Sequential decision fusion pipeline for the high-throughput species recognition of medicinal caterpillar fungus by using ATR-FTIR

Medicinal caterpillar fungus is one of the important animal-derived traditional Chinese medicines (TCMs) highly valued for its tonic and medicinal properties. Due to the worldwide importance and public interest concerning to the quality and safety of TCMs, and the interest disputes arising from the identification of some extremely highpriced fungi varieties, there is an increasing demand for accurate species authentication of caterpillar fungus on the base of various techniques. This study used attenuated total reflectance Fourier transform infrared (FTIR) spectroscopy (ATR-FTIR) to identify 10 species of caterpillar fungus and its fungal products, as well as to differentiate rare and high prices Asian fungus Ophiocordyceps sinensis (OSH) from other fungal species. To this goal, 1206 ATR-FTIR spectra of all species of samples were collected, partial least square discriminant analysis (PLS-DA) was used to construct multi-class and two-class model for differentiating 10 caterpillar fungal species and identifying OSH from other counterfeits. Subsequently, data driven soft independent modeling of class analogy (DD-SIMCA) was used to formulate species-specific models for recognize each target class. However, PLS-DA model cannot be used for species authentication since all unseen samples in the external test set were wrongly assigned to known categories. DD-SIMCA models have good performance to recognizing each target species, but the identification of multiple species requires complicated analytical steps. Therefore, this study proposed a sequential decision fusion pipeline, which combines the advantages of discriminant analysis and class-modelling for differentiating 10 caterpillar fungal species and authenticating OSH. Almost all the evaluations showed good results with sensitivity and specificity of each class greater than 98%, which demonstrated promising application of the pipeline for the high-throughput identification of caterpillar fungal species, as well as other animal-derived TCMs.