Predicting Clinical Drug Combination Efficacy Using In Vitro Monotherapy Data

Drug combinations are a cornerstone of cancer therapy, but the vast number of possible drug combinations makes it infeasible to screen them all experimentally when identifying new therapies—for example, testing all possible 4-drug combinations for 200 compounds in 100 cell lines would require more than 6 billion experiments, each requiring multiple drug concentrations and replicate measurements. To overcome this problem, efforts have been made to develop computational models capable of predicting drug combination efficacy to select lead candidates prior to experimentally testing them. While these models have traditionally aimed to predict drug synergy, recent evidence has emerged suggesting that many cancer drug combinations may derive their efficacy from independent drug action (IDA), where patients only receive benefit from the single most effective drug in a drug combination. In this thesis, I present my work to develop a method capable of using the IDA model to predict clinical drug combination efficacy using in vitro monotherapy data. This work resulted in the creation of IDACombo, an R package which enables IDA based predictions of drug combination efficacy using monotherapy data from high-throughput cancer cell line (CCL) screens. I show that IDACombo predictions closely agree with measured drug combination efficacies both in vitro (Pearson's correlation = 0.94 when comparing predicted efficacies to measured efficacies for >5000 combinations) and in a systematically selected set of clinical trials (accuracy > 88% for predicting PFS/TTP or OS benefit in 26 first line therapy trials). This work provides a framework for translating monotherapy cell line data into clinically meaningful efficacy predictions for hundreds of thousands of 2-drug combinations and millions of combinations of 3 or more drugs.