Network principal component analysis: a versatile tool for the investigation of multigroup and multiblock datasets

Complex data structures composed of different groups of observations and blocks of variables are increasingly collected in many domains, including metabolomics. Analysing these high-dimensional data constitutes a challenge, and the objective of this article is to present an original multivariate method capable of explicitly taking into account links between data tables when they involve the same observations and/or variables. For that purpose, an extension of standard principal component analysis called NetPCA was developed.

The proposed algorithm was illustrated as an efficient solution for addressing complex multigroup and multiblock datasets. A case study involving the analysis of metabolomic data with different annotation levels and originating from a chronic kidney disease (CKD) study was used to highlight the different aspects and the additional outputs of the method compared to standard PCA. On the one hand, the model parameters allowed an efficient evaluation of each group's influence to be performed. On the other hand, the relative relevance of each block of variables to the model provided decisive information for an objective interpretation of the different metabolic annotation levels.

NetPCA is available as a Python package with NumPy dependencies.

Supplementary data are available at Bioinformatics online.