Detrended Fluctuation Analysis in behavioural studies: the effects of stress induced inactivity on the fractal organization of behavioural time series

KEMBRO, J. M.; PERILLO, M. A.; MARÍN, R. H.

La Falda, Cordoba

Workshop; Segunda escuela Argentina de Matematica y Biologia; 2007

FAMAF,UNC

Fractal analyses has been used in recent animal behavioural studies to provide information about the organization and complexity of temporal patterns of behaviour, and how these patterns can be modified by stressor, drugs, etc. One of the fractal analyses most commonly used in the area is detrended fluctuation analysis (DFA).  Even though DFA is a simple procedure that can easily be performed by biologists, some considerations must be taken into account in the interpretation and calculation of results. Often some aspects are either not taken into account or not explicitly referred to in works that use DFA to study animal behaviour. For example, 1) time series must include at least 4000 data points, 2) some periodic oscillations are not eliminated by the detrending process, 3) trends in data time series of biological systems are frequent, and 4) long periods of inactivity may affect the fractal organization of the time series. This last point is particularly interesting because stressors often induce a period of inactivity at the beginning of the trial, called latency to ambulate. We used DFA to study the autocorrelation properties of the walking activity of female Japanese quails (N=22) tested in a novel environment (stressor).  In particular, we evaluated the effect of an initial period of inactivity on the fractal organization of behavioural time series. At 0.5 s intervals during 35 min we recorded the mutually exclusive states of walking and non-walking (walking time series), and the distanced walked during each interval (distance time series). No periodic oscillations were found. DFA of order 2 was used since it eliminated trends found in data. DFA showed that birds with large latencies to ambulate (>550s) never showed monofractal walking or distance time series. When the latency to ambulate data period was not included in the DFA calculation, all but one of the previous non-monofractal time series became monofractal. These results suggest that these long periods of inactivity, that are typically found in studies that apply stressors or use drugs that induce fear, can cause difficulties and possible errors in the interpretation of results obtained with DFA. Also, the fact that trends were found of order equal or less than 2 highlights the important of systematically studying trends in data in order to use an appropriate detrending order.