AutoGaitA – Automated Gait Analysis in Python
1 Forschungszentrum Jülich, Institute of Neuroscience and Medicine (INM-3)
2 University of Cologne, Department of Neurology, Medical Faculty and University Hospital Cologne
3 University of Cologne, Institute of Vegetative Physiology, Medical Faculty and University Hospital Cologne
4 University of Cologne, Institute of Zoology
5 Department of Neurology, University Hospital Frankfurt
Machine learning (ML), and particularly the field of body posture tracking, has significantly altered the study of locomotion in recent years. Software tools such as DeepLabCut (Mathis et al., 2018) and Simi Shape (Winiarski, 2003) recognise an organism’s body and limbs in a video frame and output joint coordinates in space over time, making previously used joint-markers obsolete. Despite ML algorithms having alleviated the tracking effort, making sense of the tracked coordinates still requires substantial amounts of manual labour and lacks standardisation across research labs.
To this end, we developed AutoGaitA (Automated Gait Analysis), an open-source Python toolbox designed to automate the analysis of any motor behaviour of interest performed by any species of interest. AutoGaitA’sapplicability to any kind of motor behaviour and to any species of interest makes it a valuable tool for the motor community to standardise the analysis of rhythmic behaviours across genotypes, disease states and species.
We demonstrate AutoGaitA’s capabilities with a series of proof-of-principle experiments in which the gait of flies, mice, and humans was analysed. AutoGaitA’s locomotion analysis include normalising and averaging step cycles, before extracting meaningful features from the tracked coordinates (e.g. angles, velocity, acceleration), allowing intra- and inter-animal comparison. AutoGaitA provides (cluster-extent permutation and ANOVA-based) statistical tests, principal component analyses (PCA), providing exact numerical PCA results (e.g., eigenvectors) as well as 2D and 3D scatterplots, generating rotating videos of the latter, and various additional figures and excel files storing the (run- or group-level) results.