Refining methodological approaches for analysing higher-throughput locomotor behaviour, using generalized additive...

Movement underlies most essential behaviours and is, therefore, frequently used as a behavioural endpoint in a wide range of phenotypic assessments. These include higher-throughput locomotor assays typically employing larval fish or occasionally invertebrate species, which are widely applied to assess the effects of chemical exposure or genetic modification on behaviour. Analysis of the resultant locomotor data, however, is challenging due to inter-individual variability, repeated measures (for each individual), and complex time-dependent responses. These challenges are further exacerbated when additional independent variables, such as light-dark cycling in the visual motor response (VMR) assay, are incorporated.

Here we compare use of generalised additive mixed models (GAMMs) with conventional analysis of variance (ANOVA)-based approaches to assess chemical effects on locomotor behaviour, using a case study of the impact of clozapine exposure on (light/dark) in zebrafish embryo larvae.

We show adoption of generalised additive mixed models (GAMMs) provides the most appropriate framework for analysing locomotor data, owing to their flexibility in modelling non-linear behavioural patterns, and their ability to explicitly account for dynamic inter-individual variability. While ANOVA identified a significant three-way interaction that precluded meaningful post hoc interpretation, its reliance on assumptions that are violated by these data limits the validity of the resulting inferences. In contrast, GAMMs accommodate the data structure and explain a greater proportion of the variance (59.7% compared to 54.1% for ANOVA), here revealing significant, light-dependent non-linear locomotor trajectories. GAMMs further identified treatment-specific effects, including reduced locomotion under light conditions across all clozapine treatments, and at the highest concentration only in dark conditions. Overall, our findings demonstrate that GAMMs offer high statistical rigour but with improved interpretability of treatment effects, supporting their wider adoption in analysing data from studies employing similar behavioural assessment paradigms.