Herb-disease association prediction model based on network consistency projection

Abstract

A growing number of biological and clinical reports indicate the usefulness of herbs in the treatment of complex human diseases, giving an essential supplement for modern medicine. Similar to drugs, the use of experimental validation to identify related diseases of given herbs is both expensive and time-consuming. Such validation is even more difficult because each herb always contains several components. It is alternative to design computational models to predict herb-disease associations (HDAs). Nevertheless, only a few computational models have been developed for HDA prediction. In this study, we make full use of several properties of herbs and diseases, which are collected in a public database HERB, to design a model named HDAPM-NCP for predicting HDAs. Based on these properties, six herb kernels and five disease kernels are constructed, which are further fused into one unified herb kernel and one disease kernel. These kernels and herb-disease adjacency matrix are fed into network consistency projection to quantify the strength of herb-disease pairs. The cross-validation results show the high performance of HDAPM-NCP. Such performance is higher than that of two previous models. The ablation experiments prove the effects of modules in this model. Finally, we also analyze the weakness and strength of the model, uncovering which herb-disease pairs that HDAPM-NCP can yield reliable or unsatisfied predictions, and a case study is conducted to prove that HDAPM-NCP can discover latent HDAs.

Keywords: Disease; Herb; Herb-disease association; Network consistency projection.

Fig. 1

Entire construction procedures of model HDAPM-NCP. Various information of herbs and diseases are retrieved from a public database HERB. It is used to construct six herb kernels and five disease kernels. Herb kernels and disease kernels are fused into one unified herb kernel and one disease kernel. The network consistency projection is applied to above herb and disease kernels as well as the herb-disease association matrix to generate recommendation matrix.

Fig. 2

ROC and PR curves of model HDAPM-NCP under two types of five-fold cross-validation. (A) ROC curves under global five-fold cross-validation; (B) PR curves under global five-fold cross-validation; (C) ROC curves under local five-fold cross-validation on herbs; (D) PR curves under local five-fold cross-validation on herbs; (E) ROC curves under local five-fold cross-validation on diseases; (F) PR curves under local five-fold cross-validation on diseases. HDAPM-NCP yields high performance under global five-fold cross-validation and local five-fold cross-validation on diseases, whereas its performance under local five-fold cross-validation on herbs is relatively low.

Fig. 3

ROC and PR curves using different kernel functions under global five-fold cross-validation. (A) ROC curves; (B) PR curves. The GIP kernel function yields the highest performance.

Fig. 4

ROC and PR curves using all possible combinations of single herb and single disease kernels under global five-fold cross-validation. (A) ROC curves; (B) PR curves. The AUROC and AUPR values under each kernel combination are lower than those of HDAPM-NCP, implying usage of all herb and disease kernels can improve the performance. A-F represent six herb kernels (A: , B: , C: , D: , E: , F: , see Table 1), whereas V-Z indicate five disease kernels (V: , W: , X: , Y: , Z: , see Table 2).

Fig. 5

ROC and PR curves under global five-fold cross-validation to show the performance of HDAPM-NCP on herb-disease pairs with a fixed herb. (A) ROC curves; (B) PR curves. HDAPM-NCP provides high performance to predict related diseases of some herbs, where its performance on the prediction of related diseases of a few herbs is not satisfied.

Fig. 6

ROC and PR curves under global five-fold cross-validation to illustrate the performance of HDAPM-NCP on three disease types. (A) ROC curves; (B) PR curves. HDAPM-NCP provides almost equal performance on three disease types.