A foundation deep-learning model trained on two-photon calcium imaging of mouse…

A foundation deep-learning model trained on two-photon calcium imaging of mouse visual cortices generalizes to new mice with minimal training, predicts responses to held-out stimulus classes, and accurately classifies anatomically defined excitatory cell types in the MICrONS dataset of more than 70,000 neurons.

{
  "n": 70000,
  "doi": "10.1038/s41586-025-08829-y",
  "claim": "A foundation deep-learning model trained on two-photon calcium imaging of mouse visual cortices generalizes to new mice with minimal training, predicts responses to held-out stimulus classes, and accurately classifies anatomically defined excitatory cell types in the MICrONS dataset of more than 70,000 neurons.",
  "cite_key": "Wang2025b",
  "evidence": "Demonstrates that emerging foundation models can predict functional and anatomical properties of mouse cortex from large-scale 2P data — directly addresses cluster_14 'near-horizon tools' topic.",
  "effect_size": "MICrONS dataset >70,000 neurons; model trained on eight mice",
  "text_access": "fulltext",
  "study_system": "Mouse visual cortex; two-photon imaging + MICrONS connectome; deep-learning foundation model",
  "argument_role": "supporting",
  "replication_status": "single-study",
  "claim_source_sentence": "In the Machine Intelligence from Cortical Networks (MICrONS) dataset 2 , which contains functional recordings and nanoscale anatomy of more than 70,000 neurons, our model accurately classified anatomically defined types of excitatory neurons.",
  "source_provenance_status": "ok",
  "replication_evidence_dois": [],
  "effect_size_source_sentence": "With a subset of these data, we trained a deep neural network on recordings from eight mice, producing a ‘foundation core’ that captured shared latent representations and predicted neuronal responses across mice and cortical areas."
}

[
  "paper:paper-a802f6ac77d4"
]

In the Machine Intelligence from Cortical Networks (MICrONS) dataset 2 , which contains functional recordings and nanoscale anatomy of more than 70,000 neurons, our model accurately classified anatomically defined types of excitatory neurons.

[
  {
    "ref": "paper:paper-a802f6ac77d4"
  }
]

15. Methodological limits and emerging tools — what current mouse-cortex tools cannot yet measure about E→E recurrence (subthreshold network activity, fast plasticity in vivo, millimetre-scale dynamic connectomes), and what is on the near horizon

{
  "mode": "public_source_pointer_with_short_context",
  "notes": [
    "Local review repositories are read-only inputs.",
    "SciDEX stores paper metadata, structured evidence, file pointers, and short citation contexts; it does not copy full review prose."
  ],
  "source_commit_sha": "79ce062d54a924ce05953ec90aa9d26044d2b48f",
  "source_repository_url": "https://github.com/AllenNeuralDynamics/ComputationalReviewRecurrence"
}