Correlation-Comparison Analysis as a new way of Data-Mining: Application to Neural Data

Ivan GRBATINIĆ
Bojana KRSTONOŠIĆ
Dragana SREBRO
Nemanja PURIĆ
Marija DUBAK
Vladan DUŠANIĆ
Vladimir KOSTIĆ
Nebojša MILOŠEVIĆ

Keywords

Correlation-Comparison Analysis, Inter-Sample Correlative p-Comparisons, Correlation Mismatching

Abstract

This paper tends to present a way of multidimensional data-mining termed correlation-comparison analysis (CCA). It was applied to neural data to show its utility in neuron-classification problem. The CCA represents a semi quantitative way of inter-sample comparisons. The methodology comprises the generation of inter-parametric correlation and alpha-error matrices. The main step is p-comparison for the same parametric pair defined between the two samples. This comparison has a semi-quantitative binary character that does not involve issues like false discovery rate (FDR) in multiple comparisons. As a result, the outcomes obtained are: 1) correlation match, 2) correlation mismatch of the first kind, the main type of the correlation mismatch, 3) correlation mismatch of the second kind, the strongest one but very rarely observed in biological systems and obtained on a very small number of parameters. The correlation mismatch of the first kind is the target mismatch, i.e. the mismatch of tracing interest and represents the reason why the study itself is performed. CCA application led to effective neuromorphofunctional classification of caudate interneurons into appropriate clusters and their feature-based description. CCA analysis is a very useful multidimensional bi-sampled classification tool that can be very useful for similar samples to explain their differences.