Systems Tumor Immunology: It's Big Blue Beats Kasparov All Over Again

One innovative approach to cancer therapy involves isolating immune cells (tumor-infiltrating lymphocytes or TILs) from a tumor, and then trying to expand and activate them against tumor antigens in the lab. It can work, but in practice, not an efficient practice. TILs are a mixed bag of cells with all sorts of different personalities, some of which may or may or not be interested in attacking tumor cells. Some, recruited by the ever-devious tumor cells, can even work to supress the activity of the good guys. So sometimes you get a bunch of TILs that work, sometimes not.

A new paper describes a systems biology approach to tackling the complexity of TIL populations with the aim of predicting anti-tumor activity. They profiled the reactivity of a ton of TIL populations, and correlated this inforamtion with a battery of surface markers. They fed the info into some sort of machine learning algorithim, which came up with some pretty clear-cut boolean style predictive rules that a mere organic being (aka tumor immunologist) could never have possibly concieved of with a million years of deductive reasoning and experimental testing.

Check it out:
Rule 1) If the CD8⁺CD28^-CD152^- subpopulation constitutes less than 43% of the entire TIL population AND the CD94+constitutes less than 0.4% of the entire TIL population, then the TIL population is tumor-reactive.

Also, by manipulating the relative proportions of different TIL subpopulations, they could affect reactivity as predicted by their adding machine.

Wicked. Who knew an abacus could do immunology?

See: Predicting and controlling the reactivity of immune cell populations against cancer