Chalmers Conferences, 9th European Conference on Mathematical and Theoretical Biology

Simultaneous modeling and clustering of multivariate time series - application to an animal model of alcohol addiction
Lilian Villarín Pildaín, Vineet Kumar, Franziska Matthäus

Last modified: 2014-04-01


More and more biological approaches yield quantitative data in form of multivariate time series, such as time-resolved microarray data, EEG, or the simultaneous recording of multiple components of a signaling pathway. We present a method to simultaneously characterize and classify dynamic multivariate patterns. For this, we use a generalized linear modeling (GLM) approach extended to multivariate data. Mixtures of multivariate GLMs are fitted to the data using an Expectation-Maximization algorithm.

Our method is applied to high-resolution time series data describing animal drinking behavior in an alcohol addiction protocol, where rats have free access to water and three different alcoholic solutions. Our results are not only consistent with other state-of-the-art methods, but also provide a detailed dynamic description of the pattern behavior at each stage. The dynamics contain high frequency and circadian rhythm components, whose analysis give insight into processes related to intake and metabolism of ethanol. We observe an evolution of drinking behavior over the repeated cycles of alcohol admission and deprivation, with a clear initial and specific advanced preference profiles. We further develop a new criterion to identify the alcohol deprivation effect, which quantifies relapse and presents one of the features of alcohol addiction.