Prediction Of Protein - Drug Interactions Based On Sequence

Proteins are the important biomacromolecules in life science and medicinal chemistry. Proteins are carrier of many physiological functions and are also important bearers of life activities. Therefore, it is very significant to explore the interaction mechanisms between proteins and drug, which is useful for enhancing the healthy of human beings.The identification of interactions between drugs and target proteins plays a key role in the process of genomic drug discovery. However, the traditional method of biological experiments for determining the of interactions protein and drug is time-consuming and expensive. Thus, prediction of interactions between drugs and target proteins from sequence by using pattern recognition technology has become an important task.In this study, a new sequence-based method for protein-drug interaction prediction is proposed. Evolutionary information feature PSSM_ACT and Pseudo-amino acid composition (PseAAC) feature are extracted from protein sequence, and footprint feature is generated from drug molecular structures and is transformed by discrete Fourier transform and discrete wavelet transform respectively. Next, by using feature combination method, protein features and drug features are combined to form different Protein-Drug pairs as training samples. In the prediction stage, OET_KNN, support vector machine (SVM) and Random Forest are taken as the classifiers to compute the probability of interaction for each sample, and the final decision is determined by the selected threshold. Experimental results on two benchmark datasets with leave-one-out shows that the Sequence-based Method helps improving the prediction accuracy. We also compared the proposed method with popular existing predictors, such as iGPCR_Drug, pNN_FGBF, LargeScale. It is founded out that the proposed method gets the optimal performance.