It is no use having a sequence of treatments that is as effective at killing healthy cells as cancer cells. The most elegant way to solve this problem would be to simulate the effect of each treatment tested on the healthy cells in parallel, and incorporate that effect into the score. This would then lead to the problem that resistance mutations should be applied to the cancer cells only, so the healthy steps would be skipped every other level of the tree. Which seems complicated.
A less elegant way would be to test the first and then the second treatment (in order) on the healthy cells after a treatment is chosen, and if it kills them choose the second best treatment. This seems unlikely to require too many iterations and so is likely less computationally intensive, as well as easier to implement.
It is no use having a sequence of treatments that is as effective at killing healthy cells as cancer cells. The most elegant way to solve this problem would be to simulate the effect of each treatment tested on the healthy cells in parallel, and incorporate that effect into the score. This would then lead to the problem that resistance mutations should be applied to the cancer cells only, so the healthy steps would be skipped every other level of the tree. Which seems complicated.
A less elegant way would be to test the first and then the second treatment (in order) on the healthy cells after a treatment is chosen, and if it kills them choose the second best treatment. This seems unlikely to require too many iterations and so is likely less computationally intensive, as well as easier to implement.