Multi-target-based polypharmacology prediction (mTPP): An approach using virtual screening and machine learning for multi-target drug discovery

Polypharmacology has become a new paradigm in drug discovery and plays an increasingly vital role in discovering multi-target drugs. In this context, multi-target drugs are a promising approach to treating polygenic diseases. Many in-silico prediction methods have been developed to screen active molecules acting on multiple targets. The relationship between the action of multiple targets and the drug's overall efficacy is significant for developing multi-target drugs. So, the prediction method for this relationship urgently needs to be developed. This paper introduces multi-target-based polypharmacology prediction (mTPP), an approach using virtual screening and machine learning to explore the relationship. To predict the activity of the potential hepatoprotective components, the data on the binding strength of a single ingredient with multiple targets and the proliferation rate of the compounds against acetaminophen (APAP)-induced injury L02 cells were all used to construct the mTPP model by Multi-layer Perceptron (MLP), Support Vactor Regression (SVR), Decision Tree Regressor (DTR), and Gradient Boost Regression (GBR) algorithms. Compared with MLP, SVR, and DTR algorithms, GBR algorithms showed the best performance with R2test = 0.73 and EVtest = 0.75. In addition, 20 candidates with potential effects against drug-induced liver injury (DILI) were predicted by the mTPP model. Furthermore, 2 of the 20 candidates, Chelerythrine and Biochanin A, were applied to evaluate the model's accuracy. The results showed that Chelerythrine and Biochanin A could improve the viability of APAP-induced injury cells. Thus, the mTPP model is hoped to help develop polypharmacology and discover multi-target drugs.