SAMBA: A Segment Anything Model-based tool for semi-automated Behavioural Analysis of Drosophila and other model organisms

Quantitative behavioural analysis is a powerful approach for linking genotype to phenotype, but many existing tools require specialised hardware, extensive preprocessing, or coding expertise. We present SAMBA (Segment Anything Model for Behavioural Analysis), an open-access, Google Colab-based pipeline that harnesses the Segment Anything Model 2 (SAM2) for accurate, semi-automated tracking without thresholding or background subtraction. With minimal user input, SAMBA extracts movement parameters, detects behavioural states, and supports batch processing. Validating SAMBA in a Drosophila melanogaster model of 3-hydroxyisobutyryl-CoA hydrolase (HIBCH) deficiency revealed impaired locomotion, reduced speed, and altered decision-making, highlighting its ability to capture nuanced phenotypes in neurometabolic disease. We further demonstrate adaptability to adult Drosophila and larval zebrafish, underscoring its cross-species utility. By combining foundation-model segmentation with an accessible interface, SAMBA lowers technical barriers to high-throughput behavioural phenotyping and is readily extendable to diverse model organisms, life stages, and experimental paradigms. This flexibility positions SAMBA as a valuable platform for accelerating disease mechanism studies, genetic screens, and preclinical testing.