Structure alone supports efficient visual computation in the Drosophila visual system

Eudald Correig-Fraga; Roger Guimerà; Marta Sales-Pardo

q-bio.NCOct 7, 2026

Structure alone supports efficient visual computation in the Drosophila visual system

Authors: Eudald Correig-Fraga, Roger Guimerà, Marta Sales-Pardo

Organizations: Innovamat Education, Sant Cugat del Vallès, Catalonia · Department of Chemical Engineering, Universitat Rovira i Virgili, Tarragona, Catalonia · Center for Computational Science and Applied Mathematics (ComSCIAM), Universitat Rovira i Virgili, Tarragona, Catalonia · ICREA, Barcelona, Catalonia

Abstract

Understanding the extent to which measured synaptic wiring determines computation remains a central challenge. Here, we couple the proofread adult Drosophila melanogaster connectome to an anatomically faithful model of its eye. Visual information is inputted in the eye model, then passed to the connectome, and finally read from a Kenyon-cell-centered linear decoder. This creates a connectome-only model in which the anatomical graph and eye geometry are fixed and only scalar synaptic gains and neuronal thresholds may be learned. The model supports multitask vision, including color discrimination, shape classification, and numerical discrimination that follows a ratio-dependent scaling characteristic of approximate number perception. To test whether precise connectivity is consequential under wiring economy, we compare the biological graph to randomized ensembles that increasingly preserve biological synaptic constraints. At matched wiring cost, the biological network consistently yields higher accuracy, whereas less constrained rewiring surpasses it at the cost of inflated wiring. These findings indicate that the measured connectivity and eye geometry jointly set efficient operating points for visual computation.

Figures & tables

Figure 1: Computational modeling of numerical discrimination in Drosophila melanogaster. a, Visual stimuli with varying numerosities of colored dots (blue vs. yellow) are processed through an anatomically accurate model of the fly’s compound eye. We reconstruct the 3D positions of retinal neurons from the connectome data, project them onto a 2D plane, and use Voronoi tessellation to map ommatidia, with R7 neuron positions as cell centers. Each ommatidium samples its corresponding region of the visual field, with specific photoreceptors (R1-R8) encoding different color channels based on their biological spectral sensitivities. The captured light intensity in each retinal neuron produces activation patterns that form the initial input to the neural network. These retinal neuron activations propagate via a message-passing algorithm on the graph created by the Drosophila connectome (54.5 million machine-generated synaptic contacts). Information flows via message passing for 2-6 propagation steps [ 1 ] , mimicking how neural signals travel from the retina through the optic lobe and ultimately to the Kenyon cells in the mushroom body, which serve as the decision-making locus. b, Example of an input image. c, Distribution of photoreceptor terminals colored by their spectral sensitivity (R1-6: white, broad spectrum; R7: blue, UV-sensitive; R8p: green, green-sensitive; R8y: red, red-sensitive). The Voronoi tessellation (turquoise) shows the spatial organization of these terminals. d, Pattern of neuronal activation in response to an example stimulus in b . The yellow dots activate green and red receptors (R8p and R8y, respectively) in the appropriate Voronoi cells; the blue dots activate only the blue (R7) receptors. e Biologically accurate representation of a sample of 100 neurons in the areas of interest of the connectome (left hemisphere), with retina cells in dark green, lamina in blue, medulla in lime, lobula in yellow, and lobula plate in orange. We also show mushroom body KCapbp-m (decision-making) cells in red for reference. Image e was created with https://codex.flywire.ai/

Figure 2: Connectome randomization strategies and statistics. a–e, Toy illustrations of the five connectome ensembles. a, Biological connectome. b, Unconstrained randomization, where each neuron keeps the same number of outgoing connections, but targets are randomly shuffled. This produces networks with much larger total wiring and longer synapses on average than the biological connectome. c, Connection-pruned randomization, obtained by starting from the unconstrained network and then removing connections until the total wiring matches the biological one. Even after this pruning step, the mean connection length remains above biological values. d, Synapse-bin randomization, where neuron pairs are grouped by Euclidean soma-to-soma distance into 100 bins and synapses are shuffled only within the same bin. This keeps the global distribution of connection lengths, and therefore also preserves both the average synapse length and the total wiring budget. e, Neuron-bin randomization, where for each neuron its outgoing synapses are divided into 20 bins according to wiring length, and randomization is restricted to each bin separately. In this ensemble, the network wiring pattern is unchanged, and only the synaptic weights between connected neuron pairs are redistributed. f, Distribution of the synapse lengths for the five ensembles. The least constrained randomizations have much longer average lengths than the biological and the biologically constrained ones. g, h, Fraction of neurons ( g ) and Kenyon neurons ( h ) active after each message-passing step ( $N=3$ ). In these two plots we see that the least constrained networks very rapidly saturate the connectome, whereas the constrained ones, even though they reach large amounts $(\sim 80\%)$ of neurons in very few steps, are more constrained in the amount of neurons to use for computation. i, Mean distance of activated neurons after each step, which shows how the information travels very far very rapidly for unconstrained networks. j, Propagation patterns for the biological and random ensembles. In the plots, we show one out of every hundred active neurons in their 3D position given by the connectome. We can see how, in biological and biologically constrained ensembles, the information slowly moves from the eye to the rest of the brain, whereas in the unconstrained ones the whole brain is almost immediately active.

Figure 3: Connectome performance on various visual tasks. a, Example stimuli and classification accuracy for color discrimination across the biological connectome and four randomized ensembles; all five models reach 100% accuracy. b, Shape-recognition accuracy. The unconstrained and connection-pruned ensembles perform best, followed by the biological, synapse-bin, and neuron-bin models. c, Numerical-discrimination accuracy, aggregated across the area-controlled held-out images used in d . The unconstrained and connection-pruned ensembles perform best, while the biological connectome outperforms the two distance-binned ensembles. d, Numerical-discrimination accuracy as a function of the Weber ratio $r=N_{\mathrm{more}}/N_{\mathrm{less}}$ . Accuracy generally increases with ratio, showing the ratio-dependent pattern characteristic of approximate number perception. Only the area-controlled condition is shown, in which the total surface area of each color is equalized. a,b,e are shown as mean $\pm$ 95% CI (chance level $=50\%$ for binary tasks). All tasks use $5\times 1{,}000$ test images per class (see Methods).

Explore similar work

Oct 7, 2026cs.LG

A Drosophila Whole-Connectome Network Can Learn Human-Designed Cognitive Tasks

Can a biological wiring diagram serve as a useful computational substrate beyond the behaviors for which it evolved? We use the publicly released MaleCNS v1.0 connectome, reconstructed from a single adult male Drosophila specimen, as the fixed recurrent topology of an artificial network. We train separate models for bounded addition and for a controlled grounded relational language task built from a fixed 100-word lexicon. In both models, one scalar is learned per anatomical edge. The anatomical graph reaches 92.77% mean accuracy on held-out addition, compared with 67.93% for directed degree-preserving rewires. On the strict paired language endpoint, which matches original and order-reversed scenes to their corresponding descriptions, it reaches 61.59% across four fixed interfaces, compared with 44.17% for matched rewires. At the canonical interface, it ranks first in a fixed 21-graph comparison. On the matched 48-group intervention subset, shuffling task-defined sensory features reduces its score from 60.94% to 19.27%. Together, these results show that higher-order MaleCNS wiring provides a reusable inductive bias for bounded addition and grounded relational language.

Joonghui Cho, Minchan Kang, Daeshik Kim

Korea Advanced Institute of Science and Technology (KAIST)

Cognitive Science

Ask PDF

Oct 6, 2026cs.CV

From the Drosophila Visual Connectome to General-Purpose Computer Vision

Biological connectomes encode structured solutions to visual computation that may provide reusable inductive biases for artificial vision. We develop ConnectomeX around FlyVision, a trainable architecture that preserves parallel ON/OFF processing, recurrent computation and population-level graph interaction while scaling model capacity across tasks. FlyVision reached 99.34% accuracy on MNIST with 80,608 parameters and 78.03% on CIFAR-10 with 81,408 parameters. On ImageNet-1K, FlyVision Base and Large reached 60.79% and 66.25% top-1 accuracy with 1.8 and 3.7 million parameters, while a Large local-k7 model with a learned low-frequency branch reached 66.53%, compared with 69.25% for ResNet18 with 11.7 million parameters. On a 22-class skin-disease benchmark, FlyVision Large achieved 63.78% accuracy and 95.28% macro-AUROC with 2.99 million parameters. In four-class chest radiography, ImageNet-pretrained FlyVision Base and Large reached 92.60% and 92.76% accuracy with 1.33 and 2.97 million parameters, compared with 91.56% for ImageNet-pretrained ResNet18 with 11.18 million. BrainAGE extends FlyVision to volumetric T1-weighted MRI by applying a shared ImageNet-pretrained FlyVision Large encoder to 24 sagittal, coronal and axial slices per scan and combining slice-level age estimates by confidence-modulated Gaussian voting. On 433 held-out scans, three-axis fusion achieved a mean absolute error of 5.98 years and R^2 = 0.868. Across the 224x224 classification tasks, the best FlyVision configuration remained within three percentage points of ResNet18 on ImageNet-1K and skin-disease classification and exceeded it on chest radiography with substantially fewer parameters. These results show that a conserved connectome-informed computation can scale from compact recognition to large-scale natural and biomedical vision.

Zongyu Li, Akito Yamauchi, Huaizhi Liu +4

Department of Biomedical Engineering, Columbia University, New York, NY, USA · Department of Electrical Engineering, Columbia University, New York, NY, USA · Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA +2

Topic pending

Ask PDF

Sep 29, 2026cs.NE

How much of fly walking is written in the wiring?

Connectome models of the fly nerve cord generate walking-like motor rhythms, but oscillation alone does not show that the specific wiring matters. Here we provide, to our knowledge, the first test of which features of motor output depend on the specific wiring. We simulated the leg motor systems of two independent Drosophila connectomes, with synapse counts as fixed weights and glutamatergic synapses treated as inhibitory, and compared each with six families of rewired networks that preserve progressively more of its structure, using pre-registered criteria. We find that rhythm is generic but antagonist coordination is not: many rewired networks were more rhythmic than the real ones, yet the real wiring coordinated antagonistic motor pools more strongly than every rewired network, most of all at the thorax--coxa joint. We trace this specificity to how premotor input is allocated between antagonistic pools. Both connectomes carry Sherrington's reciprocal innervation---neurons that excite one pool inhibit its antagonist---and no rewired network does. Reassigning premotor inputs between the pools abolished coordination even when motor neurons' typical input strength changed little (all pre-registered criteria met in one connectome; same direction in the other). Coordination, not rhythm, therefore reveals whether a connectome model's wiring matters.

Isabel Guan, Yuntian Zhao, Dingyuan Zhang +2

The Hong Kong University of Science and Technology · ZENBOT · Nanyang Technological University +1

Neural Circuits Neurons Musculoskeletal Wiring

Ask PDF