Activity Prediction Of Compounds Based On Biological Relevance Spectrum And The Implementation Of Biorel Webserver

In recent years, drug discovery has been faced with the dilemma of high input and low output. On the basis of recent studies, we infer that the main reason for this situation is the design of compound libraries is too dependent on combinatorial chemistry method, which lead to the insufficient diversity of compounds, and unpredictable nature of pharmacokinetic ADME/T.To address this problem, in2009, we proposed a novel index named "Biological Relevance (BR)", which can be used to quantitatively evaluate the possibility of biological origination for a compound or a database and of its importance to biological organism. Prior research results indicated that this index could be used to discriminate between compounds from different sources. Subsequently, the similarity of the objective molecule and BRCD compounds forms a2000-dimensional vector, named Biological Relevance Spectrum (BRS). BRS reflects the compounds’ distribution in biological relevant chemical space, and implies important information of molecules. Therefore, it can be used as a multi-dimensional molecular descriptor to explore the biological relevant chemical space, and screen compounds with good ADME/T properties and bioactivity.Firstly, in this paper, the characterization ability of BRS was verified by using principal component analysis method. Then43possible activity models (PoAct) were built with the BRSs of MDDR active compounds and ACD inactive compounds. The average precision rate of training set is97.26%, and the average recall rate is94.39%. The average precision rate of test set is92.65%, and the average recall rate is92.65%. Statistically, these models are with high predictive ability, which fully illustrates that BRS has a strong capacity for chemical information mining and characterizing.To facilitate the screening and exploring of potential compounds with good ADME/T properties and bioactivity based on BRS, we developed a webserver named BioRel, which integrated many useful models for predicting compounds’possible activities (PoAct), absorption (HIA), distribution (BBB, and NPL), acute toxicity (AcTox) in vivo. It can be used to rapidly screen potential drugs and greatly reduce the difficulty of BRS guiding drug discovery and design. BioRel can be accessed at http://ibi.hzau.edu.cn/biorel.