README

This repository contains jupyter notebooks and code developed to classify 2D high-content screening images based on texture and other features into 3 classes.

Data

• Coacervates data from 96-well plate, where each well has different concentrations of 2 proteins of interest
• The different concentrations lead to three categories of images: droplets, aggregates, empty.
  • Empty images look just like background
  • Droplet images show transparent droplets (just circles) of various sizes
  • Aggregates - textured image, clumps of protein, no clear structure

Analysis goal

Classify images automatically into one of the 3 categories.

Final workflow

• Based on python using libraries scikit-learn, mahotas, scikit-image, tifffile.
• Workflow
  • functions.py: contains all the helper functions
  • 1_read_images.ipynb: read and save images for training - on laptop
  • 2_train_save_classifier.ipynb: train classifier - on cluster
  • 3_do_prediction.ipynb: apply model and get predictions - on laptop
  • README_Python_Installation.pdf
  • imageclassification.yml: env file
• Additionally
  • explorations: contains additional notebooks from various explorations using deeplearning, cnn and transfer-learning
  • 1A_read_images_augment.ipynb for data augmentation part

This project was original done in collaboration with Rudrarup Bose, PhD at Tang lab (Nov 2021, MPI-CBG, Dresden). The code is published with permission.