Deciphering suitable cancer drug combinations

Saez-Rodriguez group investigates drug combos

Deciphering suitable cancer drug combinations

17 Jun 2019 - 10:31


  • Crowdsourcing project explores fundamental traits that underlie effective drug combinations for cancer cells
  • The data and computational methods suggested are freely available online
  • The work could prove useful for predicting how effective a previously tested drug combination would be on a specific patient

June 18, Cambridge – DREAM Challenges are open science initiatives that invite researchers to propose solutions for fundamental biomedical questions. A recent Nature Communications paper describes the results of the AstraZeneca-Sanger Drug Combination Prediction challenge, which explored fundamental traits that underlie effective combination treatments. The data, method description and source code produced by participants are available on the challenge webpage.

Personalised treatment

Personalised treatment targeted for a tumour’s genetics has been known to result in remarkable responses, but patients often relapse. Drug combinations have the potential to overcome drug resistance and reduce the rate of relapse. It is still unclear, however, what combinations work best on which patients.

Using computational methods to predict how a cancer cell would react to a drug combination has the potential to guide treatment decisions and become a useful tool for precision medicine.

Fundamental questions

This Drug Combination Prediction DREAM challenge was organised in 2015 by AstraZeneca, the Wellcome Sanger Institute, the European Bioinformatics Institute (EMBL-EBI) and Sage Bionetworks. As part of the challenge, AstraZeneca shared 11 576 experimentally tested drug combinations on 85 cancer cell lines, by far the largest open release of such data at the time. Over 160 teams used the data to train and test models.

The participants were encouraged to explore the following themes:

  • How to predict whether a known drug combination will be effective for a specific patient
  • How to predict which new (untested) drug combinations are likely to yield interdependent behaviours in patients
  • How to identify novel biomarkers that may reveal underlying mechanisms related to drug discovery

“Developing computational techniques that could help us estimate the success of drug combinations for specific patients is essential,” says Julio Saez-Rodriguez, former Group Leader at EMBL-EBI and now Professor at Heidelberg University.

“The format of the challenge was particularly successful because it was based on an open collaboration between industry and academia, and engaged a large number of scientists,” says Michael Menden, former PhD student at EMBL-EBI, now junior group leader in the Institute for Computational Biology, Helmholtz Munich.

“By releasing these data through the DREAM Challenge, we witnessed the full power of crowd sourcing,” explains Jonathan Dry, Director of Data Science and Bioinformatics, Early Oncology, at AstraZeneca. “The project connected a vibrant community, observing over 100 new ways to approach this problem, and revealing the common ingredients to success.”

Source paper

MENDEN, M.P., et al. (2019). A cancer pharmacogenomic screen powering crowd-sourced advancement of drug combination prediction. Nature Communications. Published online DD MM; DOI: 10.1038/s41467-019-09799-2

Contact the news team

Oana Stroe
Communications Officer
+44 (0)1223 494 369

Subscribe to the e-mail newsletter
Get a monthly round-up of the hottest news and features from EMBL, straight to your inbox.
Or stay updated with the RSS feed (EMBL-EBI only).