Dementia is characterised by a profound decline in cognitive and behavioural functions, drastically affecting daily life and social interactions. There is a pressing need for novel therapeutic strategies, particularly if these can be repurposed from available medicine – recent research has suggested antidiabetic drugs may have neuroprotective properties, but validation requires translatable pre-clinical models paired with unbiased assessment. Rocky Mountain Laboratory (RML) scrapie is a well-characterised model that mimics many of the hallmarks of neurodegeneration seen in dementia but does not require familial transgenes and has distinct motor endpoints. It has potential for rapid in vivo testing of neuroprotective agents, but unbiased phenotyping of motor signs has historically been challenging. This study validates the use of artificial neural network based pose estimation in this model for unbiased drug validation. We first used spatial-coordinate time series data of body and limb position in mice, combined with feature engineering and elimination and linear modelling, to identify multivariate motor signatures of disease. Preliminary t-SNE analysis distinguished RML and clusters by eight features. We also identified significant differences in RML and control for specific discriminating motor features, including speed variability and activity time (p<0.05), indicating the assay sensitivity. We aim to extend this paradigm to assess disease severity longitudinally across multiple assays, identify disease progression characteristics between RML and control mice, and maximise assay sensitivity for drug testing. Finally, we show proof of principle data for the neuroprotective potential of metformin and related compounds. Further work is now progressing on their mechanism.