This talk discussed the implications of combining microscopy with a FACS (Fluorescence-activated cell sorter) system. The FACS system analyses and isolates cells based on biomarkers and spatial information, while microscopy offers high-content information and spatial information at the single-cell level, providing insight into cell interactions. Yu-Hwa Lo, Professor, at UC San Diego Health, suggested that combining these tools into one modality requires leveraging AI, high-throughput technology, and spatial information.
In collaboration with Nanocellect, Lo used their Verlo System to scan cells and create images for analysis. Lo mentioned that the cells are sorted at a very low sorting pressure of 2 PSI with minimum disruption or stress on the cell. The cells that are sorted are highly viable, so that they are suitable for downstream analysis and use.
Lo gave a more in-depth overview of the workflow. He explained that the technology generates a label-free image of forward scattering or backscattering. In addition, one can create nine different fluorescent and overlay them to give detailed spatial information for individual cells. The cell of interest can then be isolated for downstream analysis. The Verlo System can support up to 12 images and each image has 15 parameters. Furthermore, it supports real-time sorting at rates up to 1000 cells per second.
The use of AI, particularly convolutional neural networks (CNNs) is essential for handling millions of high-content images. The loss function is minimised by comparing the AI-generated image with the original input image. Lo said that the system supports both supervised and unsupervised learning. The supervised learning use case showed that after training the AI achieved 100% accuracy in classifying known cell types like monocytes and granulocytes. For unsupervised learning, the AI detected subtle phenotypic changes in cells with an average accuracy of 85%. Alongside this, the unsupervised model reached 90% accuracy in predicting cell behaviour. Factors in the cell behaviour category included proliferation rate, activity rate, and genetic stability.
Breaking into the 3D world is important because moving from 2D to 3D imaging enhances morphological information. Lo used his 3D imaging flow cytometer to create 1000 cells in 3D images. He used 3D imaging to assess DNA damage on glioblastoma cells from radiation therapy. To sum up, the presentation showcased a novel platform that merges high-throughput imaging, AI, and cell sorting.
