Skip to main content

Informatics approaches for the rational selection of personalized cancer drug combinations

Periodic Reporting for period 2 - DrugComb (Informatics approaches for the rational selection of personalized cancer drug combinations)

Reporting period: 2018-12-01 to 2020-05-31

Each year we spend more than 100 billion euros on cancer medicine, however, the efficacy is far from optimal, as many cancer drugs benefit at most 25% of the patients who take them. Likewise, pharmaceutical companies have tested hundreds of cancer drugs in clinical trials, but more than 90% of them have failed. These disappointing results contribute to a surging cost of 2.6 billion euros for a drug to reach to the market. Therefore, the current treatment efficacy is largely short-term. The society would critically need more effective treatments to ensure a better quality of life for patients while keeping the cost at a sustainable level.
Despite the scientific advances in the understanding of cancer, there remains a major gap between the vast knowledge of molecular biology and effective anticancer treatments. Even when there is an initial treatment response, cancer cells can easily develop drug resistance. To reach effective and sustained clinical responses, many cancer patients who become resistant to standard treatments urgently need multi-targeted drug combinations. Drug combination therapy has the potential to enhance efficacy, reduce dose-dependent toxicity and prevent the emergence of drug resistance. However, discovery of synergistic and effective drug combinations has been a laborious and often serendipitous process. In recent years, identification of combination therapies has been accelerated due to the advances in high-throughput drug screening, but a more systems-level approach is needed. This project aims to accelerate the discovery of personalized multi-targeted drug combinations using computational approaches to (i) predict and prioritize the most effective drug combinations and (ii) to evaluate the degree of synergy in the drug combination experiment and (iii) to understand and translate the mechanisms of drug combinations into treatment suggestions for patients. Through my close connections with leading experimental and clinical researchers, the proposed computational analysis pipeline has exceptionally high potential to lead to novel, more effective and safe treatments compared to the current cytotoxic and single-targeted monotherapies.
1) We have developed a crowd-sourcing based web portal called DrugTargetCommons ( to manually curate and annotate drug-target activity data from publications. Currently there are 204,901 bioactivity data points for 4,276 compounds and 1,007 targets which are manually curated [1-2]. The drug-target activity data has been applied in the recent DREAM Challenge to predict drug combination responses for cancer cell lines [3].
2) To understand the interactions of drugs and cancer cells, we have developed a network pharmacology approach that utilizes dynamic modeling of signaling pathways. The model has been successfully validated in triple-negative breast cancer cell lines and has led to the discovery of novel combinations of Aurora B and ZAK inhibitors [4]. We are now validating the drug combinations using a larger cohort of cancer cell lines. Furthermore, we have developed a computational tool called CES (Combined Essentiality Score) to predict the potential drug targets from functional genetic screens [5].
3) To make the drug combination screening data FAIR (Findable, Accessible, Interoperable and Reusable), we have developed a web portal called DrugComb ( where the results of drug combination screening studies for a large variety of cancer cell lines are accumulated, standardized and harmonized [6]. To facilitate the data analysis, we have developed a CSS score to evaluate the drug combination sensitivity [7]. The CSS scoring tool and other previously developed drug synergy scoring tools (ie. SynergyFinder), are made freely available in the DrugComb portal for the community of cancer and drug researchers. To evaluate the statistical significance of drug response data, we have developed a numerical method to evaluate the transition probability of cell growth using birth-and-death models [8].
4) Together with my collaborators in the clinics, we have successfully applied these computational tools to evaluate the potential drugs and drug combinations for cancer patients. For example, we have utilized the network pharmacology model to predict novel drug and drug combinations using ex-vivo patient samples of T-PLL (T cell prolymphocytic leukemia) [9-10]. We have also analyzed the drug combination screen data using the SynergyFinder approach and identified synergistic combination of dasatinib or everolimus with paclitaxel in ovarian cancer patient samples [11]. More recently, we have analyzed flow-cytometry-based drug response data in multiple hematopoietic cell populations and identified drugs that are specific to distinct cell lineages [12].
[1] Tang et al. doi: 10.1016/j.chembiol.2017.11.009. [2] Tanoli et al. doi: 10.1093/database/bay083 [3] Menden et al. doi: 10.1038/s41467-019-09799-2 [4] Tang et al. doi: 10.1038/s41540-019-0098-z. [5] Wang et al. doi: 10.1016/j.ebiom.2019.10.051. [6] Zagidullin et al. doi: 10.1093/nar/gkz337. [7] Malyutina et al. doi: 10.1371/journal.pcbi.1006752. [8] Pessian and Tang. [9] He et al. doi: 10.1158/0008-5472. [10] Andersson et al. doi: 10.1038/leu.2017.252. [11] Haltia et al. doi: 10.1016/j.ygyno.2016.12.016. [12] Majumder et al. doi: 10.3324/haematol.2019.217414.
This project will develop computational systems pharmacology approaches to predict, test and understand drug combinations that go beyond the state-of-the-art in many ways. First, many new promising targeted therapies fail in clinical trials due to lack of efficacy, most likely because we have limited understanding on which patient subpopulations are expected responders and what are the most predictive molecular biomarkers for treatment response. The state-of-the-art patient stratification is based on either clinical phenotypes or genomics signatures but they do not necessarily predict drug responses. In this project, we have developed computational approaches that utilize ex vivo drug sensitivity data to stratify patient-derived samples. Secondly, while the next generation sequencing has been very successful at characterizing the heterogeneity associated with each cancer type, these findings often do not lead to therapeutic targets that can be utilized in drug combination strategies. I have developed the informatics approaches to understand drug target interaction and utilized such information in network pharmacology models to establish the mechanistic link between cancer genomics and drug sensitivity, based on which we can predict clinically actionable and synergistic drug combinations. Thirdly, we have applied the proposed drug combination discovery pipeline not only to the established cancer cell lines, but also to cancer patient samples. We have developed efficient computational tools for evaluating the significance of drug combination experimental data, which is needed for demonstrating that the drug combination predictions can be translated into treatment suggestions. The project as a whole is expected to provide widely applicable research tools, helping to fill an important gap in personalized medicine. All the informatics tools will be available in the DrugComb data portal at:
The DrugComb Data Portal Main Page
The DrugComb Prediction and Data Analysis Results