Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

Bioinformatics, Computational and Systems Biology

Bioinformatics, Computational and Systems Biology - Poster Session E

Poster L13 - Development of an automated machine learning-driven assay for robust and accessible analysis of cell migration

Saturday, October 26, 2024
10:00 AM – 11:00 AM EST
Location: Exhibit Hall E, F & G

Introduction:  Migration is a fundamental cellular process that plays a crucial role in the understanding of development, wound healing, immune response, and cancer metastasis. Current methods of quantifying 2D cell migration combine traditional scratch assays with the analysis of aggregate data such as migratory rate and wound closure area. However, traditional scratch assays lack uniformity because wounding is reliant on the skill of the user. Additionally, in recent years, a push for higher resolution data has led to a desire for capturing migration dynamics at the single-cell level. Tools such as Cellpose (CP) have used forms of artificial intelligence to improve label-free cell segmentation and allowed researchers to capture the morphology of cells with high temporal resolution, but tracking of cells and resultant analysis is still time-consuming.

In this work, we consider the more robust cell exclusion zone assay as an alternative to the traditional scratch assay and seek to improve quantification and analysis of cell migration with a modular Cellpose (CP) segmentation model and altered Linear Assignment Problem (LAP) tracking algorithm found in the ImageJ TrackMate (TM) plugin (Figure 1). The segmentation and tracking processes are automated for ease-of-use and versatility between cell types and 2D migration assays utilized. The interchangeable and automated nature of this combined methodology improves cell tracking quantification and analysis to observe single-cell migratory dynamics with greater accessibility, replicability, and higher throughput.

Materials and

Methods:  Cell exclusion zone assays were completed with 10:1 ratio SYLGARD®184 polydimethylsiloxane (PDMS) cured in a vacuum oven and room-temperature-vulcanizing silicone rubber (RTV SR) cured at room temperature for 72 h. Rectangular and circular barriers were cut from bulk material then washed and sterilized under a UV light or in an autoclave for PDMS and RTV SR respectively. Both assays are seeded with Tyk-nu ovarian cancer cells and a wash step is completed after wounding or removal of barriers. Live-cell imaging was conducted at 5% CO2 and 37 °C.

An iterative human-assisted training methodology using the CP “cyto3” base model and images taken at 4x with varied lighting and contrast were used to develop the tyk-nu_cyto3 CP segmentation model. Segmentation reaching human agreement (> 90% detection accuracy) was attained at 5 training loops.

Cell tracking is calculated via TM’s LAP algorithm. Track features for track quality, track length, and total displacement filter initial tracking results by removing tracks that fall below 2 standard deviations under the means of at least two of the three previously mentioned features. The number of tracks kept, track quality, and track length determine a tracking quality parameter that determines the ideal input values from a range of the following customizable variables: max frame gap, alternative linking cost factor, linking max distance, and gap closing max. Spot, edge, track, and Tracking Quality data are returned to the user as .csv files. The TM model and TM tracks are returned as .xml files for review.

Results, Conclusions, and Discussions: All rectangular and circular RTV SR barriers detached before or during the 24 h incubation periods and both PDMS and RTV SR circular barriers covered < 87% of the true punch area of .0314 cm2, due in part to their elasticity (Figure 2). However, while the traditional scratch assay indicated a significant difference between theoretical and actual wound size in rectangular experiments, both PDMS and RTV SR did not in either experiment, indicating an ability to create predictable wounds if the barrier itself is shaped properly (Figure 2C). Therefore, the ideal 2D migration assay involves a traditional scratch assay for its ease-of-use or a rectangular cell exclusion zone assay with a PDMS barrier. Subtractive fabrication or pre-forming with molds would improve the uniformity of barriers and allow for bulk processing.

Additionally, Directed Migration (DM) and Directed Migration Ratio (DMR) features are analyzed. Directed Migration is calculated as a projection of Total Displacement onto a unit vector in the direction of the wound. Directed Migration Ratio is described as the ratio between Directed Migration and Total Displacement. When comparing PDMS and traditional wounding, PDMS tracks had significantly different mean DM, DMR, instantaneous velocity, and track displacement than scratch tracks. However, Cohen’s d values report a difference in means of less than 0.125 pooled standard deviations for all features (Figure 3B).

While the plugin reduced the computational burden of tracking and analysis, it is reliant on inaccessible high quality live-cell imaging. Tracking data often requires high computing power for timely analysis. We hope future iterations of this assay may become more accessible through an analysis of time-point reduction, variable dimension reduction, and the use of lower resolution images.

We have provided an accessible and adaptable ImageJ plugin and alternative migration assay to improve current methods of 2D migration analysis. Future improvements may include the ability to analyze circular wounds or classify diseased cells. We are hopeful that this automated method provides researchers with a more accessible tool to track a variety of cell types with minimal human input by reducing time spent on manually fixing cell tracks by improving track accuracy.

Acknowledgements (Optional):  We thank W. Murfee, A. Andrews, and C. Casson for providing thoughtful feedback. We thank M. Tabatabaie and L. Williams for the help and materials provided for developing the PDMS constructs. All figures were created with BioRender.com. Research reported in this publication was supported by the UF Health Cancer Center, supported in part by state appropriations provided in Fla. Stat. § 381.915 and the National Cancer Institute of the National Institutes of Health under Award Number P30CA247796.  The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or the State of Florida. This work was in part supported by funding from the National Institutes of Health (NCATS 1KL2TROO1429).