What is Computational Oncology?
Introduction and Long-Term Vision
Computational Oncology is currently in its very infancy. In fact, as of this writing in December 2005, a Google search for "computational oncology" only yields about 112 hits, many of which are related to a German bioinformatics group. (Update: As of December 2006, that number has increased to 861 hits--this really illustrates the explosive growth of this new field!) Because the mathematical modeling and computer simulation of cancer are still relatively young, computational oncology has currently had its best and earliest success in the application of machine learning techniques and statistical analysis of the growing volume of genetic and proteomic data on cancer, i.e., bioinformatics. This is what computational oncology currently is.
Much more exciting, however, is what computational oncology will be. Mathematicians, oncologists, biologists, biomedical engineers, computer scientists, and others are collaborating to devise mathematical models of intracellular processes, tumor growth, the mechanics of tissue response, and other vital aspects of cancer. In the meantime, medical imaging is improving, genetic and proteomic analysis are faster, and computers are becoming faster and cheaper.
In the future, these trends will merge to push computational oncology forward as a meaningful, quantitative field. In what many consider the holy grail of the field, the clinical oncologist of the future would input a patient's medical imagery, tumor biopsies, bioarray data, and sequenced genome into a sophisticated simulator to predict the future progression of a tumor, assess its invasiveness, and choose an optimal, patient-tailored treatment.
Concrete Steps Towards the Vision
Researchers are already taking concrete steps towards the long-term vision of a patient-tailored cancer simulator. Even today, tumor models are being used to better understand the effects of nutrient transport on tumor invasiveness, examine the interplay between oncotic pressure and angiogenesis as they relate to intravenous drug delivery, and the effects of the tumor microenvironment on the morphology of a tumor.
The next logical step is to address one of the most pressing problems in cancer research: the inadequacy of the mouse model. A common expression among oncologists is, "I can cure cancer in any mouse. But in human beings, it's another story." While the mouse model has provided a wealth of insight on the nature of cancer, it has proven to be a poor predictor of treatment performance in human beings. With better models of complex tissue structures, the immune system, nutrient transport, and the interaction of tumors with their microenvironment, a well-callibrated cancer simulator could supplant the mouse model as a superior predictor of human tumor response to treatment.
Such a development would be logical. Physical experiments are much more expensive and take longer than computer simulations. A sophisticated cancer simulator could help pharmaceutical companies to examine many ideas simultaneously, allowing them to separate the wheat from the chaff, reduce the uncertainty and risk of research, and speed the time to market for new treatments, all at lower cost. At this stage, a virtual cancer model would be well-suited to the behavior of human cancers in generalized patients. However, the use of a cancer simulator in pharmaceutical research would provide vital feedback and data to improve the quality of the underlying mathematical models and increase the clinical relevance of any ongoing virtual patient project.
Some Necessary Advances
Some advances will be necessary before these goals can come to fruition. Current tumor models do not model the interaction between the tumor, its surrounding microenvironment and its host organ, nor the interplay between the tumor and more distant organs. They also do not model the spatial distribution and evolution of genetic variations within a tumor. They do not model the process of metastasis. All these problems must be overcome, and they likely will be in a gradual, step-by-step process.
As models advance toward clinical relevance, interfaces between medical imagery and the computer simulations must be developed. The interface must be able to extract tissue structure and mechanical characteristics, determine the location of the tumor, and match internal parameters to available patient data. Some techniques, such as level set-based image segmentation, already exist to do this, but it will require careful integration and extension.
Throughout the process, greater collaboration between experimenters and modelers will be required to validate, critique, and extend the mathematical models. This will provide the additional benefit of providing expertise in callibrating a model's parameters to experimental and patient data.
