Classification of Effects of Drug Combinations with Support Vector Machines

Cancer is still one of the challenging diseases to develop new therapies due to the late diagnosis and its complex progression nature. There is an urgent need for new therapy regimes for cancer patients having late stage diagnosis or recurrence. New computational approaches can help to identify more effective drug combinations as new treatment options for cancer. For this purpose, we developed a classification method to identity more effective drug pairs out of all possible combinations by using single drug treatment gene expression and biological network data. A support vector machine was trained with new features. The model was evaluated on a real drug treatment data that contains both positive (more effective) and negative (not effective) drug combinations. The classification performance reached 80% average accuracy on the test data. Although these results are promising, the model has a room for improvement with different extensions.