mzClustering

This repository contains R implementation of MSI data preprocessing, visualization, and spatial centroid segmentation using package Cardinal and Python implementation of $m/z$ clustering using convolutional neural network based deep clustering method.

$m/z$ clustering

from mz_clustering import *

Self-supervised training of the clustering network

clusterNet = clustering (spec_path, label_path, 
                        num_cluster = 7, height = 40, width = 40, KNN = True, k = 10)

spec_path : path to the .csv file of MSI spectra data

label_path: path to the cluster labels of $m/z$ if available. Default is None. It is only used to calculate the clustering accuracy.

num_clusters: specifies the number of $m/z$ clusters.

height and width: specify the height and width of ion images.

KNN: True if including KNN for pseudo labeling. Default is True.

k: k in KNN if KNN is used.

cae, CLUST = clusterNet.train(use_gpu = True)

use_gpu specifies whether to use gpu, default is True.

Running inference on all $m/z$

pred_label = clusterNet.inference(cae, CLUST)

t-SNE visualization of $m/z$ and predicted cluster memberships

clusterNet.tsne_viz(pred_label)