From Movement to Mechanism: Understanding a neural network through fly leg movements

Abstract

From Movement to Mechanism — version 0.4 How can a visible leg movement help us understand a neural network? This illustrated, English-language computational preprint follows two related experiments based on maleCNS, the male fruit-fly central nervous system wiring map. The main text explains the research in accessible language; a technical appendix specifies the models, inputs, controls and numerical settings. The original M1/P7 experiment traces a change in a repeatable 100-millisecond foreleg movement to four motor outputs, an inhibitory route and signal timing. Version 0.4 retains those findings and the original six figures, then adds movement-dependent sensory feedback and a separate six-leg experiment using the full accepted network. In the new experiment, timed stimulation of nonmotor cells produces continuous preparation, whole-foot lift and return over 4.2 seconds. The primary left forefoot reaches 137.4 micrometres above the ground and stays at least 20 micrometres clear for 246.7 milliseconds, without contact or support force, while the other five feet support the body. The complete task passes an exact repeat and separate finer neural and physical calculations. Matched no-lift-command controls remain standing without a qualifying lift. The new figures show the implemented signal loop, actual saved leg poses and traces, and the relationship between the two experiments. The purpose is understanding: precise inputs, visible consequences and controlled interventions that other researchers can inspect and build on. The cellular, sensory and mechanical rules are supplied model components, not measurements of every biological cell. Live sensory delivery is verified; its benefit or necessity for the manoeuvre is not yet established. This is neither walking nor autonomous balance. The initial mounting fixture and fixed change in muscle effectiveness are explicitly disclosed. Files and scope The PDF is the reading edition; the DOCX is an editable copy. The evidence ZIP contains a file guide, all nine figures, original and new figure exports, full-resolution clearance/support traces, exact central-input schedules, model and sensory-port summaries, protocols, six confirmation reports, the return-input comparison and the attempt catalogue. Original-source and packaged-file hashes distinguish provenance from export integrity. The slowed leg-only GIF is retained from version 0.3. It illustrates the original M1/P7 comparison, not the new whole-foot lift. The supplement supports inspection and replotting, not a self-contained simulator rerun. It does not include the full connectome, compiled body/meshes, complete simulation histories or a portable research environment. Internal study numbers identify project stages, not separately published articles. No outside-team reproduction is claimed. This preprint has not undergone journal peer review. Author and licence Tomáš Kovářík is the sole author and an independent researcher. No external funding was received; no competing interests are declared. AI-assisted literature review, implementation, analysis and writing are disclosed in the manuscript. No new experiments on living animals were conducted. CC BY 4.0 applies to the paper and the author's original documentation, figure compositions and exports. Third-party materials retain their own rights. Version 0.3 remains available at https://doi.org/10.5281/zenodo.22787902. The all-versions DOI is https://doi.org/10.5281/zenodo.22787901.