CellVoyant, a U.K.-based startup, has launched an AI platform that allows scientists to predict the future health and performance of human cells based on their appearance in microscope images—a breakthrough that could sharply reduce the cost of cell-based medical therapies.
A growing number of treatments for diseases ranging from cancer to Parkinson’s to diabetes depend on modified human cells as the core component of the treatment. These include CAR-T, where a patient’s own immune cells are extracted, genetically reprogrammed outside the body so that they can recognize and destroy cancer cells, and then reinjected into the patient. The category also includes various stem cell-based therapies.
Currently, these cell-based therapies are among the most expensive medications available. CAR-T treatments, for example, often cost hundreds of thousands of dollars per dose.
Part of the reason they are so costly is due to the labor-intensive, highly sensitive, and relatively wasteful process that is required to produce them. Scientists have to culture a lot more cells than they need, because some of the cells will not be healthy enough or exhibit the right properties to make a good treatment dose. Although it is possible to determine some aspects of cell health simply by “eyeballing” the living cells under a microscope, scientists usually cannot gauge which cells are best without performing tests on samples, usually killing those cells being tested in the process. Then they have to hope that the cells they tested are actually representative of the other cells in the culture, which is not always the case. And while tests assess a cell’s current condition, they can’t predict how the cell will develop in the future. Sometimes entire cell lines fail to perform and have to be discarded. Sometimes an entire treatment dose winds up being ineffective. All of this adds to the cost of the treatments.
CellVoyant’s new product, a platform called FateView, aims to significantly reduce the waste in this process by using AI models it has trained to classify cells by their current qualities and, critically, predict which cells will possess the right qualities in the future, simply by analyzing microscope-based imagery of the cells that use regular, white light. Currently the platform can do this for 10 different cell types—including stem cells, T-cells, cardiac cells, and blood cells—and the company is planning on training its models to work with more in the future.
