BACKGROUND & DISCLOSURES
About FlyCast
Distinguishing biological observations from engineered computational choices.
Observed Biology vs. Engineered Modeling Choices
| OBSERVED (MaleCNS Dataset) | ENGINEERED (FlyCast System) |
|---|---|
| 166,691 physical neuron reconstructions across the male fruit fly central nervous system. | 166,483 nodes filtered by superclass annotation and treated as rate-neuron computational units. |
| Observed synaptic contact counts between pre- and postsynaptic pairs. | Synaptic weights transformed via log(1 + count) and normalized per postsynaptic row (L1 = 1.0). |
| Neurotransmitter identity predictions (acetylcholine, GABA, glutamate, etc.). | Unsigned reservoir mode by default to avoid speculative sign mappings; optional heuristic mode. |
| Biological sensory receptor classifications (optic lobes, antennal lobes). | Deterministic injection of external time series into 512 sensory neurons with uniform weights. |
| Complex living biophysical membrane dynamics, ion channels, and spike generation. | Leaky tanh rate dynamics and linear Ridge regression readout for discrete-time forecasting. |
The Research Question
“Can an evolved biological network provide computational utility when used directly as a fixed recurrent computer?”
Most modern neural networks are described as “brain-inspired.” FlyCast takes a different path: it investigates whether the exact structural topology of an actual animal nervous system exhibits useful nonlinear memory, separation, and dynamic properties when driven by external data.
Data Attribution
FlyCast relies on the monumental connectomics work accomplished by the MaleCNS collaboration:
- HHMI Janelia Research Campus (FlyEM Project Team)
- University of Cambridge (Department of Zoology)
- MRC Laboratory of Molecular Biology (LMB)
- Google Research
MaleCNS data is provided under Creative Commons Attribution (CC BY). FlyCast is an independent open-source experiment created to explore reservoir computing over biological graphs.