Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 81 of the 81 references it cites were found in a registry or an index; 18 of the 42 sources its claims cite were checked for support.

A virtual rodent predicts the structure of neural activity across behaviours

Open access

Abstract

Animals have exquisite control of their bodies, allowing them to perform a diverse range of behaviours. How such control is implemented by the brain, however, remains unclear. Advancing our understanding requires models that can relate principles of control to the structure of neural activity in behaving animals. Here, to facilitate this, we built a ‘virtual rodent’, in which an artificial neural network actuates a biomechanically realistic model of the rat1 in a physics simulator2. We used deep reinforcement learning3, 4–5 to train the virtual agent to imitate the behaviour of freely moving rats, thus allowing us to compare neural activity recorded in real rats to the network activity of a virtual rodent mimicking their behaviour. We found that neural activity in the sensorimotor striatum and motor cortex was better predicted by the virtual rodent’s network activity than by any features of the real rat’s movements, consistent with both regions implementing inverse dynamics6. Furthermore, the network’s latent variability predicted the structure of neural variability across behaviours and afforded robustness in a way consistent with the minimal intervention principle of optimal feedback control7. These results demonstrate how physical simulation of biomechanically realistic virtual animals can help interpret the structure of neural activity across behaviour and relate it to theoretical principles of motor control. We built an artificial neural network to control a biomechanically realistic virtual rodent, which, when trained to imitate real rats, predicts neural activity and variability across natural behaviours.

Overview

CoChat Summary

TL;DR

The authors built a “virtual rodent”: an artificial neural network implementing inverse dynamics that actuates a biomechanically realistic rat model in MuJoCo, trained with deep reinforcement learning to imitate freely moving rats. Its network activity predicted sensorimotor striatum (DLS) and motor cortex (MC) neural activity better than movement features, and its latent variability structured action variability in a way consistent with the minimal intervention principle.

Key findings

  • The virtual rodent imitated held-out rat behavior across a 607-hour dataset; recurrent decoders outperformed other architectures, and most deviations from real kinematics were attributable to accumulated center-of-mass error.
  • In DLS (3 animals, 353.5 hours, 1249 neurons) and MC (3 animals, 253.5 hours, 843 neurons), inverse dynamics model activations predicted single-unit activity better than kinematic or dynamic representational features, with best results from the first decoder layer (p < .001, one-sided permutation test; N = 732 DLS and 769 MC significantly predicted neurons).
  • Representational similarity analysis showed that DLS and MC population activity was more similar to inverse dynamics model representations than to representational models, and this similarity correlated with imitation performance and robustness.
  • Latent variability structured action variability across behaviors; latent variability alone discriminated behavior (p < .001), and uniform or inverted variability perturbations reduced robustness and increased termination (p < .001), consistent with the minimal intervention principle; latent variability also predicted the structure of neural variability across behaviors.

Methods

  • Freely moving female Long Evans rats were recorded in a circular arena with six cameras and custom 128-channel tetrode drives targeting DLS or MC; the dataset spans 607 hours of simultaneous electrophysiology and videography.
  • The MIMIC pipeline used DANNCE for 3D pose estimation of 23 keypoints, STAC to register a 74-degree-of-freedom skeletal model (38 controllable actuators) to those keypoints, and deep reinforcement learning (MO-VMPO) to train ANNs implementing inverse dynamics to control the model in MuJoCo at 50 Hz, using 842 five-second behavioral clips.
  • Analyses included Poisson generalized linear models with cross-validation to predict spike counts from representational features versus ANN layers, representational similarity analysis using cross-validated Mahalanobis distance and whitened unbiased cosine similarity, and variability analyses that resampled the latent space 50 times per timestep and perturbed latent variability structure.

Limitations

  • The paper states that the model uses point-torque actuation and omits muscular actuation, opting for the simplest abstraction sufficient for its objectives; it also notes that the models are most appropriate for short-timescale motor dynamics rather than long-term behavioral organization.
  • The paper states that directly measuring instantaneous neural variability is not possible, so neural variability was approximated with a 1-second sliding-window variance; it also notes that alternative control models trained via supervised learning may differ from reinforcement-trained composite controllers.
  • Evident from the text: neural comparisons are correlational and cover only DLS and MC with 3 animals each, and the virtual rodent’s behavior deviated slightly from real rats (e.g., center-of-mass drift), though the inverse dynamics model still outperformed representational models.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41586-024-07633-4

Cites

81 of 81 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

9 Valid · 7 Not valid · 6 Not checked · 95 Own results and interpretations

45 more claims

Create a free account to see all 117 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 38
  • It may state two separate claims as one. 35
  • The quote only partly states this claim. 3

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

3 of 3

Citations

Cites

Show all 76 references

Research works printed but not linked: 5

  1. Hasenclever, L., Pardo, F., Hadsell, R., Heess, N. & Merel, J. CoMic: complementary task learning & mimicry for reusable skills. In Proc. 37th International Conference on Machine Learning (eds Daumé, H. & Singh, A.) 4105–4115 (PMLR, 2020).
  2. Jordan, M. I. in Handbook of Perception and Action, Vol. 2 (ed. Heuer, H.) Ch. 2 (Academic Press, 1996).
  3. Lai, L. & Gershman, S. J. in Psychology of Learning and Motivation, Vol. 74 (ed. Federmeier, K. D.) Ch. 5 (Academic Press, 2021).
  4. Kingma, D. P. & Ba, J. Adam: a method for stochastic optimization. In Proc. 3rd International Conference on Learning Representations (eds Bengio, Y. & LeCun, Y.) (2015).
  5. Maas, A. L., Hannun, A. Y. & Ng, A. Y. Rectifier nonlinearities improve neural network acoustic models. In Proc. 30th International Conference on Machine Learning (ICML) (2013).

Cited by

See all 70

Metadata

Authors
Aldarondo, D., Merel, J., Marshall, J. D., Hasenclever, L., Klibaite, U., Gellis, A., Tassa, Y., Wayne, G., Botvinick, M., & Ölveczky, B. P.
Affiliations
  • Harvard University (US)
  • Google DeepMind (United Kingdom) (GB)
Corresponding author
Aldarondo, Diego
Venue
Nature
Published
2024-06-11
Volume
632
Issue
8025
Pages
594-602
ISSN
  • 0028-0836 (ISSN-L)
  • 1476-4687
OpenAlex
W4399534689
Cited by
70
References
76
Citations by year
2026: 28 · 2025: 29 · 2024: 12
FWCI
16.71
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature (publishedVersion)
  • PubMed (publishedVersion)
  • PubMed Central (acceptedVersion)
Funders
  • National Institutes of Health (5r01ns099323-08)
  • National Institute of General Medical Sciences (R01GM136972)
  • National Institute of Neurological Disorders and Stroke (R01-NS099323, F99NS125834)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Motor Control and Adaptation
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural dynamics and brain function
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Memory and Neural Mechanisms
Keywords
motor control, physical simulation, neural decoding
MeSH terms
Virtual Reality, Deep Learning, Animals, Behavior, Animal, Female, Models, Neurological, Motor Cortex, Movement, Neural Networks, Computer, Rats, Long-Evans, Rats, Sensorimotor Cortex

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-01, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-01 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: A virtual rodent predicts the structure of neural activity across behaviours. https://app.cochat.ai/papers/10.1038/s41586-024-07633-4
The paper
Aldarondo, D., Merel, J., Marshall, J. D., Hasenclever, L., Klibaite, U., Gellis, A., Tassa, Y., Wayne, G., Botvinick, M., & Ölveczky, B. P. (2024). A virtual rodent predicts the structure of neural activity across behaviours. Nature. https://doi.org/10.1038/s41586-024-07633-4

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.