RE: Bayesian networks

Svetlana, If you could explain a little more on what type of data you are trying to model using Bayesian Networks I may be able to help you. If you simply need to specify uncertainty in your parameters you can do so in NONMEM 7 without using a Network approach. Jeff and I have previously used Bayesian Networks for disease progression modeling, however this was not linked to NONMEM. Our preliminary work was presented at the PAGE meeting in 2009 ( http://www.page-meeting.org/?abstract=1649) and we have further developed this to add more dynamic component. Our approach is using Netica software to implement the SAEM algorithm in order to 'learn' conditional probabilities directly from individual patient records. This approach was particularly helpful (and fast!) for modeling sparse multivariate data. We have piloted models which may include a drug effect in to these as well and find they are very flexible when integrating raw data with historical/literature data. As for linking to NONMEM models, the use of WinBUGS for PKPD modeling is the most direct approach to this as far as I know (see http://www.ncbi.nlm.nih.gov/pubmed/12449499). The common theme is these models with what we have done for disease progression modeling in Netica is the use of a Directed Acyclic Graph (DAG) structure, however the parameter estimation procedure may differ. There are many NMusers more adept at WinBUGS than I so they may be able to help with your problem if more suited for using this approach. Here is my favorite 'primer' on BNs: Needham CJ, Bradford JR, Bulpitt AJ, Westhead DR. PLoS Comput Biol. 2007 Aug;3(8):e129. http://www.ncbi.nlm.nih.gov/pubmed/17784779 Also see the Book : Probabilistic Networks and Expert Systems: Exact Computational Methods for Bayesian Networks By Robert G. Cowell, Philip Dawid, Steffen L. Lauritzen, David J. Spiegelhalter If you'd like to discuss further please contact me directly. Jason Jason H. Williams, PhD Manager, Clinical Pharmacology Pfizer Biotechnology Unit & Oncology Clinical Research 500 Arcola Road A5009 Collegeville, PA 19426 email: [email protected] office phone: 484.865.2495