| Wolfiporia cocos is a genuine medicinal fungus with the homology of medicine and food,and it is used as medicine with dried sclerotia.Its inner part（Poria）and epidermis（Poriae cutis）are used as two traditional Chinese medicines.Due to the complex and diverse factors affecting the quality of W.cocos from different regions and parts,as well as a series of problems such as unbalanced supply and demand in the development of resources,and lack of systematic resource evaluation,the quality of W.cocos is uneven,which seriously hinders the development of W.cocos.With the advent of“Traditional Chinese medicine+Internet”and the characteristics of multiple components,multiple targets,and integrity of traditional Chinese medicine,traditional evaluation methods cannot meet the rapid analysis of multi-source data.It is urgent to establish a robust,efficient,and comprehensive resource evaluation method in order to ensure the consistency of its（raw material）quality.Therefore,it is planned to comprehensively collect the samples of W.cocos to obtain multi-source spectral information with overall characterization,and combine modern analytical techniques such as chemometrics,two-dimensional correlation spectral（2DCOS）images,and deep learning to conduct multi-level and high-level analysis of W.cocos from different regions and parts for efficiency and comprehensiveness identification research,and preliminary screen differential compounds between different parts of W.cocos combined with network pharmacology.Methods:1.Based on the multi-information fusion strategies of Fourier transform near-infrared spectroscopy（FT-NIR）and mid-infrared spectroscopy（FT-MIR）,the partial least squares discriminant analysis（PLS-DA）,support vector machine（SVM）and extreme learning machines（ELM）models were established to conduct overall identification analysis and comparison of W.cocos in different regions.A total of 3528synchronous,asynchronous,and integrated 2DCOS images were obtained,combined with residual convolutional neural network（Res Net）and PLS-DA model,which were used for the identification and analysis of different parts and regions of W.cocos.Three types of2DCOS images were compared and analyzed,including the original spectrum and the full-band and eigen-band(8900-6850 cm-1,6300-5150 cm-1,and 4450-4050 cm-1)with pre-treated by the second derivative（SD）.3.Screening potential quality markers in different parts of W.cocos based on the measurability and effectiveness of chemical fingerprinting,machine learning and network pharmacology.Results:1.The spectral data can be optimized through pretreatment,and the statistical analysis method can cluster the samples of cloud poria in different regions and locations.Compared with FT-MIR,ELM models of single,low-level data or intermediate data fusion（except continuous projection algorithms）FT-NIR spectra have advantages and can effectively and quickly identify cloud poria of different regions.2.Synchronous SD 2DCOS spectral images are more suitable for the identification and analysis of small and complex hybrid systems with small characteristic bands of Poria and Poriae cutis.Both PLS-DA and Res Net models can successfully identify different regions of Poria and Poriae cutis with 100%accuracy.3.From the perspective of effectiveness and measurability,it is obtained that dehydrotrametenolic acid,poricoic acid A,and pachymic acid can be used as preliminary potential quality markers of Poria and Poriae cutis.This method provides a new strategy for finding potential quality markers of Poria and Poriae cutis,and provides an important reference for explaining the clinical application of other medicinal plants and the multi-pathway utilization of non-traditional medicinal sites,which is conducive to promoting regional economic development and promoting the green and sustainable development of medicinal plant resources. |
