Connectome Coarse-Graining and Whole Brain Emulation Resolution Across Scales
Must a whole brain emulation reproduce every single microscopic synapse to predict how a nervous system responds, or does functional neural dynamics survive grouping neurons into much coarser biological modules? In a November 4, 2026 preprint titled Connectome Structure Constrains Neural Activity Across Scales (bioRxiv 10.64898/2026.11.03.762695), Carter Sale and Michael J. Richardson test exactly how much structural detail can be compressed before functional prediction fails. Analyzing Caenorhabditis elegans perturbation data alongside Drosophila developmental lineages, they find that direct single-hop synaptic counts poorly predict evoked neural responses, whereas path-based connectivity predictors remain accurate even after coarse-graining the connectome into biological cell classes or roughly nine macroscale functional subsystems. Because bioRxiv 10.64898/2026.11.03.762695 is a November 2026 preprint awaiting peer review, its coarse-graining bounds stand as pre-publication evidence rather than a settled engineering rule for mammalian brains.
For our Brain Emulation research section, our Connectome Atlas Console, and the broader scientific questions tracked in how scientists race to define and test AI consciousness in 2026, Sale and Richardson address the central scaling bottleneck of mind uploading. Scanning an 86-billion-neuron human brain with 100 trillion synapses at nanometer electron-microscopy fidelity remains decades away. Knowing which anatomical details govern evoked neural activity and which details average out safely under biologically aligned coarse-graining directly shapes every serious roadmap for brain emulation.
Why Direct Synapses Underperform Two-Step Path Predictors
Connectomics projects invest massive imaging budgets to count individual chemical synapses and gap junctions between pairs of neurons. Sale and Richardson begin by asking a direct empirical question on C. elegans, the model organism where both complete synaptic wiring diagrams and whole-brain single-neuron stimulation datasets exist. When you stimulate neuron A, does the direct synaptic weight between A and B accurately predict whether neuron B responds?
Their analysis shows that direct one-hop synaptic connectivity is a relatively poor predictor of evoked neural activity. In contrast, a parameter-free path-based predictor performs markedly stronger across the same perturbation experiments. Crucially, predictive accuracy depends heavily on two-step paths and the broader spatial arrangement of connections across the network rather than raw direct synapse counts alone. Indirect polysynaptic routing shapes functional communication even in a 302-neuron nervous system.
Coarse-Graining Into Cell Classes and Nine Subsystems
Having established a baseline functional predictor on the full wiring graph, Sale and Richardson systematically coarse-grain the connectome, merging individual neurons into progressively larger clusters and testing whether the compressed graph retains predictive accuracy.
- Grouping C. elegans neurons according to their canonical biological cell classes preserves functional prediction accuracy even though fine-grained individual path structure is discarded.
- Arbitrary or randomized groupings of the same cluster size sharply degrade functional predictions, proving that biological organization carries real compressed information rather than mere dimensionality reduction.
- Optimizing partitions directly from the network’s connectivity structure reveals alternative coarse-grainings that preserve evoked-response predictability all the way down to roughly nine macroscale groups corresponding to major functional subsystems of the worm’s nervous system.
- Partitions inferred from newly hatched juvenile C. elegans connectomes retain predictive power on adult brains, and extending the structural test to Drosophila shows that developmental lineage-based partitions preserve path structure significantly better than random partitions of equal size.
| Emulation / Modeling Scale | Anatomical Unit Preserved | Structural Detail Retained | Functional Prediction Status (Sale & Richardson 2026) |
|---|---|---|---|
| Molecular / Subcellular | Ion channels, receptor kinetics, neuromodulators | Highest biophysical detail | Not modeled directly in static connectome graph |
| Single-Synapse Graph | Individual neuron-to-neuron chemical and electrical counts | Full microscopic wiring | Direct 1-hop weights perform poorly; 2-step path predictors perform well |
| Biological Cell Classes / Lineages | Shared cell-type identities (C. elegans) or clonal lineages (Drosophila) | Microscopic paths compressed | Preserves evoked functional predictions cleanly over random nulls |
| ~9 Functional Subsystems | Connectivity-derived macroscale partitions | Extreme structural compression | Retains predictive power across major nervous-system subsystems |
Connecting Coarse-Graining to Sandberg-Bostrom, Worm Reservoirs, and MaleCNS
Reading Sale and Richardson’s November 2026 preprint beside three other studies already archived on this site resolves a striking apparent tension in connectomics.
First, in Felix Reimers, Stefano Nichele and colleagues’ September 2026 benchmark of C. elegans connectomes as echo-state reservoirs, randomized null graphs frequently matched or outperformed biological wiring on generic machine-learning tasks. Sale and Richardson show the crucial complementary side. When you evaluate actual evoked biological responses and respect biological cell classes or developmental lineages, real connectome organization decisively beats randomized groupings, even after aggressive coarse-graining.
Second, in the MaleCNS adult Drosophila whole-connectome cognitive task and Intel Loihi 2 simulation study, scaling up to roughly 140,000 to 166,700 neurons requires mapping sparse recurrent pathways efficiently onto neuromorphic silicon. Sale and Richardson’s finding that developmental lineage partitions in Drosophila preserve path geometry points toward principled ways of compressing insect-scale and eventually mammalian-scale connectomes without losing their macroscale communication structure.
Third, the result speaks straight to the level-of-emulation taxonomy introduced in Anders Sandberg and Nick Bostrom’s Whole Brain Emulation Roadmap and updated in the 2025 State of Brain Emulation Report. Emulation roadmaps have always faced an open trade-off between simulating down to molecular biochemistry versus abstracting up to circuit motifs. Sale and Richardson provide concrete empirical evidence that evoked functional dynamics can survive at resolutions significantly coarser than raw microscopic synapse maps, provided the coarse-graining follows genuine biological cell classes and developmental lineages rather than arbitrary geometric grids.
What Connectome Coarse-Graining Tells Us About Consciousness
Why does a coarse-graining study on worm and fly wiring graphs bear directly on consciousness science? It probes the exact fault line separating functionalist emergentism from fine-grained causal-structure and biological-substrate theories of mind.
- Under functionalist emergentism, if a biologically aligned coarse-graining preserves the macroscale functional dynamics, sensorimotor transformations, and subsystem interactions of a nervous system, then an emulation running at that functional grain captures the computational organization relevant to cognition and emerging consciousness.
- Under Integrated Information Theory (IIT 4.0), coarse-graining raises an acute theoretical question explored in Erik Hoel’s work on causal emergence. Compressing micro-elements into macro-subsystems can either increase or destroy intrinsic cause-effect power depending on whether the macro-scale grain constitutes the system’s true maximal scale of causal integration.
- Under biological naturalism, critics would note that predicting evoked linear or path-based neural responses in C. elegans still leaves out extrasynaptic neuropeptides, volume-transmitted neuromodulators, and metabolic homeostasis, all of which operate outside pure synaptic wiring graphs.
Sale and Richardson do not claim that nine coarse-grained worm subsystems experience anything. Rather, they demonstrate empirically that functional organization is far more resilient to biologically grounded abstraction than raw anatomical wiring diagrams suggest.
Comparison to The Consciousness AI
This resilience to biologically principled abstraction sits right at the foundation of The Consciousness AI. Rather than waiting for complete nanometer-resolution connectomes of vertebrate brains, our architecture builds a functionalist modular hierarchy grounded in neuroanatomical subsystems (sensory tectum, affective valuation core, oscillatory binding layer, and global workspace competition), tested openly against explicit probes and recorded episodes.
At the same time, our project treats that architectural abstraction as a testable hypothesis rather than an assumed equivalence. As documented in docs/instrument_inventory.md, 0 of 21 consciousness instruments are trusted yet, and verified probes operate on closed-set 72-stimulus tests. Sale and Richardson’s method offers a clear benchmark principle for future multi-scale experiments on our Connectome Atlas and Brain Emulation hub. Any coarse-grained neural architecture earns its validity only when it outperforms randomized null partitions at predicting actual evoked dynamics across biological subsystems.
Preprint Limits and Open Research Questions
Three clear boundaries frame bioRxiv 10.64898/2026.11.03.762695 ahead of peer review. First, evoked single-neuron stimulation datasets currently exist at whole-brain scale only for C. elegans, so extending functional coarse-graining tests to Drosophila relies on structural path preservation rather than full evoked-activity measurements across all 140,000 fly neurons. Second, static connectome paths exclude wireless neuropeptide signaling, which plays a major documented role in C. elegans neural communication. Third, whether mammalian neocortex with its massive recurrent dendritic integration obeys comparable cell-class coarse-graining bounds remains an open empirical question for future mouse-cortex projects.
Sources
- Carter Sale and Michael J. Richardson. Connectome Structure Constrains Neural Activity Across Scales. bioRxiv preprint
10.64898/2026.11.03.762695, November 4, 2026. doi:10.64898/2026.11.03.762695. - Felix S. Reimers, Ola Huse Ramstad, Aliaksandr Hubin, and Stefano Nichele. Benchmarking the Connectomes of Caenorhabditis elegans within the Reservoir Computing Framework. arXiv preprint
arXiv:2609.30508, September 24, 2026. arXiv:2609.30508. - The Consciousness AI Public Repository (
tlcdv/the_consciousness_ai). Architecture documentation anddocs/instrument_inventory.md.
Researchers covered here
- Carter SaleComputational Connectomics and Neural Dynamics ResearchMulti-scale connectome coarse-graining and path-based prediction of evoked neural activity across C. elegans and Drosophila
- Michael J. RichardsonComputational Neuroscience and Connectome Scaling ResearchMulti-scale connectome structure-function modeling, coarse-graining bounds, and dynamical systems in nervous systems