| Register
\(\newcommand{\Cat}{{\rm Cat}} \) \(\newcommand{\A}{\mathcal A} \) \(\newcommand{\freestar}{ \framebox[7pt]{$\star$} }\)

3. Other modeling-centric problem areas

    1. Problem 3.1.

      Accounting for and modeling metastasis
          Main questions:

      How do we consider modeling the metastatic tumor cells as opposed to the primary tumor cells, and/or the transition from primary to metastatic cells (known as the epithelial-mesenchymal transition (EMT))?


      (a) Issue of understanding what about the primary tumor will cause it to spread, versus studying already-spread metastatic tumor cells.

      (b) Making sure we are testing and developing models for the relevant state, if we give drugs to patients who already have metastatic cancer.

      (c) Clinical trials likely occur in metastatic patients, but if that drug is successful, it will likely be used on non-metastatic patients. Does that affect drug effectiveness?

      (d) For an example of a systems approach to studying the epithelial-mesenchymal transition, see Kim et al. (2011).
        • Problem 3.2.

          Uncertainty in model structure
              What kind of methods/experiments are available, or should be available, to efficiently test model structures?


          (a) For examples of methods to explore model structures, see Ciaccio et al. (2010) and Morris et al. (2011).
            • Problem 3.3.

              Uncertainty in model parameters
                  Main questions:

              Can we create a ’best practices’ approach for estimating model parameters? Can we better define local vs. global parameter fits?


              (a) What do people mean by ‘global’ (e.g., global sensitivity analysis), given that there are not well established/widely accepted methods for this purpose.

              (b) If we are doing parameter estimation, how do we efficiently crawl through the search space, assuming the problem is not convex.

              (c) This topic was partially discussed in a break out group. They thought that a balance must be established between simple and understandable methods, and methods that are easily parallelizable. For example, Kalman filtering is not parallelizable, but particle filtering is parallelizable, even though it is more complicated to do Particle filtering. Thus it may be useful to look for tools that are inherently parallelizable.

              (d) There are methods that can be useful for moderately sized problems, but fail for higher dimensional problems.

              (e) Must work with experimentalists to get a solid understanding of what are valid and reasonable estimates of parameter values a priori.

                  Cite this as: AimPL: Systems approaches to drug discovery and development in oncology, available at http://aimpl.org/systemsoncology.