Connectome-constrained neural networks are often evaluated against sparse random controls and then interpreted as evidence that biological graph topology improves learning efficiency. We revisit that claim in a controlled flyvis-based study using a Drosophila connectome, a naive self-loop-matched random graph, and a degree-preserving rewired null. Under weak controls, in which both models were recovered from a connectome-trained checkpoint and the null matched only global graph counts, the connectome appeared substantially better in early loss, mean activity, and runtime. That picture changed under stricter controls. Training both graphs from a shared random initialization removed the early loss advantage, and replacing the naive null by a degree-preserving null removed the apparent activity advantage. A five-sample degree-preserving ensemble and a pre-training activity-scale diagnostic further strengthened this revised interpretation. We also report a descriptive mechanism analysis of the earlier weak-control comparison, but we treat it as behavioral characterization rather than proof of causal superiority. We show that previously reported topology advantages in connectome-constrained neural networks can arise from initialization and null-model confounds, and largely disappear under fair from-scratch initialization and degree-preserving controls.
Figures & tables
Figure 1: Control ladder used in the revision study. The original observation compared a connectome graph to a self-loop-matched random graph after checkpoint recovery. The corrected analysis then removed checkpoint initialization and strengthened the null model by preserving the directed degree sequence. The substantive scientific conclusion changes across these control levels.
Figure 2: Architecture used in the study. A MovingEdge stimulus drives a graph-constrained recurrent flyvis network whose connectivity mask is given by either the empirical connectome, a naive random graph, or a degree-preserving random graph. A linear decoder reads pooled central-cell activity and predicts the 2D motion direction target. The key comparison in the paper is not between different decoders or optimizers, but between different graph masks under matched nodes, edges, self-loops, and parameterization.
Graph
Nodes
Edges
Self-loops
Trainable
Fixed
Connectome
45,669
1,513,231
12,380
734
2959
Naive random
45,669
1,513,231
12,380
734
2959
Degree-preserving random
45,669
1,513,231
12,380
734
2959
Table 1: Graph-level matching constraints used throughout the study. The recurrent network parameterization is identical across all graph conditions.
Figure 3: Matched-step training curves under checkpoint-based initialization and a naive random null. The connectome appears to outperform the control across all three metrics.
Figure 4: Matched-step summary at 5 and 10 steps under weak controls. All three metrics favor the connectome prior to applying stricter controls.
Figure 5: Degree-preserving ensemble variability at 5 matched steps. Each point represents one sample-seed comparison. Loss differences remain near zero, activity differences are slightly negative, and runtime differences are modest.
Comparison
Steps
Δ loss
Δ activity
Δ elapsed (s)
Original checkpoint + naive random
5
0.1841
1.2056
57.00
Original checkpoint + naive random
10
0.0583
0.6397
99.11
Random init + naive random
5
-0.0020
0.0323
57.60
Random init + naive random
10
-0.0020
0.0202
369.59
Random init + degree-preserving
5
0.0003
-0.0106
8.38
Random init + degree-preserving
10
-0.0018
-0.0087
3.98
Table 2: Control progression across experiments. Deltas are control minus connectome.
Figure 6: Control progression. Each bar shows control-minus-connectome differences across successive control levels. Loss differences vanish under shared random initialization, and activity differences vanish under the degree-preserving null.
Metric
Connectome
Naive random
Mean absolute activity
0.6648 ± 0.0481
1.7355 ± 0.0150
Node activity Gini
0.4319 ± 0.0372
0.4694 ± 0.0033
Top 10% node activity fraction
0.2807 ± 0.0308
0.3333 ± 0.0027
Edge usage Gini
0.7786 ± 0.0080
0.8232 ± 0.0025
Top 10% edge usage fraction
0.6031 ± 0.0159
0.6884 ± 0.0034
Mean total activity over time
30,358.9 ± 2,195.8
79,258.3 ± 686.7
Table 3: Descriptive activity and usage metrics for the connectome and naive random graphs under the weak-control comparison. These values summarize behavior in that setting and are not interpreted as causal evidence under the corrected controls.
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)
Connectome-constrained neural networks make biological components directly accessible to intervention, but their mechanistic conclusions can still depend on the comparison model, intervention, circuit representation and trained solution. We stress-test these dependencies in flyvis, a Drosophila motion-vision model, using its 50-model ensemble, task-matched causal CNN controls and replicated interventions. At the level of model behaviour, three-level reverse-phi reversal in flyvis is strongly period- and velocity-dependent and also appears in task-matched CNNs; neither family reproduces the fly's stationary-sawtooth response. Adversarial direction reversal requires a perturbation bound of 0.21--0.27 times stimulus contrast in flyvis versus 0.13--0.25 in CNNs, but the gap over a parameter-matched CNN largely closes for larger CNNs. At the level of mechanism, matching attribution to the intervention matters: for held-out-mean replacement, first-order edge attribution patching reaches pooled AUROC 0.970 and Spearman 0.872, versus 0.933 and 0.126 when its zero-aligned score is evaluated against the same ground truth. Circuit conclusions also vary with representation and ensemble member. Greedy cell-type sets meet a model-relative T4/T5 tuning criterion in 25 of 50 models, while T4/T5 output silencing reduces the tested optic-flow behaviour to at most two of eight clips in 40. In six models where the cell-type search first succeeds, the corresponding connection sets can be pruned to 13--31 added classes; under an additional two-sided response-amplitude check, only two of those six first-success cell-type sets still meet the tuning threshold. These results support a narrower use of connectome-constrained models: they make mechanistic hypotheses directly testable, but the resulting claims must be scoped to the model family, intervention, representation and trained solution.
Aashish Dhawan
Department of Computer & Information Science & Engineering University of Florida
Randomised copies of a connectome are the usual baseline for asking whether measured wiring matters, and the answer depends on what the randomisation preserves. We evolved embodied foraging agents whose brains are a compressed adult Drosophila connectome (FlyWire v783; 512 cell-type groups and 1,000 Kenyon cells) alongside agents built on randomised wiring, in pre-registered experiments with ten seeds, four ecologies and 600 generations. Two standard randomisations, a column shuffle and degree-preserving edge swaps, route 10.6 to 10.7 % of olfactory output directly onto descending motor groups, against 0.012 % in the connectome. On the registered primary endpoint, fitness averaged over the run, no difference was detected; at the last common-garden probe the connectome was behind both controls (-0.22 and -0.20 fitness units on seed means). Against controls that keep every sensory-output and motor-input edge and rewire only the interior, the seed-mean difference lay within a +/-0.10 equivalence bound (+0.002 and -0.074, unchanged under a calibration that also matches activity spread), although per ecology the interior column shuffle was ahead by 0.26 in one of four ecologies at ten seeds, a lead that ten further pre-registered seeds did not replicate. Rewiring the connectome so that it acquires the shortcut raised its fitness by 0.44 (10 of 10 seeds) and its dependence on olfaction from 0.15 to 0.99; graded doses raised both in step; at comparable swap counts the full dose was ahead of an interior-only sham by 0.53 (10 of 10 seeds); and a sham that rewired the same boundary edges without creating shortcuts matched the connectome (+0.007) while the full dose was ahead of it by 0.60. What a null preserves at the sensory-motor boundary can decide an evolutionary connectome comparison, and sensory-to-motor path statistics belong next to the degree statistics a null is said to preserve.