DeepDefocusCTF

DeepDefocusCTF is a physics-constrained neural network designed to estimate defocus in both U and V directions, along with defocus angles, using the Power Spectral Density (PSD) of a micrograph in cryoEM.

Repository Structure

DeepDefocusCTF/
│── prepare_training_dataset.py   # Script to prepare training dataset
│── train_model.py               # Script to train the model
│── predict.py             # Script for inference/predictions
│── utils/                      # Utility functions
│   ├── metrics_and_model.py
│   ├── plotting.py
│   ├── processing.py
│── models/                     # Neural network model definition
│   ├── deep_defocus_model.py
│── data_generator/              # Data loading and augmentation
│   ├── data_generator.py
│── trained_models/              # Directory to store trained models
│── README.md                   # Project description and usage
│── requirements.txt             # Dependencies
│── environment.yml              # Conda environment setup
│── .gitignore                   # Ignore unnecessary files
│── LICENSE                      # License file

Installation

To set up the environment, use Conda:

conda env create -f environment.yml
conda activate deepdefocus_env

Or install dependencies using pip:

pip install -r requirements.txt

Usage

1. Prepare Training Dataset

Run the following command to prepare the dataset:

python prepareTrainingDataset.py --input <input_folder> --output <output_folder>