F.A.Q

Here I provide a growing list of frequently asked questions and helpful tips for running ABC-SysBio and performing analyses using ABC.

INSTALLATION

  1. How do I install ABC-SysBio?
    Installation instructions on Linux and Mac are provided in the README.txt file included in the package distribution.
  2. What about installation on Mac OSX 10.6 Snow Leopard?
    ABC-SysBio installs on Snow Leopard but we have had only partial success in installing libSBML and matplotlib. Conflicts have been caused due to the shift to 64bit libraries and it will take some time for the dependent packages to catch up. If people have had success then contact me and I will update this page accordingly.
  3. What about Windows?
    Unfortunately I don’t have time to support ABC-SysBio on Windows. We have succeeded in installing ABC-SysBio and dependencies on Vista in the past. If anyone has instructions for installing and running on any Windows platform then get in touch and I can update this page.

INFERENCE

  1. How do I choose the epsilon schedule?
    This can be chosen via trial and error keeping the acceptance rates at reasonable values (> 0.05 for example). There is now also the option for automation calculation of the epsilon schedule. This is done by examining the previous population distance distribution and choosing a quantile of this as the next threshold (this is the alpha parameter). This is experimental and you must try different alpha values as the performance seems to be problem dependent and it is also possible to choose alpha such that the algorithm gets stuck in a local minimum.
  2. How do I choose the kernels?
    At first users are strongly advised to use uniform kernels in the input file (default). This should insure that the posterior is covered adequately but it is no guarantee and different kernels should then be investigated to make sure they have no effect on the conclusions of the inference.
  3. How can I infer initial conditions?
    This is now possible in version 2.x. Initial conditions are treated as parameters that must have priors placed on them.