Fast deep neural correspondence for tracking and identifying neurons in C. elegans using semi-synthetic training

  1. Xinwei Yu
  2. Matthew S Creamer
  3. Francesco Randi
  4. Anuj Kumar Sharma Ph.D.
  5. Scott W Linderman
  6. Andrew Michael Leifer  Is a corresponding author
  1. Princeton University, United States
  2. Stanford University, United States

Abstract

We present an automated method to track and identify neurons in C. elegans, called 'fast Deep Neural Correspondence' or fDNC, based on the transformer network architecture. The model is trained once on empirically derived semi-synthetic data and then predicts neural correspondence across held-out real animals. The same pre-trained model both tracks neurons across time and identifies corresponding neurons across individuals. Performance is evaluated against hand-annotated datasets, including NeuroPAL [1]. Using only position information, the method achieves 79.1% accuracy at tracking neurons within an individual and 64.1% accuracy at identifying neurons across individuals. Accuracy at identifying neurons across individuals is even higher (78.2%) when the model is applied to a dataset published by another group [2]. Accuracy reaches 74.7% on our dataset when using color information from NeuroPAL. Unlike previous methods, fDNC does not require straightening or transforming the animal into a canonical coordinate system. The method is fast and predicts correspondence in 10ms making it suitable for future real-time applications.

Data availability

All datasets generated as part of this work have been deposited in a public Open Science Foundation repository DOI:10.17605/OSF.IO/T7DZU.

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

Article and author information

Author details

  1. Xinwei Yu

    Department of Physics, Princeton University, Princeton, United States
    Competing interests
    The authors declare that no competing interests exist.
  2. Matthew S Creamer

    Princeton Neuroscience Institute, Princeton University, Princeton, United States
    Competing interests
    The authors declare that no competing interests exist.
  3. Francesco Randi

    Department of Physics, Princeton University, Princeton, United States
    Competing interests
    The authors declare that no competing interests exist.
  4. Anuj Kumar Sharma Ph.D.

    Department of Physics, Princeton University, Princeton, United States
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5061-9731
  5. Scott W Linderman

    Department of Statistics, Stanford University, Stanford, United States
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3878-9073
  6. Andrew Michael Leifer

    Department of Physics and Princeton Neuroscience Institute, Princeton University, Princeton, United States
    For correspondence
    leifer@princeton.edu
    Competing interests
    The authors declare that no competing interests exist.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-5362-5093

Funding

Simons Foundation (543003)

  • Andrew Michael Leifer

Simons Foundation (697092)

  • Scott W Linderman

National Science Foundation (IOS-184537)

  • Andrew Michael Leifer

National Science Foundation (PHY-1734030)

  • Andrew Michael Leifer

National Institutes of Health (R21NS101629)

  • Andrew Michael Leifer

National Institutes of Health (1R01NS113119)

  • Scott W Linderman

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

Reviewing Editor

  1. Gordon J Berman, Emory University, United States

Version history

  1. Received: January 9, 2021
  2. Accepted: July 13, 2021
  3. Accepted Manuscript published: July 14, 2021 (version 1)
  4. Version of Record published: August 16, 2021 (version 2)

Copyright

© 2021, Yu 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

  • 2,770
    views
  • 267
    downloads
  • 25
    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. Xinwei Yu
  2. Matthew S Creamer
  3. Francesco Randi
  4. Anuj Kumar Sharma Ph.D.
  5. Scott W Linderman
  6. Andrew Michael Leifer
(2021)
Fast deep neural correspondence for tracking and identifying neurons in C. elegans using semi-synthetic training
eLife 10:e66410.
https://doi.org/10.7554/eLife.66410

Share this article

https://doi.org/10.7554/eLife.66410

Further reading

    1. Neuroscience
    Jongkyun Kang, Guodong Huang ... Jie Shen
    Research Article

    Mutations in leucine-rich repeat kinase 2 (LRRK2) are the most common genetic cause of Parkinson’s disease (PD). However, whether LRRK2 mutations cause PD and degeneration of dopaminergic (DA) neurons via a toxic gain-of-function or a loss-of-function mechanism is unresolved and has pivotal implications for LRRK2-based PD therapies. In this study, we investigate whether Lrrk2 and its functional homolog Lrrk1 play a cell-intrinsic role in DA neuron survival through the development of DA neuron-specific Lrrk conditional double knockout (cDKO) mice. Unlike Lrrk germline DKO mice, DA neuron-restricted Lrrk cDKO mice exhibit normal mortality but develop age-dependent loss of DA neurons, as shown by the progressive reduction of DA neurons in the substantia nigra pars compacta (SNpc) at the ages of 20 and 24 months. Moreover, DA neurodegeneration is accompanied with increases in apoptosis and elevated microgliosis in the SNpc as well as decreases in DA terminals in the striatum, and is preceded by impaired motor coordination. Taken together, these findings provide the unequivocal evidence for the cell-intrinsic requirement of LRRK in DA neurons and raise the possibility that LRRK2 mutations may impair its protection of DA neurons, leading to DA neurodegeneration in PD.

    1. Computational and Systems Biology
    2. Neuroscience
    Mu Qiao
    Tools and Resources

    Understanding how different neuronal types connect and communicate is critical to interpreting brain function and behavior. However, it has remained a formidable challenge to decipher the genetic underpinnings that dictate the specific connections formed between neuronal types. To address this, we propose a novel bilinear modeling approach that leverages the architecture similar to that of recommendation systems. Our model transforms the gene expressions of presynaptic and postsynaptic neuronal types, obtained from single-cell transcriptomics, into a covariance matrix. The objective is to construct this covariance matrix that closely mirrors a connectivity matrix, derived from connectomic data, reflecting the known anatomical connections between these neuronal types. When tested on a dataset of Caenorhabditis elegans, our model achieved a performance comparable to, if slightly better than, the previously proposed spatial connectome model (SCM) in reconstructing electrical synaptic connectivity based on gene expressions. Through a comparative analysis, our model not only captured all genetic interactions identified by the SCM but also inferred additional ones. Applied to a mouse retinal neuronal dataset, the bilinear model successfully recapitulated recognized connectivity motifs between bipolar cells and retinal ganglion cells, and provided interpretable insights into genetic interactions shaping the connectivity. Specifically, it identified unique genetic signatures associated with different connectivity motifs, including genes important to cell-cell adhesion and synapse formation, highlighting their role in orchestrating specific synaptic connections between these neurons. Our work establishes an innovative computational strategy for decoding the genetic programming of neuronal type connectivity. It not only sets a new benchmark for single-cell transcriptomic analysis of synaptic connections but also paves the way for mechanistic studies of neural circuit assembly and genetic manipulation of circuit wiring.