Author: Nikita Lukianets Email: nikita.lukianets@unice.fr
- Neuroscience: Michele Studer, http://ibv.unice.fr/EN/equipe/studer.php
- Statistics: Franck Grammont, http://math.unice.fr/laboratoire/interactions/neurosciences
This set of MATLAB scripts has been developed to perform automatic neuron morphometric analysis using clustering approach, as well as a feature-by-feature comparison. Neurons are treated as objects in the multidimensional feature space. Cluster algorithm groups neuron according to similarities in their feature values. The script is based on K-means++ clustering, but unlike the original method, it doesn't require explicit specification of the cluster number as an input that makes it less biased.
Input should be organized as the datatable where objects (observations) are each placed in separate raws and features organized into a separate columns. Feature selection is done in the header of the scripts and stored as a feature_range variable. Next, features_selected matrix is transmitted as a clustering input.
- clustering_CBBP.m:
- clustering function for execution in the loop
- clustering_CBBP_standalone.m:
- standalone clustering function
- crosscorrelated.m:
- Cross-correlations of the morphology features
- distinctivef_CBBP.m:
- comparison of the populations by distinctive features
- runtestcases_clustering_CBBP.m:
- execution of the all possible test cases with selected sets of features
- regression_CBBP_standalone.m:
- machine learning using logistic regression for incomplete set of features to classification biocytin neurons based on a trained set from CBBP neurons
- plot.ly visualization https://plot.ly/matlab/
- export_fig.m https://github.com/altmany/export_fig
- combinator.m by Matt Fig http://fr.mathworks.com/matlabcentral/fileexchange/24325-combinator-combinations-and-permutations
- mwwtest.m by http://www.mathworks.com/matlabcentral/fileexchange/25830-mann-whitney-wilcoxon-test
The work became possible with the support of the SIGNALIFE PhD within University of Nice Sophia-Antipolis.