Non-Negative Matrix Factorization-Based SIMCA Method to Classify Traditional Chinese Medicine by HPLC Fingerprints

Non-negative matrix factorization (NMF) is a novel technique that decomposes multivariate data into a smaller number of basis vectors and encodings under non-negative constraints. SIMCA (Soft independent modeling by class analogy) is a statistical method for supervised classification of data proposed by S. Wold. In this paper, each classification model is built separately using NMF algorithm rather than principal component analysis (PCA), and each group is described by NMF basis vectors. Then new observations are projected into each NMF model. The residual distances from the new objects to each NMF model are calculated and F-test is used to predict whether a new object belongs to some specific group. The capabilities of this novel SIMCA method, named as NMF-based SIMCA, for discriminating two origins of traditional Chinese medicine, Xiangdan injection, are studied and compared with classical PCA-based SIMCA method. The results show that NMF can more clearly separate samples than PCA. Better classification accuracy (over 90.0%) is achieved. Factors selection and robust evaluation of classification are studied. It is indicated that in some cases NMF-based SIMCA outperformed PCA-based SIMCA method. NMF and NMF-based SIMCA show promising features and can be practiced in pattern recognition field for discrimination purposes.