Data dimensionality reduction technique for clustering problem of metabolomics data

Abstract

In metabolomics studies, independent analyses or replicating the metabolite concentration measurements are often performed to anticipate errors. On the other hand, the size of the dataset is increasing. For clustering purposes, obtaining representative information chemically from independent analyses is needed. The objective of this study is to develop a data reduction method such that a dataset that represents chemical information is obtained. Overall a proper data reduction method would simplify the clustering of metabolite data. We propose the modified Weiszfeld algorithm (MWA) to reduce independent analyses. To obtain comprehensive results, we compare MWA with some other well-known reduction methods, including PCA, CMDS, LE, and LLE. Then reduced datasets are clustered using the fuzzy c-means (FCM) algorithm with the Tang Sun Sun (TSS) index and silhouette index as the cluster validity indices. The results show that MWA, together with PCA, present the optimal number of clusters, namely four clusters. This result aligns with the optimal number of clusters before dimensionality reduction. The present results show that MWA is robust to perform dimensionality reduction of independent analyses while maintaining chemical information on the reduced dataset. Therefore, we recommend the reliability of MWA as one of the chemometric techniques, and the present finding has enriched chemometric techniques in metabolomics studies.