Drug-target Interaction Prediction And Application Based On Ensemble Learning Method

The process of new drug research and development is often time-consuming,costly,and has some randomness and blindness.Typically,developing a new drug will take an average of 10-15 years and more than 800 million US dollars from research and development to successful launch.Despite the large investment in drug development,the output is not ideal.Therefore,accelerating the development of innovative drugs has become a global consensus.As the origin of drug development,the identification of drug-target interactions plays an important role in drug discovery.However,traditional biological experiments and clinical trial methods require not only a large amount of human and financial resources,but also have high false positive rate and false negative rate.With the rapid development of machine learning and bioinformatics,computer-aided drug-target interaction prediction methods have become a fast and accurate means for drug target identification.This paper uses ensemble learning to predict drug-target interactions,the main contents are as follows:(1)Numerical characterization of drug compounds and target protein sequences.Considering that the molecular descriptor can convert the chemical information encoded in the molecular symbol representation into useful values according to the physical properties of the drug,the number of atoms,and the number of chemical bonds,to distinguish different drug molecules.Therefore,this article uses molecular substructure fingerprints to characterize drug compounds.In order to contain as much as possible the biological evolution information of the target protein,a Position-Specific Score Matrix(PSSM)is used herein to represent the target protein sequence.(2)For the target protein scoring matrix after numerical characterization,this paper uses the local phase quantization descriptor and the Legendre moment feature description algorithm to objectively and efficiently extract representative biometric features.Thus,an eigenvector representation of the drug and the target protein can be obtained,preparing for the next drug-target interaction prediction.(3)Design of predictive model based on ensemble learning.This paper uses two ensemble learning classifiers,namely Rotation Forest(RoF)and Random Forest(RF)to predict drug-target interaction by combining two feature extraction algorithms respectively.Finally,the proposed model is applied to four mainstream gold standard datasets and compared with support vector machine classifier and other representative models.The experimental results show that the prediction model based on ensemble learning method proposed in this paper can effectively predict drug-target interaction.