The above two Mechlorethamine hydrochloride methods are predictive approaches that provide with novel, testable small molecule-target associations, while the text mining-based approach is a way to gather information previously existing in the literature that would probably have been missed. Additionally, it also suffers from an inability to detect new biological findings, and their efficiency is generally hampered by the redundancy of the compound and gene names in literature. Therefore, the genome-wide application of LBVS, SBVS and texting miningbased methods still has many limitations. An effective means that might overcome these problems is not to considerate each drug or target independently from other drugs or targets, but to take the standpoint of chemical genomics which could open up new opportunities to identify new drug leads or therapeutic targets instead. Chemical genomics aims at exploiting the whole chemical space, which corresponds to not only the space of the small molecules but also of those proteins interacting with the molecules. Recently, several chemical genomics approaches, including the ligand-based, targetbased or target-ligand methods have been developed to predict the interactions between compounds and proteins. The ligand-based method that integrated the protein targets was designed at the level of families or subfamilies which is appropriate for some specific protein families such as GPCRs. Based on the ligand binding site similarity, Frimurer et al developed a target-based approach which clustered the Albaspidin-AA receptors and pooled together the known ligands for each cluster to infer shared ligands. Different from these two approaches, the target-ligand approach combines the ligand chemical space, target space and the currently known drug-target networks information to construct a complex forecast system, with purpose to predict ligands or targets for a given target or ligand without prior attempting to define a special set of similar receptors or ligands. For instance, the amino acid sequences, 2-dimensional chemical structures and mass-spectrometry data have been collected together to build a statistical method for predicting compound-protein interactions based on 519 approved drugs and their 291 associated targets. Similarly, the chemical functional groups and biological features have also been adopted to establish the classification models for predicting the drug-target interaction network. Interestingly, without the negative samples, the semi- supervised machine learning algorithm NetLapRLS has been developed based on heterogeneous biological data, which could effectively predict the interaction of each chemical-protein pair. Furthermore, based on DrugBank database, the sets of chemical substructures and protein domains have also been collected and effectively analyzed using Sparse Canonical Correspondence Analysis and Support Vector Machine methods to identify molecular recognition rules between drugs and targets. However, all these aforementioned methods might suffer from the small receptor space which only focused on certain protein families or the limited chemical space only covered by the FDA approved drugs. To predict the drug-target interactions, we have designed a set of in silico tools by incorporating the chemical, genomic and pharmacological information into an integrated framework using the largest available dataset of DrugBank database. The predictions are based on extraction of conserved patterns from subdivided interaction vectors involving both proteins and their corresponding ligands.