Dynamical Interrogation of Serpentine Medulla Circuits in the Drosophila Optic Lobe

Abstract

Dense connectomes invite a tempting leap: if we can "wire up" a circuit, perhaps its dynamics (and even its function) will emerge. In practice, connectomic constraints leave large biophysical degrees of freedom (synapse signs/strengths, neuromodulation, gap junctions), so qualitative dynamical claims can be fragile. We present a simple forensic workflow for testing connectome-to-dynamics hypotheses that is deliberately hostile to wishful thinking: each stage is gated by sanity checks, run across multiple random seeds, and reported with effect sizes and negative controls. As a case study, we instantiate the Serpentine Medulla (Sm) interneuron family from FlyWire as a current-based adaptive exponential integrate-and-fire (AdEx) network and test a sharp narrative: whether strong recurrence supports a working-memory-style bistable attractor. Under strict attractor criteria (noise removed after the kick), we find no robust evidence for bistability (paired Wilcoxon p = 0.099, within-subject dz = - 0.25). For oscillations, a corrected reproduction across 20 random seeds shows that recurrent coupling strongly amplifies integrated 30-80 Hz power when comparing the fully coupled condition (Full W) to an uncoupled control (W = 0) (paired Wilcoxon p = 1.9 x 10-6; dz = 5.96). We also observe a fast negative-coupling signature in input decomposition (mean peak correlation r {approx} - 0.40 at [~] 4 ms lag), while stressing that these signatures are conditional on the chosen model class and operating point. Overall, the main contribution is methodological: an auditable template for falsifying attractive connectome-based stories rather than polishing them.