New Way to Spot Good Drug Pairs

Scientists have found that many single‑drug treatments for cancer do not work well. Because of this, doctors try to give two drugs together. But there are so many possible pairs that it is hard to guess which ones will help the patient. Most computer tools try to give one number that says how well a pair works. That number hides the details of how the drugs behave at different doses. When a tool only gives one score, it can’t tell if a drug pair is powerful or simply safe. The result is that the predictions are shaky and not useful for real experiments. A new computer model, called DeepSynBa, changes the game. Instead of one single score, it predicts a whole table that shows how every dose combination works.

It does this by first estimating the shape of the response surface and then filling in all the values. The researchers tested DeepSynBa on two big drug‑combination data sets, NCI‑ALMANAC and O'Neil. They compared it to the best existing tools. In most tests, DeepSynBa produced more accurate tables. It worked well even for new drug pairs, new cell lines and new types of cancer cells. Because DeepSynBa can show the entire dose grid, scientists can pick doses that give the best effect while keeping side effects low. The model also still gives a single synergy score, but that score is now based on a richer data set. The code and the data are open for anyone to use. Researchers can download them from a public repository.

Actions