This repository is used to generate the figures published in the bioRxiv paper, Ascending neurons convey behavioral state to integrative sensory and action selection centers in the brain, found here https://www.biorxiv.org/content/10.1101/2022.02.09.479566v1

This pipeline was developed to be run on Ubuntu 18.04.3 LTS (GNU/Linux 4.15.0-76-generic x86_64).

• Installation
• Download preprocessed experimental data
• Reproducing the figures

To be able to run all of the scripts, Python and R must be installed.

To install the environment with specific versions of requried Python packages, please install Anaconda first if you haven't already.

Install Anaconda --> https://www.anaconda.com/

Clone repository:

$ cd ../to/where/you/want/
$ git clone https://github.com/NeLy-EPFL/Ascending_neuron_screen_analysis_pipeline.git

Change the directory to your ../Ascending_neuron_screen_analysis_pipeline folder where AN_env_public.yml is located:

$ cd ../../Ascending_neuron_screen_analysis_pipeline

Install the AN Python environment with packages in AN_env_public.yml:

$ conda env create -f AN_env_public.yml

$ git clone https://github.com/DeepLabCut/DeepLabCut.git

1. Go to the folder of your clone DeepLabCut:

$ cd DeepLabCut/

2. Change to the commit number of v2.1.8.1

$ git checkout f2f69166bb11e37fa9784273191468362f9c699d

3.You can check the update record:

$ git log

4. Finally, you can check your clone DeepLabCut folder which should have the contents in the github reository of your desired version.

$ cd /DeepLabCut/conda-environments

Create the DLC-CPU environment and install the CPU version of DeepLabCut:

$ conda env create -f DLC-CPU.yaml

Now DeepLabcut should be installed and ready to use.

We use R 3.6.1 for some of the anaylsis pipeline. You can find more information about R here.

Install R version 3.6.1 (2019-07-05) and the r-base-dev package to compile R packages:

$ sudo apt-get update 
$ sudo apt-get install r-base-dev=3.6.1

Launch R to install associated packages by entering R in the terminal:

$ R

Then install the R packages for this manuscript:

> install.packages("ggpmisc")
> install.packages("ggplot2")
> install.packages("tidyverse")
> install.packages("tidyr")
> install.packages("dplyr")
> install.packages("reshape2")

Optional

Now the AN conda environment, DeepLabCut, and R should be installed. If you want to check Python scripts seperately, remember to activate the AN environment manually before running each .py script:

$ source activate AN
$ python name.py

If you want to use DeepLabCut independently, remember to activate the DLC-CPU environment manually before running the script:

$ source activate DLC-CPU
$ python name.py

If source doesn't work, try using conda instead.

There is no environment for R. To run the R script in the terminal independently:

$ Rscript name.R

Download the data from Harvard dataverse and make sure that the location of the data inside each numbered folder is organized as it is below:

Ascending_neuron_screen_analysis_pipeline
├── 00_behavior_data_preprocess
│   ├── CO2puff_regressors
│   ├── df3dResults_ballRot_captureMeta
│   └── PE_regressors
├── 01_behavior_annotation_for_behavior_classifier
│   └── behaviour_annotations.csv
├── 03_general_2P_exp
│   ├── MAN
│   ├── R15E08
│   ├── R30A08
.   .   ...
.   .   ...
.   .   ...
│   └── SS52147
├── 04_mcfo_traced_singleAN_exp
│   ├── MCFO
│   ├── VFB
│   └── VNC
├── 05_offBall_onBall_2P_exp
│   ├── SS38631
│   └── SS51017
├── 06_SS36112-air_vs_co2_puff
│   └── SS36112
├── 07_SS31232-PE_exp
│   └── SS31232
├── DLC_model_for_labelling_proboscis
│   ├── config.yaml
│   ├── dlc-models
│   ├── evaluation-results
│   ├── labeled-data
│   ├── training-datasets
│   └── videos

Note: Before running the following scripts, make sure that the Python environment and R packages are installed (see the Installation guide above). To start running the analysis pipeline, execute the following .run script from the folder Ascending_neuron_screen_analysis_pipeline/scripts_for_public/:

$ cd ../../Ascending_neuron_screen_analysis_pipeline/scripts_for_public/

Execute the script by typing ./ before the script fileame. For example:

$ ./_00-0-plot_FigS1-exemplar_DLC_proboscisLabel_PEevt_detection.run

The intermediate results must be generated in the order shown below to generate plots. Please check the diagram in each session that depicts the workflow from a particular input dataset to specific ouput plots.

• Figures from dataset 00_:

  • FigS1: Semi-automated detection of proboscis extensions

• Figures from dataset 00_, 01_ and 03_:

  • Fig4a, Fig5a, Fig6a, Fig7b, Fig8a, Fig9a, Fig10a: Prediction of neural activity
  • Fig7c: Prediction of neural activity differences between left and right neurons due to turning
  • Fig10b_right, Fig10c: Prediction of neural activity from convolved PEs
  • FigS10: Behavior classifier confusion matrix
  • Fig2, FigS4 top: GLM of joint angle, leg, leg pair, and behavior for predicitng neural activity
  • Fig7a: Explained variance matrix of turning for predicting neural activity

• Figures-from-dataset-03_:

  • FigS2: Joint angle and behavior covariance matrix
  • FigS3: Behavioral event-triggered average neural activity
  • Fig4b, Fig5b, Fig6b, Fig7b, Fig8b, Fig9b: Behavioral event-triggered average neural activity
  • Fig6c, Fig7e: Neural activity related ball rotations
  • FigS5: ANs from SS36112 respond to puff stimulation rather than backward walking

• Figures-from-dataset-04_:

  • Fig3, FigS4: Large-scale anatomical quantification of ascending neuron projections to the brain and ventral nerve cord

• Figures-from-dataset-05_:

  • FigS7: ANs that are active for movements off of the spherical treadmill

• Figures-from-dataset-06_:

  • FigS6: ANs from SS36112 respond to puffs of either air or CO2

• Figures-from-dataset-07_:

  • Fig10b_left: Summary of neural activity for each PE during PE trains

$ ./_00-0-plot_FigS1-exemplar_DLC_proboscisLabel_PEevt_detection.run

Outputs: /output/FigS1-exemplar_PEevt_detection/

The intermediate file must first be generated:

$ ./_00_01_03-1_train_behavior_classifier.run
$ ./_00_01_03-2_predict_behavior_and_sync_beh_w_DFF.run

Outputs: /output/Fig2_S4-GLM_jangles_legs_beh_DFF/

Note: The intermediate resuls from the behavior classifier and synchronized dataframes can now be substituted with the those used in this work from /02_Fig2_output_of_published_version if you want to skip _00_01_03-1_train_behavior_classifier.run and _00_01_03-2_predict_behavior_and_sync_beh_w_DFF.run.

Once the intermediate results are generated, the following plots can be made using the following bash scripts:

$ ./_00_01_03-2-plot_Fig4a_5a_6a_7b_8a_9a_10a-Plot_prediction_rawDFF_traces.run

Outputs: /output/Fig4a5a6a7b8a9a10a-representativeDFF_traces/

$ ./_00_01_03-2-plot_Fig7c-Plot_dDFF_traces_fit_turning.run

Outputs: /output/Fig7a_7c-turning/

$ ./_00_01_03-2-plot_Fig10b_right_10c-Plot_convPE.run

Outputs: /output/Fig10a_10c-PE_analysis/

This figure number is corresponding to the version of Nature Neuroscience.

$ FigED4_ED5-overview_matrices_jangle_Rest.R

$ FigED4_ED5-Plot_GLM_rest.py

Outputs: /output/FigED4_ED5-overview_matrices_jangle_Rest/

$ ./_00_01_03-2-plot_FigS10-beh_jangle_confusionMat.run

Outputs: /output/FigS10-confusionMat_beh_classifier/

To plot and visualize the GLM matrix, the following intermediate GLM results must first be generated:

$ ./_00_01_03-3_glm_of_beh_leg_legPair_jangle_DFF.run

Overall significance of our models using an F-statistic to reject the null hypothesis that the model does not perform better than an intercept-only model. Note: GLM intermediate results used in this paper can be found in four /02_Fig2_output_of_published_version/overview_... folders. Use them if you want to skip this GLM step

Then, plot with:

$ ./_00_01_03-3-plot_Fig2abcd_S4-Plot_glmmat_of_beh_leg_legPair_jangle_DFF.run

Outputs: /output/Fig2_S4-GLM_jangles_legs_beh_DFF/

To visualize and plot the turning R-squared matrix, the following intermediate results must first be generated:

$ ./_00_01_03-4_turn_mat_analysis.run

Overall significance of our models using an F-statistic to reject the null hypothesis that the model does not perform better than an intercept-only model. Then, plot using:

$ ./_00_01_03-4-plot_Fig7a-Plot_turn_analysis_matrix.run

Outputs: /output/Fig7a_7c-turning/

$ ./_03-0-plot_FigS2-jangle_beh_covariance.run

Outputs: /output/FigS2-jangle_beh_covariance/

First the intermediate results of averaging neural activity for each behavioral epoch need to be generated:

$ ./_03-1_prep_DFF_beh_mat.run

Then, the results can be visualized using the following scripts:

$ ./_03-2-plot_FigS3-Plot_DFFmat.run

Outputs: /output/FigS3-DFF_mat/

$ ./_03-2-plot_Fig4b_5b_6b_7d_8b_9b-Plot_BehEvt_avgDFF.run

Outputs: /output/Fig4b_5b_6b_7e_8b_9b-Beh_avgDFF/

Intermediate results of neural event detection first need to be generated:

$ ./_03-3_prep_DFFevt_analysis.run

Then, the results can be visualized using the following scripts:

$ ./_03-3-plot_Fig6c_7d-Plot_DFFevt_analysis.run

Outputs: /output/Fig6c_7d-DFF_event_corresponding_ballRot/

Intermediate results of puff and backward walking event detection first need to be generated:

$ ./_03-4_prep_SS36112_independentBW_vs_CO2puffBW_analysis.run

Then, the results can be visualized using the following scripts:

$ ./_03-4-plot-FigS5-SS36112_independentBW_vs_CO2puffBW_analysis.run

Outputs: /output/FigS5-CO2puff_BW_analysis_SS36112/

Intermediate results of puff and backward walking event detection first need to be generated:

$ ./_04-1_prep_singleAN_innervation_mat.run

Then, the results can be visualized using the following scripts:

$ ./_04-1-plot_Fig3_S4-Plot_singleAN_innervation_mat.run

Outputs: /output/Fig3_S4-single_AN_innervation_mat/

$ python FigS8-morphoLateralization-activitySymmetry.py

Outputs: /output/FigS8-morphoSymmetry-activitySymmetry/

To avoid the false-positive result of statistical comparison due to over-large population, the data were bootstrapped into 10 groups with sample size 30 and the majority of Mann-Whitney U-test results determines whether to reject null hypothesis.

$ ./_05-0-plot_FigS7-offBall_onBall_comparison.run

Outputs: /output/FigS7-offballActive_ANs/

To avoid the false-positive result of statistical comparison due to over-large population, the data were bootstrapped into 10 groups with sample size 30 and the majority of Mann-Whitney U-test results determines whether to reject null hypothesis.

$ ./_06-0-plot_FigS6-Plot_air_vs_CO2puff.run

Outputs: /output/FigS6-air_vs_CO2_puff_comparison/

$ ./_07-0-plot_Fig10b_left-DFF_per_PE.run

Outputs: /output/Fig10b_left-PE_dynamic/