Abstract
Recent advances in targeted chemotherapy have dramatically improved the survival odds of patients suffering from specific molecular subtypes of cancer. Nonetheless, while a fortunate few enjoy a durable and lasting response to treatment, the majority will eventually relapse after some grace period in remission due to the emergence of drug-resistant cancer cells.
Acquired resistance is a complex phenomenon that likely involves intra-tumoral heterogeneity, tumor-stromal interactions, and natural selection due to the pressure of chemotherapy. One promising approach to overcome resistance involves utilization of two or more drug compounds in combination, where each compound targets a different member of a signaling pathway important in the disease. By targeting more proteins, the hope is that the probability of any particular cancer cell failing to respond will decrease as the barrier to survival increases.
While drug combinations are an exciting and promising direction for chemotherapeutic research, such studies present additional challenges and opportunities. Interpretation of synergistic and efficacious drug combinations is not always straightforward and pairs of drugs may interact in surprising ways that deviate from expectation based on the molecular pathway relationships between their targets. Moreover, the question of which drugs to prioritize for testing in combination is an important one to which there is currently no standard solution. The brute force approach is not cost-effective due to the factorial growth in the number of possible higher-order drug combinations.
In this work, we present and describe an algorithmic approach to (1) predict synergistic and efficacious drug combinations based on pathway relationships between their targets and (2) classify drug combination response surfaces into canonical archetypes of greatest pharmaceutical interest. Our approach borrows from classical work in image analysis and recognition. We trained and tested our algorithm on two publicly available datasets of drug combinations. We further trained on a private dataset of cancer drug combination data and made predictions that were later experimentally validated in vitro. To our knowledge, our approach is the first proof-of-concept algorithm that utilizes the actual shapes of response surfaces to categorize and prioritize drug combinations for further testing.
Citation Format: Manway Liu, Thomas Horn, Matthew Greene, Joseph Lehar. Image-based classification of cancer drug combination response surfaces. [abstract]. In: Proceedings of the AACR Precision Medicine Series: Drug Sensitivity and Resistance: Improving Cancer Therapy; Jun 18-21, 2014; Orlando, FL. Philadelphia (PA): AACR; Clin Cancer Res 2015;21(4 Suppl): Abstract nr B51.