Poster Presentation

Poster F in Poster Session F: Thursday, August 6, 1:45 – 3:30 pm, Kimmel Center, Shorin & Rosenthal Rooms

Peptide-mediated timescale separation underlies Bayesian causal inference during multisensory integration

Madeleine C Snyder¹; ¹Harvard University

Presenter: Madeleine C Snyder

Multisensory integration is critical for a diversity of behaviors in ancestral nervous systems , but the capacity of the first "true" brains to perform complex computations such as inference has not been fully explored. Here we argue that one of the first bilatarian brains may perform approximate causal inference to determine whether separate sensory channels reflect a common input source . Our model of Bayesian causal inference uses timescale-separated computation to allow dynamic weighting of sensory inputs, which drives adaptive phototaxis behavior. We propose that neuropeptides serve as a putative biological substrate for this timescale separation due to their robust and global modulatory effect on small, basal brains . To test our model, we used a dataset of phototaxis behavior of planarian flatworms (Schmidtea mediterranea) exposed to coherent and conflicting ocular (520 nm) and extraocular (365 nm) stimuli . Planarians exhibit a characteristic delayed onset of stimulus tracking under UV-only conditions, immediate tracking when UV and visible light are presented coherently, and graded responses under conflict conditions. RNAi knockdown of two neuropeptides, ppp-1 and spp-1, eliminated the delay and produced immediate tracking, consistent with a disruption of slow evidence accumulation. Our model, which compares an integrated single-source model against a segregated two-source model via log-evidence accumulation, qualitatively reproduces stimulus tracking and ignoring behavioral trends across six lighting conditions, as well as the peptide knockdown phenotype. These results support the hypothesis that temporally separated computation enables planarians to perform approximate Bayesian causal inference and allocate sensory bandwidth to the most informative channel given accumulated evidence about source coherence.

Topic Area: Methods, Tools, Theory & Neural Coding