ManifoldAnalysis

This is the code for the paper "Cell Typing and Sub-typing Based on Detecting Characteristic Subspaces of Morphological Features Derived from Neuron Images" under consideration
Preprint: https://www.biorxiv.org/content/10.1101/2023.09.26.559442

This python package is used to find the optimal feature subspace. Currently it has been applied in analysis of single neuron morphologies and neuron type classification.

OptimalSubspace.py: algorithm to find the optimal feature subspace
KANNDBSCAN.py: an adaptive parameter dbscan algorithm

pre-request python packages

os copy numpy sklearn scipy matplotlib multiprocessing argparse

command line

python main.py -i input_feature_table -o outputfolder -v True/False 

i: csv format table. row: data points column: features selected.

o: target folder you want to put all the results in.

v: visualization of all figures involved in feature screening process. If False, then only final result will be generated. default is True.

Output

1. v==True: ScreeningResult Folder containing all figures related in the screening process of features
2. FeatureSubspace.npy: statistics for each dimension reduction step, e.g feature deletion order; outlier, number of clusters and scores generated; DBSCAN parameters
3. OptimalSubspaceResult.jpg: 3D plot of the final optimal feature subspace
4. scoreplot.png: score changes vs. dimension
5. clusteringlabel.txt: clustering label result for each data point

Notice

1. The number of parallel threads can be modified in "KANNDBSCAN.py", please adjust it according to the machine used