Abstract In this project it was aimed to determine if (i) the selective a5 GABAA receptor inverse modulator a5IA-II would improve memory in the rat; (ii) the selective partial a2 and a3 (and full a5 agonist) GABAA receptor modulator NS.A would be effective in aputative animal model of schizophrenia; and, finally (iii) to develop a statistical model appropriately describing the pre-clinical data and enable a suitable test of drug effects. The effects of a5IA-II and NS.A in the models utilised were compared in all cases to the non-selective GABAA receptor modulator alprazolam, a benzodiazepine. All three GABAA receptor modulators were assessed in male SPRD rats tested in the following models: fear conditioning, trace fear conditioning and pre-pulse inhibition. In the latter model, pre-pulse inhibition, the three GABAA receptor modulators were tested for their ability to reverse PCP or amphetamine induced deficits which arguably reflect sensorimotor gating deficits seen in schizophrenia. Whilst initial studies indicated that a5IA-II tended to improve fear conditioning notably during and after the tone period on the test day, this was not reproducible. In the trace fear conditioning experiment, the inclusion of a trace between the offset of the tone and onset of the shock was anticipated to retard memory relative to a normal fear conditioning group (i.e., offset of tone and onset of shock co-terminate). However, animals trained with and without a trace showed equivalent memory 24 hours after training. Therefore, any interpretation of the data for a5IA-II or other GABAA receptor modulators is equivocal. In pre-pulse inhibition experiments, NS.A did not affect PCP induced pre-pulse inhibition impairment and surprisingly exacerbated the amphetamine induced pre-pulse inhibition impairment. From a statistical approach it was found that a bias as well as variation is induced by the experimental equipment which finds expression in heterogeneity across the startle boxes. In order to assess the effect of the drugs in fear conditioning experiment a linear mixed effects model was defined. In order not to violate the model assumptions the heteroscedasticity obtained for the startle boxes was modelled by means of a variance function with the assignment of a variance parameter for each box. An exponential correlation structure was used to model residual autocorrelation. Type Master's thesis [Industrial collaboration] Year 2007 Publisher Informatics and Mathematical Modelling, Technical University of Denmark, DTU Address Richard Petersens Plads, Building 321, DK-2800 Kgs. Lyngby Series IMM-Thesis-2007-08 Note Supervised by Associate Professor Klaus Kaae Andersen (IMM, DTU), and Naheed Mirza, Karin Sandager Nielsen (NeuroSearch). Electronic version(s) [pdf] BibTeX data [bibtex] IMM Group(s) Mathematical Statistics