PCA-n2v

This repository contains the implementation of PCA-n2v (Noise2Void), based on the package n2v.

Running the example notebook

This code should run on recent versions of Python, though the package n2v cannot be installed on 3.12+.

From a clean virtual or conda environment install the requirements:

pip install -r requirements.txt

If this fails due to missing git, install a copy of git from here. We must install n2v from GitHub in order to have certain bugfixes which are not yet released to PyPi.

If you have a CUDA-capable GPU then you can also run:

pip install -r "tensorflow[and-cuda]<2.16"

though it would be best to follow the n2v readme for configuring Tensorflow and CUDA appropriately.

From this folder launch the Jupyter notebook server:

jupyter notebook 

then load the file Example_denoising.ipynb.

Example data

The example data for the repository are available at the following Zenodo page. We use the file Alga_raw.npz placed in the ./data directory.

Note on data import

Appropriate data import is necessary before running PCA-n2v. MSI data should than be transformed into a matrix (.npy) or better a sparse matrix (.npz) before further processing. Keep in mind that the format float16 is not supported in .npz matrixes.

For IonTOF systems, we recommend exporting all the data in .txt format. This creates as many files as are selected peaks. The function iontof_to_matrix has been especially implemented for this purpose.

For PHI data, we recommend exporting the data as .tif images in a dedicated folder. Warning: this functions only if the maximum value of pixels is 256, otherwise you might need to split one peak into several peaks. The function tif_to_matrix performs the task of turning a set of .tif images into a matrix.

We have not implemented functions for other systems but we will be very happy to include the functions you build for your own data into this Github as a service to the community.