RAINBAR: A Live-Cell CRISPR Screening Platform

Description

Preliminary code for decoding RainBar, will be updated in the future. NFkB Assays code will be uploaded soon.

Table of Contents

• Description
• Installation
• Usage
• Features
• License
• Contact

Installation

• Install conda
• Create RainBar virtual environment with: conda create --name RainBar, then conda activate RainBar to activate the venv
• pip/conda install the following packages: cellpose numpy scikit-image matplotlib scipy pandas scikit-learn keras tensorflow
• also install or ensure you can import: os, re, gc, multiprocessing, datetime

Usage

• Open _op_image_analysis.py and set these variables for your image analysis:
  • image_dir -- point the image analysis script to the directory containing the images you wish to segment and quantify.
  • experiment_name -- use this variable to give a descriptive name to all output files.
  • custom_model_path[s] -- point the image analysis script to your cell and nucleus segmentation models.
  • additionally, you can set which intermediate files (masks, annotations) you would like to save and their output destinations by scrolling through the script and modifying these lines (io.imsave).
  • num_workers -- you may also adjust parallelization using this variable near the bottom of the script.
• Open _op_rb64_analysis_mlp.py and set these variables for your data analysis:
  • Cell 1 (Singles Data Import and Pre-Process):
    • rb64_subfilter (main data) -- point the data analysis script to the intensity measurements you generated in the previous step.
    • experiment_name -- use this variable to give a descriptive name to all output files.
    • rb64_subfilter_ids -- create a .csv file with the identities of your single construct wells and provide this path here.
    • qc_rb64_subfilter -- modify circularity, area, or others filters as necessary for your analysis.
    • ch_to_drop -- choose which channels you want to (optionally) drop from the dataset.
    • optionally save the intermediate output.
  • Cell 2 (Ratio Dataframe Generation):
    • if not saving the intermediate output from the previous step, use the variable name for your main data.
    • optionally save the intermediate output.
  • Cell 3 (Training MLP Model on Singles Wells):
    • if not saving the intermediate output from the previous step, use the variable name for your ratio data.
    • this cell will give you information to evaluate your model on 30% withheld test data.
  • Cell 4 (Pooled Data Import, Pre-Process, and Ratio Dataframe Generation):
    • analyze your pooled wells in the same fashion as your singles using the _op_image_analysis.py script.
    • pools_name -- use this variable to give a descriptive name to all output files.
    • rb64_subfilter_pools -- point the data analysis script to the pools intensity measurements you generated in the previous step.
    • optionally drop low quality wells as well.
    • optionally save the intermediate output.
  • Cell 5 (Pooled Data Predictions):
    • if not saving the intermediate output from the previous step, use the variable name for your pools data.
  • Cell 6 (T-SNE Visualization):
    • well_number -- set the pools well number you wish to visualize over grayed-out test data.
    • produce a visualization from the penultimate layer of the MLP model with the first plot as a T-SNE of the 30% withheld test data predictions, and the second plot as the pooled well of your choice overlaid on grayed-out test data