Behavioral Analysis

Purpose

• Pose estimation with DeepLabCut (DLC)
• Behavior labeling with Mouse Action Recognition System (MARS) and BENTO
• Behavioral classification with Simple Behavioral Analysis (SimBA)

Workflow overview

• DLC → extract frames → label body parts in napari → train DLC network → output estimated pose coordinates (.csv/.h5) and videos (.mp4)
  • Follow this tutorial walkthrough to create a DLC project
  • Annotating 100-200 frames from 10 videos would already give you a pretty good result
• MARS/BENTO → manually label frames → output labeled behavioral bouts with start/end time stamps (.annot)
  • Have clear inclusion/exclusion criteria for each behavior
  • Label in an actor-agnostic way
• SimBA → create project → import DLC pose estimation (.csv) and MARS behavior labels (.annot) train behavioral classifiers → analyze videos
  
  • Follow this tutorial walkthrough to create a SimBA project
  • SimBA trains a separate random forest classifier for each behavior
  • You can annotate videos using the SimBA GUI, but I do it using Caltech's MARS/BENTO
  • Import MARS annotations to SimBA

Pre-requisites

1. Install Python 3.x version
2. Install Miniconda3
3. Ideally work on a workstation with GPU

Install DLC

1. Open Anaconda Prompt
2. conda create --name deeplabcut python=3.12
3. conda activate deeplabcut
4. pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128 # analysis workstation 1
5. pip install deeplabcut[gui]
6. python -c "import torch; print(torch.cuda.is_available())" # should print out "True"
7. python -m deeplabcut # launch dlc

Install BENTO

1. Download the zip file from my forked directory here
2. Open Anaconda Prompt
3. cd path_to_bento_folder
4. conda env create -f bento.yml
5. conda activate bento
6. pip install colour-science==0.4.6 --no-deps
7. pip install colour-demosaicing==0.2.6 --no-deps
8. python src/bento.py # launch bento

Install SimBA

1. Open Anaconda Prompt
2. conda create --name simba python=3.6
3. conda activate simba
4. pip install simba-uw-tf-dev
5. simba # launch simba

Citations

DeepLabCut
Mathis, Alexander, et al. "DeepLabCut: markerless pose estimation of user-defined body parts with deep learning." Nature neuroscience 21.9 (2018): 1281-1289.

MARS/BENTO
Segalin, Cristina, et al. "The Mouse Action Recognition System (MARS) software pipeline for automated analysis of social behaviors in mice." Elife 10 (2021): e63720.

SimBA
Goodwin, Nastacia L., et al. "Simple Behavioral Analysis (SimBA) as a platform for explainable machine learning in behavioral neuroscience." Nature neuroscience 27.7 (2024): 1411-1424.