Related Experiment Video
Updated: Sep 19, 2025

07:26
In Situ Microscopy for Real-time Determination of Single-cell Morphology in Bioprocesses
Published on: December 5, 2019
8.0K
Unsupervised Machine Learning-Based Process Analytical Tools for Near Real-Time Cell Morphology Analysis During CAR-T
Nidhi G Thite1, Michael Yarnell2,3, Terry J Fry2,3
1Department of Chemical and Biological Engineering, University of Colorado Boulder, Boulder, Colorado, USA.
Biotechnology and Bioengineering
|June 16, 2025
Summary
Machine learning using flow imaging microscopy monitors Chimeric Antigen Receptor (CAR)-T cell production in real-time. This approach tracks cell changes, enabling early detection of process deviations and ensuring final product quality.
Area of Science:
- Biotechnology and Pharmaceutical Manufacturing
- Cell Therapy Production
- Machine Learning in Healthcare
Background:
- Chimeric Antigen Receptor (CAR)-T cell therapy utilizes living cells as active pharmaceutical ingredients, posing manufacturing challenges.
- Current quality control (QC) methods for CAR-T cells rely on end-point testing, leading to high failure rates and product heterogeneity.
- Real-time process monitoring is crucial for improving CAR-T cell production but is hindered by analytical tool limitations with heterogeneous cell products.
Purpose of the Study:
- To demonstrate unsupervised image-based machine learning as a Process Analytical Tool (PAT) for near real-time monitoring of CAR-T cell production.
- To quantitatively track morphological changes in T cells during CAR-T cell manufacturing using machine learning.
- To assess the utility of machine learning for monitoring patient-to-patient variability and detecting process deviations.
Main Methods:
- Collected flow imaging microscopy (FIM) images of T cells from nine healthy donors throughout CAR-T cell production stages (activation, transduction, expansion).
- Trained a Variational Autoencoder (VAE) model on FIM images to quantitatively track cell morphology changes.
- Correlated VAE-derived cell population density with transduction efficiency measured by traditional flow cytometry.
Main Results:
- A novel, transient cell population was identified in CAR-T cells expressing the CAR protein, absent in non-transduced cells.
- The density of this new population correlated directly with transduction efficiency, as confirmed by flow cytometry.
- The VAE model effectively tracked patient-to-patient variability in cell morphology during CAR-T cell production.
Conclusions:
- Unsupervised machine learning, specifically VAEs applied to FIM data, serves as a valuable PAT for CAR-T cell manufacturing.
- This approach enables near real-time monitoring of cell morphology, facilitating early detection of process deviations and potential failures.
- The VAE-based method can quantify patient-specific cellular changes, improving understanding and control over manufacturing variability.

