A Machine Learning Network to Automatically Track Prairie Voles in Cohabitation: Oxytocin Receptor KO Males Reveal No Behavioral Deficits Towards their Partners Restricted; Files Only

Mahmood, Rahil (Spring 2023)

Permanent URL: https://etd.library.emory.edu/concern/etds/8336h328h?locale=en


Quantifying the nature of social interactions displayed by prairie voles in cohabitation can be a useful way to understand the neural mechanisms underlying various social behaviors. However, the current standard of behavioral analysis involves the annotation or scoring of experimental recordings, which in addition to being a time-consuming process, is open to biases and great variability across human annotators. Using supervised machine learning principles, we developed a robust network capable of automatically tracking prairie voles in cohabitation. Trained with the help of the open-source pose estimation tracking software DeepLabCut, our tracking network incorporated an in-house autoencoder and was built on a dataset of over 500,000 frames of video. Ultimately, the network was capable of increasing the efficiency of the annotation process by nearly 100%. Oxytocin has previously been shown to play a crucial role in regulating the social behaviors that govern the formation of pair bonds in prairie voles and other monogamous mammals. We used our auto-tracking network to quantitatively compare the differences in social behaviors displayed by wildtype and CRISPR/Cas-9 mediated oxytocin receptor knockout male prairie voles in cohabitation with their wildtype female conspecifics. We found no significant differences between the two genotypes with regards to the average total duration and frequencies of prosocial behaviors displayed at the time of cohabitation. These results were consistent across both automated and manual tracking processes. In concordance with previous studies, we posit that pair bond maintenance, and hence the maintenance of the social behaviors regulating pair bonds, can be achieved in the absence of the oxytocin receptor. This is likely due to the cross-talk that has been shown to exist among the oxytocin, vasopressin, and dopaminergic pathways in the brain. Ultimately, we propose that our automated tracking model can be used to significantly eliminate the subjectivity inherent in the process of behavioral annotation and eventually serve to facilitate the standardization of annotation protocols in the field of computational neuroethology.

Table of Contents

1) Introduction

2) Methods


Behavioral Recording and Manual Annotation

Training Neural Network (DLC)


Data Analysis

3) Results

Comparing Auto-Tracked and Manually-Derived Results Independent of Genotype


Social Investigation

Comparing Quantitative Differences in Social Behavior Across Genotypes

4) Discussion

5) Future Directions

6) References

7) Appendix


About this Honors Thesis

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
  • English
Research Field
Committee Chair / Thesis Advisor
Committee Members
Last modified Preview image embargoed

Primary PDF

Supplemental Files