Which is a new contribution in the field of drug discovery and development. As well known, RF introduces two sources of randomness into the trees: random training set and random input vectors. Each tree is grown using a bootstrap sample of training data at each node, with best split chosen from random sample of mtry variables instead of all variables. The Bagging method causes the low bias and high variance of the unpruned trees, but the variance is reduced by averaging the bootstrapped trees. Because each node is split using the best among a subset of predictors Albaspidin-AA randomly chosen at that node, this strategy turns out to perform very well compared with many other classifiers such as ANN and SVM, and is more robustness against overfitting. However, it should be also pointed out that SVM allows us to use a tensor product space, with no extra calculation time with respect to the joint space, and versatile choice of similarity measures for proteins and ligands, which is a clear advantage of SVM with respect to this proposed RF approach. As shown in the Results section, the proposed method enables to identify the unrelated proteins that may share structurally similar pockets in the 3D space, the advantage effect of which is derived from the learning ligand similarities. If the unrelated receptors do not possess similar ligands in the learning dataset, then the protein encoding plays a key role for the unrelated targets identification, which have been well applied for the prediction of protein structural and functional classes, protein-protein interactions and subcellular locations. Particularly, most descriptors of this encoding such as Composition of Solvent Accessibility and Distribution of Charge are highly useful for representing and distinguishing interaction profiles, which is essential for the successful application of statistical learning methods in predicting the interaction profiles between drugs and targets. Although this encoding can effectively describe the interaction profiles of the whole protein, it is limited by the description and extraction of the 3D structure features of the ligand-binding domains. Therefore, obtaining optimal protein encoding method will require further research. The main advantages of our proposed approach are summarized as follows: 1) In building models, no explicit procedure is needed to select the information shared among all drug-target recognitions. 2) Our system is suitable for simultaneously screening huge numbers of drug candidates and candidate targets from a systematic level; 3) Compared with the structure-based simulation methods, this approach is not limited by the 3D structure data of targets; 4) The structure similarity of 
a chemical with ligands that bind to proteins which this compound is not known to bind can be used by our method to discover unknown Catharanthine sulfate activities. 5) The system is able to identify those ligands/receptors that exhibit scaffold hopping/low sequence similarity; and 6) this approach can aid in discovery of multi-target drugs by recognizing the group of proteins targeted by a particular ligand. Based on all these, we conclude that the present systematic biology-based “integrationist mindset” is appropriate for modeling and understanding complex drug-target interaction networks. This perspective could devote to the recapturing of known small molecules and the explication of mechanisms of drug side effects, and finally is anticipated to help fill in the existing gap between chemical genomics and network pharmacology. Massive UDPS analysis enabled sensitive baseline detection of minor viral populations associated with NA resistance.