Global and local approaches to detect subtle behavioural changes in large-scale behavioural screens

The central nervous system produces diverse behaviours, like muscular responses, observable via video recordings. Current advancements in genetics, large-scale behaviour tracking, and machine learning facilitate the understanding of how behaviour and neural activities correlate. In organisms like the Drosophila larva, it's now feasible to map this at large scales, covering millions of animals and individual neurons. This enables the pinpointing of neural circuits linked to specific behaviours. High-throughput behavioural screens are invaluable as they relate the activation or deactivation of neurons to behaviour sequences in millions of organisms, showcasing the vast range of neural responses to a single stimulus. However, extracting nuanced behaviours from these screens and interpreting them at a broader scale remain challenges.

We propose multiple robust methods to mine detailed information from behavioural screens:

1. - A generative method to regularize inference of larva shape across the entire screen.
   
   - An unsupervised kernel-based statistical test to spot slight behavioural deviations.
   
   - A generative model to identify significant changes in behavioural sequences.
   
   - A suffix-tree-based method to group genetic lines based on frequent action sequences. We've successfully applied these methods to a screen analyzing the behaviours of 292,639 larvae across 593 genetic lines in response to an air puff.