Latent state dynamics in female Drosophila during social interactions

Umesh Singla, Albert Lin, Jonathan W. Pillow, Mala Murthy

Keywords: social behavior, multisensory, statistical modeling, behavioral quantification, Drosophila melanogaster

TL;DR: A GLM–HMM uncovers latent states that shape how female Drosophila respond to male courtship cues.

Abstract: Social interactions across animal species are governed by the interplay between multimodal sensory information and an animal's internal state. Here, we investigate this interplay in female Drosophila as she engages with the male during courtship, a highly dynamic social behavior. While male behaviors during courtship, such as song production, have been well characterized moment by moment, the female’s actions have not been described with similar temporal precision, despite her central role in determining copulation outcomes. Her behavior displays high variability and is often viewed as volitional, raising questions about its structure and predictability at fine temporal resolution. To address this, we used a state-space model that combines Generalized Linear Models (GLMs) with a Hidden Markov Model (HMM) to uncover latent states that modulate the relationship between male sensory cues and female responses. We find that, overall male cues weakly predict female behavior, but that predictive power varies substantially across inferred states: some states exhibit clear cue-driven structure, while others show reduced sensitivity to external cues and more internally-driven behavior. At short timescales, female behavior appears only weakly predictable and highly variable, yet, at longer timescales, a rich latent state structure emerges, hinting at internal gating and evaluation of social signals over time. This work provides the first moment-by-moment characterization of female behavior during courtship, taking a crucial step toward closing the loop in social behavior modeling.

Submission Number: 40