Using Deep Machine Learning to Enhance Monoclonal Cell Line Identification
What is Monoclonalization?
Monoclonalization is the process by which a single cell from a heterogeneous population is isolated and expanded to create a genetically homogeneous cell culture. These types of cell populations are termed monoclonal and are considered to have maximum uniformity.
Monoclonal cell lines are used for the generation of induced pluripotent stem cells (iPSCs), therapeutic protein production including monoclonal antibody manufacturing, gene therapy, genetic engineering, drug discovery research, and a diverse array of biological research where high-quality cell lines are required.
Monoclonal cell culture is critical for many research areas however, the process to establish monoclonal cell lines is time-consuming, and many methods used to monitor monoclonality of cell culture still rely on manual and low-throughput methods. This causes scalability to remain a major bottleneck in the cell line development workflow.
The Celigo image cytometer can overcome some of the challenges for monitoring clonality of samples, as it can rapidly image and quantify cell culture in situ as well as reduce bias from manual inspection of colony growth. Equipped with brightfield imaging and four fluorescent channels, it is routinely used to monitor kinetic viability, cell health, killing, and immunofluorescence-based assays. Celigo’s ability to image standard culture plates (6-well to 1536-well plates), suspension cells, adherent cells, and 3D culture models makes it extremely versatile for use in a diverse array of experimentation.
Celigo image cytometer
The New York Stem Cell Foundation Research Institute has previously described an automated workflow to derive, identify, and characterize iPSCs that can be used for a variety of research purposes.
In this automated workflow, the team uses the Celigo image cytometer throughout multiple stages to assess cell growth, proliferation, expression of differentiation and pluripotency markers, monitoring clonal expansion, and tracking embryoid body and colony formation.
Figure 1. Schematic of the NYSCF automated cell line characterization workflow
Building off this automated workflow in which they use Celigo to image cell culture over time, the team at NYSCF developed a new deep machine learning workflow that enables the automatic detection of colony presence and can identify monoclonality from the time-course imaging captured on Celigo. This workflow, called Monoqlo, has substantial implications for the monoclonalization process.
Figure 2. Overview of the daily automation workflow which generates data for training and real-time use with Monoqlo. Note: imager graphic updated from current schematic to reflect actual imager being used.
Monoqlo incorporates four different convolutional neural networks to establish the machine learning computational workflow. The algorithm is based on the principle that cell culture inherently has directionality and cells that are deposited on day zero will either die or proliferate over time. Monoqlo uses the images captured on Celigo and analyzes them in reverse chronological order to assess the clonality of each well without manual confirmation.
Figure 3. Schematic representing a broad overview of Monoqlo’s design and algorithmic logic.
Figure 4. Zoomed representative images captured on Celigo analyzed by Monoqlo to determine clonality of a detected colony using reverse chronology.
Using this approach and a series of 5000 manually annotated images to serve as a training set, this framework is currently reporting greater than 93% accuracy across all parameters tested. Coupling this algorithm with traditional image processing strategies enables the rapid analysis of thousands of data sets acquired by Celigo in an extremely short timeframe.
This report is the first-ever instance of automating the identification of clonality using a deep machine learning object detection approach. This algorithm has the potential for infinite scalability, removing significant bottlenecks from a vast array of research areas requiring monoclonal cell line generation. Monoqlo coupled with the Celigo and its automated imaging capabilities truly has the potential to revolutionize the process of monoclonalization.
Modular deep learning enables automated identification of monoclonal cell lines
Brodie Fischbacher, Sarita Hedaya, Brigham J. Hartley, Zhongwei Wang, Gregory Lallos, Dillion Hutson, Matthew Zimmer, Jacob Brammer, The NYSCF Global Stem Cell Array® Team, Daniel Paull
bioRxiv 2020.12.28.424610; https://www.biorxiv.org/content/10.1101/2020.12.28.424610v1.full.pdf