Abstract

Brains are not engineered solutions to a well-defined problem but arose through selective pressure acting on random variation. It is therefore unclear how well a model chosen by an experimenter can relate neural activity to experimental conditions. Here we developed 'Model identification of neural encoding (MINE)'. MINE is an accessible framework using convolutional neural networks (CNN) to discover and characterize a model that relates aspects of tasks to neural activity. Although flexible, CNNs are difficult to interpret. We use Taylor decomposition approaches to understand the discovered model and how it maps task features to activity. We apply MINE to a published cortical dataset as well as experiments designed to probe thermoregulatory circuits in zebrafish. MINE allowed us to characterize neurons according to their receptive field and computational complexity, features which anatomically segregate in the brain. We also identified a new class of neurons that integrate thermosensory and behavioral information which eluded us previously when using traditional clustering and regression-based approaches.

Data availability

All data generated in this study is publicly available. Links are provided in the 'Materials and Methods - Code and data availability' section.

The following data sets were generated
The following previously published data sets were used

Article and author information

Author details

  1. Jamie D Costabile

    Department of Neuroscience, The Ohio State University, Columbus, United States
    Competing interests
    Jamie D Costabile, is employed by Hitachi Solutions America, Ltd., Irvine, CA.
  2. Kaarthik A Balakrishnan

    Department of Neuroscience, The Ohio State University, Columbus, United States
    Competing interests
    No competing interests declared.
  3. Sina Schwinn

    Department of Neuroscience, The Ohio State University, Columbus, United States
    Competing interests
    No competing interests declared.
  4. Martin Haesemeyer

    Department of Neuroscience, The Ohio State University, Columbus, United States
    For correspondence
    haesemeyer.1@osu.edu
    Competing interests
    No competing interests declared.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-2704-3601

Funding

National Institutes of Health (5R01NS123887-02)

  • Jamie D Costabile
  • Kaarthik A Balakrishnan
  • Martin Haesemeyer

The Ohio State University Wexner Medical Center

  • Sina Schwinn
  • Martin Haesemeyer

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Ethics

Animal experimentation: Animal handling and experimental procedures were approved by the Ohio State University Institutional Animal Care and Use Committee (IACUC Protocol #: 2019A00000137 and 2019A00000137-R1).

Copyright

© 2023, Costabile et al.

This article is distributed under the terms of the Creative Commons Attribution License permitting unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,567
    views
  • 178
    downloads
  • 3
    citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Jamie D Costabile
  2. Kaarthik A Balakrishnan
  3. Sina Schwinn
  4. Martin Haesemeyer
(2023)
Model discovery to link neural activity to behavioral tasks
eLife 12:e83289.
https://doi.org/10.7554/eLife.83289

Share this article

https://doi.org/10.7554/eLife.83289

Further reading

    1. Computational and Systems Biology
    2. Developmental Biology
    Rosalio Reyes, Arthur D Lander, Marcos Nahmad
    Research Article

    Understanding the principles underlying the design of robust, yet flexible patterning systems is a key problem in developmental biology. In the Drosophila wing, Hedgehog (Hh) signaling determines patterning outputs using dynamical properties of the Hh gradient. In particular, the pattern of collier (col) is established by the steady-state Hh gradient, whereas the pattern of decapentaplegic (dpp), is established by a transient gradient of Hh known as the Hh overshoot. Here we use mathematical modeling to suggest that this dynamical interpretation of the Hh gradient results in specific robustness and precision properties. For instance, the location of the anterior border of col, which is subject to self-enhanced ligand degradation is more robustly specified than that of dpp to changes in morphogen dosage, and we provide experimental evidence of this prediction. However, the anterior border of dpp expression pattern, which is established by the overshoot gradient is much more precise to what would be expected by the steady-state gradient. Therefore, the dynamical interpretation of Hh signaling offers tradeoffs between

    1. Computational and Systems Biology
    2. Neuroscience
    Sebastian Quiroz Monnens, Casper Peters ... Bernhard Englitz
    Research Advance

    Animal behaviour alternates between stochastic exploration and goal-directed actions, which are generated by the underlying neural dynamics. Previously, we demonstrated that the compositional Restricted Boltzmann Machine (cRBM) can decompose whole-brain activity of larval zebrafish data at the neural level into a small number (∼100-200) of assemblies that can account for the stochasticity of the neural activity (van der Plas et al., eLife, 2023). Here, we advance this representation by extending to a combined stochastic-dynamical representation to account for both aspects using the recurrent temporal RBM (RTRBM) and transfer-learning based on the cRBM estimate. We demonstrate that the functional advantage of the RTRBM is captured in the temporal weights on the hidden units, representing neural assemblies, for both simulated and experimental data. Our results show that the temporal expansion outperforms the stochastic-only cRBM in terms of generalization error and achieves a more accurate representation of the moments in time. Lastly, we demonstrate that we can identify the original time-scale of assembly dynamics by estimating multiple RTRBMs at different temporal resolutions. Together, we propose that RTRBMs are a valuable tool for capturing the combined stochastic and time-predictive dynamics of large-scale data sets.