Two existing models of brain dynamics in epilepsy, one detailed (i.e., realistic) and one abstract (i.e., simplified) are compared in terms of behavioral range and match to in vitro mouse recordings. A new method is introduced for comparing across computational models that may have very different forms. First, high-level metrics were extracted from model and in vitro output time series. A principal components analysis was then performed over these metrics to obtain a reduced set of derived features. These features define a low-dimensional behavior space in which quantitative measures of behavioral range and degree of match to real data can be obtained. The detailed and abstract models and the mouse recordings overlapped considerably in behavior space. Both the range of behaviors and similarity to mouse data were similar between the detailed and abstract models. When no high-level metrics were used and principal components analysis was computed over raw time series, the models overlapped minimally with the mouse recordings. The method introduced here is suitable for comparing across different kinds of model data and across real brain recordings. It appears that, despite differences in form and computational expense, detailed and abstract models do not necessarily differ in their behaviors.
From the University of Memphis (A.S.W.), University of Chicago (H.C.L., M.B., R.L.S., M.H.), and the Argonne National Laboratory (R.L.S., M.H.).
Supported by a Department of Energy Computational Science Graduate Fellowship (to A.S.W.), the Falk Foundation (to H.C.L.), and by the Office of Advanced Scientific Computing Research, Office of Science, U.S. Department of Energy, under Contract DE-AC02-06CH11357.
Presented as a poster at Tools for Epilepsy Research: Tutorials and Updates, Chicago, IL, August 6–8, 2009.
Address correspondence and reprint requests to Anne S. Warlaumont, School of Audiology and Speech-Language Pathology, The University of Memphis, 807 Jefferson Ave., Memphis, TN 38105, U.S.A.; e-mail: email@example.com.