Abstract:

Objective    To clarify internal relation and quality relationship between the material composition and the Chinese herbal medicine property (CHMP), to establish the Bayesian network model among ‘composition’ and ‘CHMP’， and to infer and restore the CHMP by means of network.  Methods    According to the reductionism mode,  high-performance liquid chromatography (HPLC) technology was combined with the Partial Least Squares Discriminant Analysis (PLS-DA) to identify the CHMPmarkers. Then based on the system theory, the Bayesian network model of CHMP-markers and CHMP was structured.  Results    37 CHMP-markers were selected by the PLS-DA model to build the Bayesian network. It had high sensitivity, specificity (AUC=0.98) and discriminative power. Using this network to recognize the CHMP, the discriminative accuracy of the training set was 93.88% and the predictive accuracy of the testing set was 100%.  Conclusion     The Bayesian network has apparent modular construction，  and it could explain the compatibility and pathways among the composition and the CHMP.

Key words: Bayesian network； Chinese herbal medicine property； High-performance liquid chromatography