Anxiety in aquatics: Leveraging machine learning models to predict adult zebrafish behavior

Accurate analysis of anxiety behaviors in animal models is pivotal for advancing neuroscience research and drug discovery. This study compares the potential of DeepLabCut, ZebraLab, and machine learning models to analyze anxiety-related behaviors in adult zebrafish. Using a dataset comprising video recordings of unstressed and pre-stressed zebrafish, we extracted features such as total inactivity duration/immobility, time spent at the bottom, time spent at the top and turn angles (large and small). We observed that the data obtained using DeepLabCut and ZebraLab were highly correlated. Using this data, we annotated behaviors as anxious and not anxious and trained several machine learning models, including Logistic Regression, Decision Tree, K-Nearest Neighbours (KNN), Random Forests, Naive Bayes Classifiers, and Support Vector Machines (SVMs). The effectiveness of these machine learning models was validated and tested on independent datasets. We found that some machine learning models, such as Decision Tree and Random Forests, performed excellently to differentiate between anxious and non-anxious behavior, even in the control group, where the differences between subjects were more subtle. Our findings show that upcoming technologies, such as machine learning models, are able to effectively and accurately analyze anxiety behaviors in zebrafish and provide a cost-effective method to analyze animal behavior.