Pathway-Based Drug Combinatory Synergy Prediction Using Gene Expression And Essentiality Data

Background: Experiment based drug combinatory synergy discovery is limited in a moderate scale. Systems biology models can substantially widen the search space, and discover novel mechanisms. While most of systems pharmacology models were designed to optimize the predictive performance for the drug combination synergy, they were limited by their lack of mechanistic interpretations. The gene essentiality data generated from RNAi screening have not been integrated into the systems pharmacology model yet. Methods: Drug combinatory synergy data from NCI-ALMANAC was used in this research. Cancer cell line gene expression and essential score (RNAi screening from DepMap portal), drug targets and KEGG pathways were used in our systems pharmacology model. Two types of models were constructed. The first model predicts a drug combination9s synergy using various pathways among various cell lines. The second model predicts various drug combinatory synergy in a cell line. Machine learning approaches, such as SVR, Ridge regression, Random forest and Logistic regulation were investigated in developing this model. Our predictive models were built upon 43 cell lines with both baseline gene expression and essential scores, 165 KEGG pathways, and 198 drug pairs who have at least synergistic drug combinatory effects in 3 or more cell lines. Results: In predicting drug combinatory synergy from both predictive models, the mean R square was merely 0.006 using individual gene expression or essentiality data. The mean R-square increased to 0.022 when using pathways. In drug combination specific predictive model (model 1), 64 and 62 drug combinations were related to more than 10 pathways which had enriched essential and expression genes, respectively. The correlation between expression enrich pathways and essential enriched pathways is only 0.04. In the cell line specific predictive model (model 2), drug combination synergy score from 14 cell lines can be predicted with more than 150 pathways, and 5 cell lines can be predicted with less than 10 pathways. Conclusion: Pathway-based models are more powerful than gene level-based models in predicting drug combination synergy. Gene expressions and essential genes provide much different information in predicting drug combination synergy. Drug combination synergy prediction and its related pathways are dramatically different among different cancer cell lines. Citation Format: Jin Li, Xue Wu, Yang Huo, Enze Liu, Zhi Zeng, Lijun Cheng, Lang Li. Pathway-based drug combinatory synergy prediction using gene expression and essentiality data [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 4397.