Abstract 4290: ReCorDE: A novel computational framework to discover potential combinations of anti-cancer drugs

Background: Cancer is usually treated with a combination of drugs rather than a single agent. Treating cancer with a drug combination decreases the likelihood of the cancer acquiring resistance to therapy. It also allows for lower dosages of individual drugs, which reduces the occurrence and severity of side effects. Using public datasets, we investigated potential anti-cancer drug combinations using a novel framework, ReCorDE (Recurrent Correlation of Drugs with Enrichment), which identifies correlated drug response patterns for drugs with different primary mechanisms of action. Methods: 250 drugs from CTRPv2, GDSCv2, and PRISM datasets were examined using normalized logIC50 or AUC measurements. ReCorDE consists of a correlation and an enrichment step. For each dataset, we constructed pairwise drug-drug relationships using Spearman’s correlation. Combinations significant (p < 0.05) in less than two datasets after Benjamini-Hochberg (BH) correction were pruned. To turn each drug combination into a drug class combination, all drugs were mapped to their corresponding fourth-level Anatomical Therapeutic Chemical (ATC) codes. Then, using a hypergeometric test, for each class combination in the subset of significant drug combinations (the “inside” set), we tested for enrichment of that class combination in the inside set compared to the class combination universe, the distribution of class combinations for all unique pairwise drug-drug associations (250 unique drugs; 31,125 distinct drug combinations mapped to 1596 distinct class combinations). Multiple testing during enrichment was corrected using the BH method with a significance cutoff of p< 0.05. Results: We found 2764 pairs to be significant at a = 0.05 using the AUC measure. With a correlation coefficient of 0.77, vincristine and YK-4-279, an EWS-FLI1/RNA Helicase A inhibitor, had the lowest computed p-value of of 1.39 × 10−120 Adjusted p-values for >15% of these combinations had p-values < 5 × 10−6. Class combination enrichment on the same set showed 132 class pairs were enriched in the inside set compared to the class combination universe after BH correction. Taxanes/Plk1 inhibitors (OR = 69.5, adjusted p = 5.6 × 10−11); Aurora Kinase inhibitors/histone modifying agents (OR=9.2, adjusted p = 6.13 × 10−11); and pyrimidine analogues/CDK inhibitors (OR=3.43, adjusted p = 0.02) are a few notable enriched class combinations. The enrichment of these class combinations in the inside set suggests that combinations of drugs from these classes may be particularly effective in treating cancer compared to other drug combinations. Conclusions: Our framework, ReCorDE, demonstrates that finding potential drug combinations and characterizing novel, frequently perturbed pathways outside of a drug's primary mechanisms of action can be accomplished by identifying correlated drug-drug pairs from large, publicly accessible databases. Citation Format: Emily T. Ghose, August John, Huanyao Gao, Krishna R. Kalari, Liewei Wang. ReCorDE: A novel computational framework to discover potential combinations of anti-cancer drugs. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 4290.