CTD² DREAM Challenges: Develop Predictive Algorithms to Identify Effective Cancer Treatment Strategies

December 1, 2019, by Justin Guinney, Ph.D.

Crowdsourcing the analysis of highly complex and massive data has emerged as one way to incentivize and match experts from around the world to scientific problems. When crowdsourcing is done in the form of scientific competitions—or Challenges—the validation of the analytical approach is automatically incorporated into the study design. Challenges foster open innovation, creating communities that collaborate directly or indirectly to solve important biomedical problems. The Dialogue on Reverse Engineering and Assessment Methods (DREAM) Challenges are special instances of biomedical Challenges that have spawned a community of solvers committed to advancing important science questions using open and reproducible methods. Since its inception in 2006, DREAM has hosted dozens of Challenges across a wide spectrum of biomedical domains, disease areas, and data modalities that include genomics, imaging, and clinical data.^1,2

In the past, DREAM has partnered with NCI’s Cancer Target Discovery and Development (CTD²) Network to host the Gene Essentiality Prediction Challenge. The goal of this challenge was to evaluate and develop computational algorithms that predicted gene dependencies using gene expression and copy number features. This led to benchmarks and insights into factors influencing gene essentiality from functional genetic screens.³

DREAM and CTD² have partnered again to host two new Challenges: the CTD² Pancancer Drug Activity DREAM Challenge and the CTD² BeatAML DREAM Challenge.

CTD² Pancancer Drug Activity DREAM Challenge

CTD² members at the Columbia University developed Pancancer Analysis of Chemical Entity Activity (PANACEA), a comprehensive repertoire of dose dependent cellular responses and post-treatment molecular profiles to drug treatments. PANACEA covers a broad spectrum of cellular contexts representative of poor outcome malignancies, including rare ones such as GastroIntestinal Stromal Tumor (GIST) sarcoma and GastroEnteroPancreatic NeuroEndocrine Tumors (GEP-NETs).

PANACEA is uniquely suited to support DREAM Challenges related to the elucidation of drug mechanism of action (MOA), drug sensitivity, and drug synergy. Specifically, this Challenge is posing three questions or sub-Challenges:

• Inference of targets using the transcriptional data collected 24h after treatment with chemotherapeutic compounds
• Prediction of cell line compound sensitivity using baseline transcriptional profiles
• Identification of optimal compounds to sensitize three KRAS-mutant cell-lines to treatment with the MEK inhibitor selumetinib

Data provided to participants will include drug perturbational profiles from cell lines, as well as drug-sensitivity measurements from a panel of compounds. Participants are encouraged to utilize large public databases such as Connectivity Map,⁴ Cancer Cell Line Encyclopedia,⁵ and Genomics of Drug Sensitivity in Cancer,⁶ as well as insights and models developed from previous DREAM Challenges^7-9 in the development or training of algorithms. The details for the CTD² Pancancer Drug Activity DREAM Challenge can be viewed at: https://www.synapse.org/#!Synapse:syn20968331/wiki/597042.

CTD² BeatAML DREAM Challenge

Oregon Health and Science University (OHSU), in collaboration with academic medical centers, pharmaceutical, and biotechnology companies, developed the BeatAML research initiative. This study integrates molecular alterations data with ex vivo drug sensitivity for a large number of clinically annotated Acute Myeloid Leukemia (AML) cases. One of the primary goals of this multi-center study is to prioritize drugs that could yield new drug target hypotheses and discover predictive biomarkers of therapeutic response. Patient samples were subjected to whole-exome sequencing (WES), transcriptomic sequencing (RNA-seq), and ex vivo functional drug sensitivity screens.¹⁰ This rich resource enables the discovery of molecular correlates of drug response and putative patient populations most likely to respond to targeted agents. Indeed, analysis of these data has already revealed numerous correlations of drug sensitivity or resistance with a variety of mutational subsets of disease, as well as numerous gene expression signatures that correlated with drug sensitivity/resistance.¹⁰

The overall goal of the BeatAML DREAM Challenge is to define patient subpopulations tailored to specific treatments by discovering (genomic and transcriptomic) biomarkers of drug sensitivity. This Challenge is posing two sub-Challenges:

• Predict quantitative ex vivo drug sensitivity to targeted and chemotherapeutic agents using genomic alterations and gene expression data
• Stratify patients into clinical responders (i.e. those that did not have a relapse within two years of standard induction therapy) and non-responders based on ex vivo drug sensitivity data, genomic alterations, and/or gene expression data

OHSU will provide training (Beat AML waves 1 and 2) and validation (Beat AML wave 3) data. The details for the CTD² BeatAML DREAM Challenge can be viewed at https://www.synapse.org/#!Synapse:syn20940518/wiki/596265.

References

1. Saez-Rodriguez J, Costello JC, Friend SH, et al. Crowdsourcing biomedical research: leveraging communities as innovation engines. Nat Rev Genet. 2016 Jul 15;17(8):470-86. (PMID: 27418159)
2. Ellrott K, Buchanan A, Creason A, et al. Reproducible biomedical benchmarking in the cloud: lessons from crowd-sourced data challenges. Genome Biol. 2019 Sep 10;20(1):195. (PMID: 31506093)
3. Gönen M, Weir BA, Cowley GS, et al. A Community Challenge for Inferring Genetic Predictors of Gene Essentialities through Analysis of a Functional Screen of Cancer Cell Lines. Cell Syst. 2017 Nov 22;5(5):485-497.e3. (PMID: 28988802)
4. Subramanian A, Narayan R, Corsello SM, et al. A next generation connectivity map: L1000 platform and the first 1,000,000 profiles. Cell. 2017 Nov 30;171(6):1437-1452.e17. (PMID: 29195078)
5. Barretina J, Caponigro G, Stransky N, et al. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature. 2012 Mar 28;483(7391):603-7. (PMID: 22460905)
6. Iorio F, Knijnenburg TA, Vis DJ, et al. A landscape of pharmacogenomic interactions in cancer. Cell. 2016 Jul 28;166(3):740-754. (PMID: 27397505)
7. Bansal M, Yang J, Karan C, et al. A community computational challenge to predict the activity of pairs of compounds. Nat Biotechnol. 2014 Dec;32(12):1213-22. (PMID: 25419740)
8. Costello JC, Heiser LM, Georgii E, et al. A community effort to assess and improve drug sensitivity prediction algorithms. Nat Biotechnol. 2014 Dec;32(12):1202-12. (PMID: 24880487)
9. Menden MP, Wang D, Mason MJ, et al. Community assessment to advance computational prediction of cancer drug combinations in a pharmacogenomic screen. Nat Commun. 2019 Jun 17;10(1):2674. (PMID: 31209238)
10. Tyner JW, Tognon CE, Bottomly D, et al. Functional genomic landscape of acute myeloid leukaemia. Nature. 2018 Oct;562(7728):526-531. (PMID: 30333627)