temsis/Cell-Painting-VAE

TIFF Image Processing and Analysis Pipeline

A Python library for processing multi-channel TIFF images, creating tiles, and generating embeddings using a VAE model.

Project Structure

Core Files

• main.py - Main entry point that orchestrates the image processing pipeline:
  • Processes TIFF files from input directory
  • Creates tiles from composite images
  • Generates embeddings using VAE model

Source Code directory (src/)

Image Processing (src/process_images/)

• preprocess.py - Handles TIFF file preprocessing:
  • Loads multi-channel TIFF images
  • Normalizes and resizes images
  • Creates composite arrays
• create_tiles.py - Creates fixed-size tiles from processed images:
  • Generates 128x128 pixel tiles
  • Saves tiles as parquet files

Deep Learning (src/generate/)

• generate.py - Handles VAE model inference:
  • Loads pretrained VAE model
  • Generates embeddings from image tiles
• weights/model.py - VAE model architecture definition

Data Loading (src/load_dataset/)

• dataset.py - PyTorch dataset implementation for tile loading

Usage

Basic usage to process images and generate embeddings:

python main.py -d sample_images

The pipeline will:

1. Load and process TIFF files from the input directory
2. Create tiles in the tiles_output directory
3. Generate embeddings using the VAE model

Output

• Processed tiles are saved in tiles_output/
• Generated embeddings are saved as numpy arrays
• Processing logs are written to log_error.log

Credit to: Dr Mark Pinches Dr Nina Vyas Temitope Femi Temiola