cellstate-pred 0.1.0

Predict cell state (hypusine marker) based on deep learning morphological features

cellstate-pred

A Python package for cell segmentation from drug assays and inference with pretrained deep learning models.

Overview

cellstate-pred is designed to analyze microscopy images from drug screening assays by automatically segmenting individual cells and extracting morphological features using pretrained deep learning models. The package provides tools for:

• Cell Segmentation: Automated segmentation of cells from microscopy images using advanced image processing techniques
• Feature Extraction: Extraction of morphological features from segmented cells using pretrained deep learning models
• Drug Assay Analysis: Specialized workflows for analyzing cellular responses in drug screening experiments

Key Features

• Support for Opera microscope image formats
• Otsu thresholding and watershed segmentation for robust cell detection
• Integration with Hugging Face transformers for feature extraction
• Batch processing capabilities for high-throughput analysis
• Flexible data loading with PyTorch DataLoader integration

Installation

Requirements

• Python 3.10+
• Dependencies as specified in pyproject.toml

Install from source

git clone https://github.com/rendeirolab/cellstate-pred.git
cd cellstate-pred
pip install -e .

Quick Start

from cellstate_pred.image_utils import OperaImage
from cellstate_pred._inference import inference
from pathlib import Path

# Load and segment cells from a microscopy image
image = OperaImage()
image.load_image("path/to/your/image.tiff")
image.segment_cells(min_distance=7)

# Extract features using a pretrained model
inference(
    batch_size=32,
    image=image,
    model_name="microsoft/resnet-50",
    output_dir=Path("./features"),
    num_workers=4
)

Documentation

For detailed usage examples and API documentation, see the notebooks in the scripts/ directory.

Contributing

This repository was created with a cookiecutter template, version 0.4.1dev.

License

[Add your license information here]