Close

A computation opponent process model describes and predicts cocaine self-administration among naïve rats

Abstract

Aims: To develop, calibrate and validate a computational opponent process model to cocaine self-administration data in rats. Methods:Outcomes. Numbers of injections per session and time between injections.Data.We used data from two 30-day rat cocaine self administration experiments (Mihindouet al., 2011, 2013).Daily sessions lasted for 6h in one and 3h in another experiment.Model. We used a control theory computational model (4 difference equations) that simulates self-administration controlled by an adaptive self-stimulating threshold (Newlin et al., 2012). The threshold is modeled as a function of both: the drug effect and the delayed opponent process (allostatic adjustment). Calibration and analysis. We calibrated the model to a time series based on 6-h sessions.We used a non-linear fit algorithm and mean square error (between the data and the model-generated trajectories) as the goodness of fit criterion. We used this calibrated model to predict the numbers of injections in the 3-h session time series. We examined variance of times between the injections among 3-h and 6-h sessions. Results:The model calibrated to the 6-h session almost perfectly predicted long-term (after 5 days) numbers of injections in 3-h sessions. In the first 5 days, experimental data showed qualitative differences in stabilization of the variances of times between injections between the 3-h the 6-h sessions. The patterns of increase in daily numbers of injections in the first 5 days were also qualitatively different (concave vs. convex). This difference could be potentially attributed to the longer time the rats needed to reach stable allostatic process (learn the effect of the drug). Model improvement is proposed to capture these differences. Conclusions: A computational model of self-administration well describes and predicts long-term cocaine self-administration of initially naive rats. Accounting for the initial learning of drug effects (first 1–7 days) is the next step in model improvement.

MIDAS Network Members

Citation: