The DREAM8 Challenges: Spring 2013
Sage Bionetworks and the DREAM Initiative are convinced that running open computational Challenges focused on important unsolved questions in systems biomedicine can help advance basic and translational science. By presenting the research community with well-formulated questions that usually involves complex data, we effectively enable the sharing and improvement of predictive models, accelerating many-fold the analysis of such data. The ultimate goal, beyond the competitive aspect of the Challenges, is to foster collaborations of like-minded researchers that together will find the solution of vexing problems that matter most to citizens and patients.
During the “Challenge season” spanning from May to September 2013 Sage Bionetworks and DREAM plan to run the following Challenges. High-level descriptions of these Challenges are provided below.
Click here to read the April 19, 2013 DREAM8 Challenge press release.
Click here to view information about the DREAM8 Challenges opened for competition.
Sponsored by the Heritage Provider Network (HPN) and the NCI Division of Cancer Biology, and in collaboration with the labs of Drs. Joe Gray and Paul Spellman from the Oregon Health and Science University, and Dr. Sach Mukherjee from the Netherlands Cancer Institute, Sage Bionetworks and DREAM are running the Heritage-DREAM Breast Cancer Network Inference Challenge in 2013. This Challenge will build upon the lessons extracted from the NCI-DREAM Challenge, run by DREAM in 2012 (with the full support of the NCI Division of Cancer Biology). In that 2012 Challenge, the work of over 50 teams throughout the 4-month Challenge period rigorously showed the feasibility to predict the sensitivity of 18 mammary cancer cell lines to 35 different therapeutic agents. However, many improvements in model prediction are still needed to make these models valuable in a clinical or drug discovery setting.
The Heritage-DREAM Breast Cancer Network Inference Challenge intends to motivate further model development in the area of signaling network inference, complementing the lessons learned in the NCI-DREAM Challenge by exploring mechanistic aspects of drug action. For this we will leverage the extensive and unique proteomics data collected under the auspices of the NCI Division of Cancer Biology by Drs. Gray, Spellman and Mukherjee in collaboration with Dr. Gordon Mills of the MD Anderson Cancer Center. This dataset was generated using Reverse Phase Protein Array (RPPA) quantitative proteomics technology, and represents the responses of 4 breast cancer cell lines to 88 therapeutic compound/stimuli pairs during a 72 hour time course (11 time points). Approximately 170 proteins and phosphoproteins were measured, yielding over 600,000 data points in total.
The goals of the Challenge will be manyfold. First, participants will be asked to use these data to build network models that represent the active pathways and their response to different stimuli during drug treatment. A second goal will be to predict the dynamic response of the different measured phosphoproteins to different drug perturbations. The third goal will be for participants to propose novel visualization strategies for these high dimensional data sets. The best performing models will be useful not only in drug development decision-making but also for understanding and guiding validation studies that determine how breast cancer drugs work or how different drugs interact with one another.
Best performing teams in the network inference and time-course prediction aspects of the challenge will be awarded monetary prizes and the possibility of subsequent validation experiments of their predictions, as well as an invitation to present at the DREAM conference with travel expenses covered. The winner of the visualization challenge will be offered the possibility of implementation of the visualization scheme to be deployed in Synapse by Sage Bionetworks software engineers. The Heritage Provided Network has generously donated prize funding. Besides these awards, the organizers will work with the editors of high impact journals to develop a partnership that will allow appropriate dissemination of the winning methodologies and results.
The NIEHS-NCATS-UNC DREAM Toxicogenetics Challenge represents a groundbreaking new direction for toxicity testing and is intended to help us understand how genetic variation affects individual response to common environmental and pharmaceutical chemicals. For this Challenge, Sage Bionetworks and DREAM are teaming up with scientists at the National Institute for Environmental Health Sciences (NIEHS), the National Center for Advancing Translational Sciences (NCATS), and the University of North Carolina at Chapel Hill (UNC). These groups have generated population-scale toxicity data in a human ex-vivo model system. To do this, they leveraged the 1000 Genomes Project, which provides public access to genotype and transcriptomic data derived from cell lines collected from thousands of individuals representing 9 distinct geographical populations with defined genetic heterogeneity. The NIEHS/NCATS/UNC team conducted the largest ever population-based ex-vivo cytotoxicity study by treating 920 of these cell lines with approximately 170 common, pharmaceutical or important environmental chemicals. The NIEHS-NCATS-UNC DREAM Toxicogenetics Challenge aims to ask the “crowd” of researchers to use the 1000 genomes genetics, genomics and cytotoxicity data to build models that can predict: (1) the toxic response of individuals to each chemical based on genetics and genomics data; (2) the parameters of distribution for the toxic effects of each chemical (e.g., mean and variance of toxic response across the population) based primarily on chemical information about the compounds being evaluated.
For every chemical that has been tested for toxicity, there are myriad others that still remain untested. Thus the best models resulting from the NIEHS-NCATS-UNC DREAM Toxicogenetics Challenge that can accurately predict either what groups of individuals will be most sensitive to chemicals or the range of toxicity for different types of chemicals, will provide powerful new tools for EPA and other government agencies to do more targeted experimentation based on computational predictions.
Best performers will be invited to present at the DREAM conference with travel expenses covered by the Sage/DREAM organizers. We are working with high impact journals to ensure that the methodology developed by the best performer will be considered for publication under the challenge-assisted peer review format.
The “DREAM Whole-cell parameter estimation challenge” is a crowd-sourcing effort organized by the DREAM project and Sage Bionetworks, in collaboration with Prof. M. Covert’s lab from Stanford University, in which the ability of the community to infer the kinetic parameters underlying biological processes will be tested. Building upon two previous DREAM challenges to infer the parameters of small biological networks, this challenge will be based on a first-of-its-kind whole-cell computational model of the human pathogen Mycoplasma genitalium [Cell, 150 2, 389-401 (2012)]. The model is unique in that it integrates diverse mathematical approaches to account for all the essential cellular processes as well as all annotated genes, resulting in a number of biological predictions that were validated experimentally. The challenge consists of predicting a subset of the kinetic parameters used in the model to represent fundamental biological processes. Many parameter estimation processes done in the modeling community field start from already existing experiments, from which researchers fit the parameters of their models. The present challenge will be based on a parameter inference process comprising an experimental design component that forms an integral part of the optimization procedure. The challenge will be structured around a credit budget system in which participants can purchase in-silico generated data of their choice (within a restricted set of possibilities) and use it to infer the model parameters. This selection can be done iteratively until the budget is exhausted, each round choosing the experiment most needed according to the current model constrains. This iterative data-acquisition setup proved very efficient for modelers to accurately find the actual parameters underlying the biological networks in the two previous DREAM parameter estimation challenges. The present challenge, however, presents modelers with a substantial leap in computational demands as hundreds of parameter values will have to be predicted from a model that is orders of magnitude more complex than in the previous DREAM runs. As the model requires extensive computational power, we expect to provide participants with a platform to run the whole-cell model and test their parameter optimization methods. When applied to real experimental systems, the results of this challenge will show how well we can reverse engineer the kinetics of cellular processes. In the longer run this understanding will allow us to model cellular behavior, predict a cell’s response to external agents such as therapeutic or toxic compounds and allow for drug target identification among many potential applications.
As in the other challenges, the best performing team will be invited to present at the DREAM conference with travel expenses covered. Sage/DREAM organizers are also working with high impact journals to ensure that the best performer methodology could be considered for publication under the challenge-assisted peer review format.