Research And Implementation Of Active Ingredient Group Prediction Method For Herb Pair And Classical Chinese Prescription Based On Heterogeneous Network

The classical Chinese prescriptions are an essential part of traditional Chinese medicine(TCM),and the herb pairs are the combination of two kinds of herbs that often appear in the prescriptions.TCM acts on different targets through active ingredient group(AIG).The prediction of active ingredient group can help to clarify the pharmacodynamic action of TCM.Currently,researches on the AIG prediction are mainly based on the target interaction network.The key targets can be screened through the target interaction network,and the ingredients corresponding to the key targets are taken as AIG.However,the prediction of AIG through key targets only uses the the target level information,ignoring the pathway and biological process information.And this kind of method fails to conform to the characteristics of multi-ingredient,multi-target and multi-pathway synergistic mechanism.In this thesis,the heterogeneous network is used to model the influence of ingredients on the target,related pathways and biological processes,and two methods for the AIG prediction of classical prescriptions are proposed based on the heterogeneous network.The main research work of this thesis is as follows:(1)A method based on the semi-supervised heterogeneous networks representation learning for AIG prediction(HRTCM)is proposed.HRTCM integrates the TCM and western medicine by constructing a heterogeneous network.The AIG is obtained by heterogeneous network representation learning.For some specific diseases,western medicine molecules have relatively complete disease information in the database.HRTCM uses this additional information to construct labels,and applys them in the process of network representation learning to improve the effect of network representation learning.The experimental results show that this method can effectively predict the AIG.(2)A method based on the self-supervised heterogeneous networks representation learning for AIG prediction(SELFTCM)is proposed.Most pharmacological studies of TCM and Western medicine may have not been covered for many complex diseases,and the labeled data is not enough for semi-supervised network representation learning.Therefore,Self-supervised representation learning is applied to reduce the dependence on labeled data,and its effectiveness is verified by experiments.(3)Based on the above research,this thesis designs a tool for predicting the active ingredient group of classic prescriptions,which can be used to predict the active ingredient group of prescriptions.