, Ola Spjuth

и другие.

Nature Methods, Год журнала: 2024, Номер unknown

Опубликована: Дек. 5, 2024

Язык: Английский

Процитировано

3
Capturing cell heterogeneity in representations of cell populations for image-based profiling using contrastive learning DOI Creative Commons
Robert van Dijk, John Arévalo, Mehrtash Babadi

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2023, Номер unknown

Опубликована: Ноя. 16, 2023

Image-based cell profiling is a powerful tool that compares perturbed populations by measuring thousands of single-cell features and summarizing them into profiles. Typically sample represented averaging across cells, but this fails to capture the heterogeneity within populations. We introduce CytoSummaryNet: Deep Sets-based approach improves mechanism action prediction 30-68% in mean average precision compared on public dataset. CytoSummaryNet uses self-supervised contrastive learning multiple-instance framework, providing an easier-to-apply method for aggregating feature data than previously published strategies. Interpretability analysis suggests model achieves improvement downweighting small mitotic cells or those with debris prioritizing large uncrowded cells. The requires only perturbation labels training, which are readily available all datasets. offers straightforward post-processing step profiles can significantly boost retrieval performance image-based

Язык: Английский

Процитировано

2