Redundancy analysis for characterizing the correlation between groups of variables - Applied to molecular descriptors

摘要: Abstract Redundancy analysis (RA) estimates the extent of linear relationships between blocks variables that are given for a set objects (samples). RA has only rarely been used in chemometrics. Basic principles and limits discussed, is briefly compared with canonical correlation (CCA) partial least-squares (PLS2) regression. The significance redundancy index estimated by permutation tests. For PLS2, an determining similarity variable can be derived equivalent to measure correlation, CMC. applied 3708 molecular descriptors (created software Dragon) 6458 chemical structures (AMES database). 27 descriptor groups characterized their indices, which allow comparison multivariate information content. results guide selection most different groups, perform better discrimination task (classification mutagenicity) than entire groups.