Researchers at the University of California, San Francisco (UCSF), in partnership with a team at IBM Research, have built a virtual molecular library of thousands of “command sentences” for cells, based on pairings of “words” that assisted engineered immune cells to seek out and relentlessly destroy cancer cells. These “command sentences” are based on methods for machine learning that are newer than those previously used.
This is the first time that such powerful computational tools have been used to a discipline that up until now has evolved mostly through ad hoc experimentation and designing cells with existing rather than synthesizing molecules. The work was published online in Science on December 8, 2022.
Because of this breakthrough, researchers are now able to determine which components, whether natural or synthetic, should be included in a cell in order to give it the specific behaviors that are necessary to properly respond to complicated illnesses.
According to Wendell Lim, Ph.D., the Byers Distinguished Professor of Cellular and Molecular Pharmacology at UCSF, who also serves as the director of the UCSF Cell Design Institute and was the primary investigator of the work, the research is a critical change for the discipline. The researchers can only get to a situation where they can rapidly build new cellular treatments that carry out the necessary behaviors if they have that ability of prediction.
A significant portion of the process of therapeutic cell engineering includes selecting or manufacturing receptors that, if added to the cell, will make it possible for the cell to carry out a specific function. Molecule known as receptors cross the plasma membrane of cells in order to gather information about their environment and send those findings back to the nucleus as instructions for how the cell should function in response to its surroundings.
A kind of immune cell known as a T cell can be reprogrammed to detect and destroy cancer cells if the appropriate receptor is introduced into the cell. These chimeric antigen receptors, often known as CARs, have been shown to be effective against certain types of cancer, but not others.
Lim and Kyle Daniels, Ph.D., a researcher working in Lim’s laboratory, focused their attention on a component of a receptor that is found inside the cell and is known as a motif because it contains sequences of amino acids in a repeating pattern. Each motif performs the function of a command “word,” so controlling an action that takes place within the cell. The way in which these words are put together to form a “sentence” will influence the orders that are carried out by the cell.
Many of today’s CAR-T cells are created with receptors that instruct them to destroy cancer cells but also to pause after a brief period of time. This is analogous to telling the cells, knock out some renegade cells and then take a break. As a direct result of this, the tumors are able to keep growing.
The group had the belief that by combining these “words” in a variety of different ways, they would be able to develop a receptor that would provide the CAR-T cells the ability to complete the task without pausing for a break. They created a library of roughly 2,400 command sentences that were randomly mixed and tested hundreds of these randomly combined command sentences in T cells to see how successful they were in attacking leukemia.
After that, the researchers formed a collaboration with Simone Bianco, Ph.D., a computational biologist who, at the time of the study, was working as a research manager at IBM Almaden Research Center. Currently, she is the Director of Computational Biology at Altos Labs. Bianco and his team of researchers, which included Sara Capponi, Ph.D., who was also working at IBM Almeden and Shangying Wang, Ph.D., who was working as a postdoc at IBM at the time but is now working at Altos Labs, applied novel machine learning methods to the data in in order to produce entirely new receptor sentences that they anticipated would be more successful.
The researchers modified some of the terms in the sentence, and in doing so, they gave it a new meaning. The new phrase instructed the T cells to kill those renegade tumor cells out, and continue at it. As a result, the researchers predictively built T cells that destroyed cancer without taking a break.
The new research model that is created by combining cellular engineering and machine learning is very useful.
Bianco asserted that the whole is obviously bigger than the sum of its parts. It enables the researchers to obtain a greater idea not just of how to create cell treatments, but also of the laws underlying life itself and how life forms do what they do.
Capponi continued by saying that as a result of the accomplishments of the research, they will apply this method to a varied collection of experimental data and perhaps rethink T-cell design.
The investigators are of the opinion that this method will result in cell treatments that can be used for regenerative medicine, autoimmune, and other purposes. Daniels is enthusiastic about the prospect of developing self-renewing stem cells, which would do away with the demand for donated blood.
The true strength of the computational method lies not only in the ability to make command phrases but also in the capacity to comprehend the grammar of the molecular instructions.
It is the key to developing cell treatments that do precisely what the researchers want them to do. Cell therapies do exactly what the researchers want them to do. Using this method makes it easier to make the transition from comprehending the science to developing its implementation in the actual world.
Kyle G. Daniels et al. (2022). Decoding CAR T cell phenotype using combinatorial signaling motif libraries and machine learning, Science. DOI: 10.1126/science.abq0225. www.science.org/doi/10.1126/science.abq0225