Contrastive learning of cell state dynamics in response to perturbations

Soorya Pradeep, Alishba Imran, Ziwen Liu, Eduardo Hirata-Miyasaki, Taylla Milena Theodoro, Ivan E. Ivanov, Madhura Bhave, Sudip Khadka, Hunter Woosley, Carolina Arias, Shalin B. Mehta

Keywords: contrastive learning, dynamics, cell biology

TL;DR: Integrating cell tracking with contrastive learning enables self-supervised embedding of cell dynamics for sensitive classification of cell states

Abstract: We introduce dynaCLR, a self-supervised framework for modeling cell and organelle dynamics via contrastive learning of representations of time-lapse datasets. Live cell imaging of cells and organelles is widely used to analyze cellular responses to perturbations. Supervised modeling of dynamic cell states encoded in 3D time-lapse data is laborious and prone to bias. dy- naCLR leverages single-cell tracking and time-aware contrastive sampling to map images of cells at neighboring time points to neighboring embed- dings. We illustrate the features and applications of dynaCLR with the following experiments: analyzing the kinetics of viral infection in human cells, detecting transient changes in cell morphology due to cell division, and mapping the dynamics of organelles due to viral infection. Temporally regularized embeddings computed with dynaCLR models enable efficient and quantitative annotation, classification, clustering, or interpretation of the cell states. The models reliably embed, i.e., generalize to, data from un- seen experiments with different microscopes and imaging contrasts. Models trained with dynaCLR consistently achieve > 95% accuracy in mitosis and infection state classification, enable the detection of transient cell states and reliably embed unseen experiments. dynaCLR provides a flexible framework for comparative analysis of cell state dynamics due to perturbations, such as infection, gene knockouts, and drugs. We provide PyTorch-based implementations of the model training and inference pipeline and a napari plugin user interface for the visualization and annotation of trajectories of cells in the real space and the embedding space.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 2024