The working assumption behind most whole brain emulation proposals is that the connectome — the complete map of synaptic connections between neurons — is the primary information substrate that would need to be copied to transfer a mind. If you reconstruct the wiring, the argument goes, the function follows. A February 2026 bioRxiv preprint challenges the completeness of that assumption without rejecting the goal. It does so by revealing how much of what neurons do is organized by their molecular identity, not just their synaptic position.

The platform is called WARP. It stands for whole-brain ARchitecture mapping Platform, and it integrates three technologies that have previously been used separately: whole-brain calcium imaging of neuronal activity during active behavior, expansion-assisted spatial transcriptomics, and cellular-level registration that aligns both datasets to the same anatomical reference frame. The result is a single experiment that simultaneously records what neurons do and which genes they express, across every major brain region, at cellular resolution.

What the Research Did

The team developed WARP as an experimental-computational platform for the larval zebrafish, a model organism widely used in neuroscience because its optical transparency allows non-invasive whole-brain imaging without the need for surgical procedures. It is one of the few vertebrate species where this scale of simultaneous imaging is biologically feasible.

The platform identifies more than 2,000 molecularly defined neuronal subpopulations by linking their behavior-evoked activity, gene expression profiles, and anatomical locations. This is a substantially finer-grained taxonomy than previous functional or genetic atlases of the zebrafish brain achieved separately. At the genetic level, the key finding is that gene-matched neurons — neurons belonging to the same molecularly defined cluster — consistently show stronger activity correlations with each other than neurons of different molecular types located in adjacent tissue.

The implication is direct: gene expression is not merely an annotation on neural circuitry. It is a principal organizer of functional architecture. Neurons that share molecular identity tend to participate in common computational processes, even when their anatomical positions are not immediately adjacent. The wiring alone does not fully predict this pattern.

Why This Complicates the Connectome-First Model

The simplest version of the mind uploading argument runs as follows: the brain is, at the functional level, an information-processing network. Map every connection with sufficient precision — every synaptic weight, every anatomical pathway — and you capture the substrate of cognition. Silicon can then implement the same computation.

The WARP findings add a different layer to this picture. If molecular identity is an independent organizer of functional activity, then a connectome that records only synaptic anatomy may miss a co-determining variable. Two neurons with identical synaptic inputs and outputs but different gene expression profiles may behave differently, because their intrinsic properties — membrane channels, receptor densities, signaling cascades — differ based on which genes they express.

This is not a new theoretical concern. Neuroscientists have known for decades that intrinsic neuronal properties vary enormously across cell types, and that this variation is largely genetically determined. What the WARP platform contributes is a systematic, whole-brain demonstration of how tightly molecular identity tracks functional behavior across thousands of defined cell populations simultaneously.

The broader implication for WBE is that a complete upload may need to preserve the molecular identity of each neuron alongside its synaptic wiring — not merely the architecture of the network, but the molecular state of each node. This is a deeper challenge than the connectome challenge alone.

Connecting to the Amyloid Memory Layer

A related finding published in January 2026 from the Stowers Institute found that the nervous system deliberately forms amyloid protein aggregates inside neurons as a mechanism for storing long-lasting memories. That research proposed that the molecular intra-cellular layer holds information that purely electrophysiological models collapse away. The WARP findings push in the same direction from a different angle: gene expression represents another sub-synaptic organizational layer that connectomics, as currently practiced, does not capture.

Both findings suggest that what the brain needs to preserve for memory to survive is not simply a graph of connections, but a molecular state distributed across individual cells. The M. Schons Asimov Press essay on the core requirements for brain emulation identifies this as structural fidelity — but structural fidelity defined at what resolution? The WARP paper suggests that molecular-level cell-type identity is part of the answer.

The Complementarity with Sequencing-Based Connectomics

The Connectome-seq paper published in Nature Methods in April 2026 demonstrated that barcode sequencing can map synaptic connections between 1,000+ neurons at single-synapse resolution, at roughly ten times the cost-efficiency of electron microscopy based approaches. That approach maps wiring. WARP maps molecular identity and activity. Neither captures the other.

A comprehensive brain emulation pipeline would need both. The sequencing side provides the anatomical wiring diagram. A WARP-equivalent approach provides the molecular cell-type annotations that determine how individual nodes in that diagram behave intrinsically, independent of their synaptic inputs.

Research teams are aware of this integration issue. The WARP preprint explicitly frames the platform as a foundation for cross-experiment discovery, high-throughput function-to-gene mapping, and scalable circuit modeling — all of which are steps toward building integrated brain atlases where wiring and molecular identity are recorded in the same reference space. The Allen Institute’s ongoing Human Cell Atlas of the Brain is pursuing similar integration at the structural level, though not yet with simultaneous functional readout.

AI-assisted segmentation tools are rapidly accelerating anatomical connectomics. The equivalent acceleration in molecular-functional mapping is less mature. WARP represents early progress on closing that gap.

The Translational Gap

A word of caution about distance. The work was done in larval zebrafish, a model with roughly 100,000 neurons in its brain at the larval stage. Its relevance to mammalian neuroscience is indirect. Translating this platform to a mouse, let alone a human, requires navigating biological differences — mammalian brains have far greater cellular diversity, are far larger, and are not optically transparent. The calcium imaging methods used in WARP rely on larval zebrafish transparency. Different imaging modalities would be needed for mammalian tissue, most of them with lower throughput or resolution tradeoffs.

The significance of the platform is therefore primarily methodological and conceptual at this stage. It demonstrates that simultaneous whole-brain activity and gene expression mapping is achievable in a vertebrate nervous system, establishes a proof-of-concept framework for integrating these modalities, and produces data that reveals functional organizational principles invisible to anatomy alone. Whether those principles are conserved in the mammalian brain — and whether they would need to be preserved in a mind upload — remain open questions that WARP’s authors would be the first to acknowledge.

TRL for this platform is 3: a laboratory proof of concept demonstrated in a model organism, with substantial work remaining before it can be adapted for mammalian or human tissue mapping.

What a Molecular-Aware Upload Would Actually Require

If the WARP findings are accepted as pointing to a general principle — that molecular cell-type identity is a co-organizer of brain function — an upload-quality brain scan would need to record, for every neuron:

1. Its synaptic connectivity (which other neurons it connects to, and with what weights)
2. Its gene expression profile (which of the thousands of neuronal cell types it belongs to)
3. Its epigenetic state (which genes are currently active, since expression changes with experience)
4. Its protein distribution (which molecular machinery is actually present, generated by expressed genes)
5. Its current firing history and neuromodulatory context (which background signals are modulating its current excitability)

The physics wall of non-destructive brain scanning makes acquiring even the first item from a living brain technically infeasible at synaptic resolution. Adding the molecular-level requirements items two through five multiplies the problem. For a fixed post-mortem brain, electron microscopy methods capture structural anatomy. Gene expression sequencing can be applied to preserved tissue. But recovering epigenetic state and real-time protein distribution from a dead brain is a fundamentally different challenge.

WARP is not a problem-solver for this layer of the WBE challenge. It is a clarifier. It shows, empirically, that the molecular layer matters more than purely connectomics-first models assumed. That makes it an important paper for anyone thinking rigorously about what uploading a mind would actually require.

Official Sources

• bioRxiv preprint (Feb 10, 2026): Whole-Brain Co-Mapping of Gene Expression and Neuronal Activity at Cellular Resolution in Behaving Zebrafish — https://www.biorxiv.org
• NIH coverage of the WARP platform: https://www.nih.gov
• Platform: WARP (Whole-Brain ARchitecture mapping Platform) — zebrafish larval brain, cellular resolution, simultaneous activity + transcriptomics
• TRL: 3 (validated in zebrafish; mammalian translation pending)