endpts.com/focused-on-the… One of many companies trying to make cell therapy more efficient.
Cell therapy means taking cells out of a patient or donor, genetically modifying them, and putting them back in to treat a disease. CAR-T therapy does this with immune cells and has remarkable results in some cancers. But it costs $100k to produce one dose.
The potential patient population is huge; the world’s production capacity is tiny.
You effectively have to produce a custom “drug” in the hospital, for each patient.
You effectively have to produce a custom “drug” in the hospital, for each patient.
With $20B flowing into the industry last year and investment numbers continuing to rise, a lot of people are going to lose a lot of money unless cell therapy companies figure out how to serve more than their present ~5000 patients/year.
Obviously, everyone who works in the field knows that, so there’s a lot of competing approaches to solving the manufacturing problem.
Cellares produces a machine, the Cell Shuttle, that does automated end-to-end cell therapy production. Patient puts in cells, machine spits out a finished cell therapy.
A competing approach is allogeneic cell therapy: using donor cells rather than the patient’s own cells, which requires some extra fiddling to prevent graft-vs-host issues.
The thing about manipulating live mammalian cells is that lots and lots of variables affect how they grow and “perform” in the body, and nobody has accurate predictive models yet.
The world of designing and optimizing the manufacture of cell therapy is a *great* application for machine learning.
We can observe cells quite well, in great detail — under a microscope, through single-cell expression profiling, through immunoassays, etc.
How many cells does your machine kill?
How much does your gene transfection platform “stress” cells, changing their gene expression such that they are less effective against cancer or cause more side effects?
All measurable. All nearly impossible to predict a priori.
How much does your gene transfection platform “stress” cells, changing their gene expression such that they are less effective against cancer or cause more side effects?
All measurable. All nearly impossible to predict a priori.
What happens to cells if you tweak the settings in your cell therapy making machine? Can you efficiently search for the most “cell-friendly” settings configuration? Perhaps this will even depend on the patient/donor!
The extreme personalization and unpredictability of cell therapy manufacturing, combined with the obvious financial incentive to get efficiency up, and the availability of very-high-content, cell-specific measurement tech, means this is an unusually good fit for ML approaches.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
