A New Optimized Clustering Applied to Sparse Spearman Correlation Estimation Network to Investigate Cognition

This paper presents a combined approach to clustering individuals based on the correlation structure of their cognitive domains. The proposed methodology includes (i) a robust Spearman correlation estimation, via permutation test and corrected for multiple-comparison, (ii) optimized clustering via Frobenius norm distance, and (iii) network visualization tools. The approach allows for the identification of subgroups characterized by distinct correlation structures, which is particularly relevant in behavioral science, medicine, and neuropsychological contexts. Unlike traditional clustering applications, this method creates clusters based on the relationships among cognitive variables rather than relying solely on individual cognitive scores. It exemplifies how advanced statistical techniques can be leveraged to explore latent constructs such as cognition, in this case.