Talk at the NCI 4th Annual PS-OCs Network Investigators’ Meeting on April 19, 2013


Exploring possibilities for next-generation computational cancer models that work

together (a conversation starter)



Cancer heterogeneity—ranging from genetics to phenotype to prognosis—has emerged as a critical cancer research topic. Computational modelers are grappling with a different sort of heterogeneity: a large and growing variety of modeling techniques, generally applied to different cancer problems, with little interoperability. This complicates model comparison and cross-validation and hampers efforts to interconnect models and data from different groups. This stands in contrast with other fields faced with complex data and dynamics. For example, hurricane forecasting has advanced tremendously over the last 40 years, not only due to improved data collection and individual model refinement, but also because forecasters combine them into ensemble models that leverage the variety of model assumptions to mitigate the impact of uncertain weather physics and make better predictions.

In this talk, we will explore possibilities for a next-generation ecosystem of compatible models that can cooperate to combat uncertainty in patient parameters and the biophysics of cancer. There may never be a single best model that correctly incorporates all the important biology of cancer, but ensembles of high-quality models could leverage diverse expertise to improve predictions. And interconnected systems of compatible models may be able to integrate diverse data sets to help explain patient outcomes, assess and refine leading cancer biology hypotheses, and ultimately guide clinicians and their patients in their treatment choices. We will discuss key challenges and a possible way forward to advance this vision, explore potential dividends for interdisciplinary cancer research, education and open science, and invite others to join us in exploring these ideas.