How sparse can a fly-brain model be? Connection strength, wiring placement, network state and task jointly determine function-preserving compression of the Drosophila connectome

Abstract

Whole-brain connectomes are increasingly used as wiring diagrams for spiking simulations and as structural priors for artificial networks, usually after an ad hoc synapse-count threshold whose effect on validated function has never been measured. We define function-preserving sparsification—removing edges under a fixed budget while preserving validated behaviour and population responses—and study it in the leaky integrate-and-fire whole-brain model of Shiu et al. (2024). Six operators—weak-synapse pruning, uniform and weight-proportional random removal, degree-preserving rewiring, weight shuffling and a task-aware criterion—were compared at identical edge budgets across three validated circuits (sugar-evoked proboscis extension, its bitter suppression, grooming pathway specificity), silent and spontaneously active network states, and multiple seeds. Weak-synapse pruning preserves 79% of feeding output with 6% of edges, with a cliff at k* ≈ 9 synapses, whereas random, degree-preserving and weight-shuffled controls at the same budget abolish output, and weight-proportional random sampling survives only while it retains the strong-edge set: function resides in the conjunction of wiring placement and strength, not in degree statistics or synaptic mass. The cliff is task-dependent—grooming specificity collapses below k = 2 through disinhibition. In the active state failure changes character: output over-excites instead of vanishing, and bitter suppression is lost progressively from k = 5 and reverses at k = 31. A task-aware criterion preserves its training stimulus at 0.4% of edges but loses held-out inhibition first. These results provide a first benchmark for connectome sparsification and show that what must be kept depends on task and network state. Author summary Neuroscientists now have a map of every neuron and connection in the fruit-fly brain. Computer scientists have begun to use that map directly—as a circuit to simulate, or as a fixed "wiring diagram" inside artificial networks that steer robots or move simulated bodies. But the map is large—15 million connections—and almost every study quietly discards part of it before use, typically by deleting connections with only a few synapses. Nobody had checked what that deletion does to the behaviours the model is supposed to reproduce. We asked how much of the fly connectome can be removed from an experimentally validated brain simulation before it stops working, and what decides which connections matter. The answer was not simply "keep the strong ones": choosing the same number of connections at random, rewiring them while preserving each neuron’s connection count, or keeping their positions but shuffling their strengths all destroy behaviour. Which circuits survive also differed—a grooming circuit that must distinguish two sensory pathways collapsed much earlier than a simple feeding circuit, and when the brain was spontaneously active, inhibition was the first thing to go. The results give principled grounds for the thresholds connectome-based models use and provide a test bed for future compression methods.