UniFORM normalization pipeline operates both on pixel-level and feature-level, enabling robust, scalable preprocessing for high-dimensional CyCIF data. This pipeline allows users to normalize CyCIF images and data across multiple samples to facilitate comparative and quantitative analysis.
- numpy
- matplotlib
- pandas
- dask
- zarr
- scikit-image
- tifffile
- pyometiff
- scikit-learn
- scipy
- scikit-fda
This repo is currently under development.
⚙️ Installation
You can install the latest version of the packages above using environment.yml
git clone https://github.com/kunlunw/uniform_cycif.git
cd uniform_cycif
conda env create -f environment.yml
conda activate cycif-normalization-env🧪 Workflow
UniFORM-calculate-histogram -> UniFORM-landmark-finetuning (optional) --> UniFORM-normalization
# Importing the Package
import uniform_cycif
# 1. Initialize the model with your AnnData object
uniform_model = uniform_cycif.UniFORM(adata=combined_adata)
# 2. Calculate histograms for specific markers
results_range, results_hist, gmm_curves = uniform_model.uniform_calculate_histogram(
adata=combined_adata, # anndata object
marker_to_plot=["CD45", "ECAD"], # Markers to process, to process all markers, do not flag this argument
rep="X" # Use raw data
)
# 4. Optional step to perform landmark finetuning
uniform_model.normalize_and_plot_distributions_plotly(
adata=combined_adata, # anndata object
histograms=results_hist, # Histograms from the previous step
gmm_curves=gmm_curves, # GMM curves for each marker and sample
markers_to_plot=["CD45", "ECAD"], # Markers to include in the plot
output_path="/path/to/output", # Directory to save output HTML files
)
# 5. Normalize data and calculate FFT-based shifts
shifts_fft_dict = uniform_model.normalize_and_plot_distributions(
adata=combined_adata, # anndata object
histograms=results_hist, # Histograms from the previous step
markers=["CD45", "ECAD"], # Markers to process
reference=references, # Reference sample for each marker
landmarks=user_landmark, # Optional fine-tuned landmarks for normalization, don't flag this for fully automatic pipeline
)
shift_in_log_pixels_dict = uniform_model.calculate_shift_in_log_pixels(
data_range=range_dict, # Data range dictionary
shifts_fft_dict=shifts_fft_dict, # Shifts from the previous step
)
# 6. Normalize data and save the output to an AnnData object
uniform_model.generate_normalized_feature(
shift_in_log_pixels_dict=shift_in_log_pixels_dict, # Shift values
reference_sample=references, # Reference samples for each marker
output_directory="/path/to/output" # Output directory for AnnData
)