UCSF

Large-scale experimental determination of the protein targets of small molecules is both time-consuming and costly. Computational methods can be used to predict interactions between small molecule and targets, which can help experimentalists find new therapeutic targets or off-targets responsible for undesired side-effects. A data fusion framework for combining multiple similarity computations and a novel method for drawing relationships between drugs based on their clinical effect was developed. Small molecules may be quantitatively compared based on 2D topological structural considerations, based on 3D characteristics directly related to binding, and based on their clinical effects. Given a new molecule along with a set of molecules sharing some biological effect, a single score based on comparison to the known set is produced, reflecting either 2D similarity, 3D similarity, clinical effects similarity or their combination. The methods were systematically applied to a large set of FDA approved drugs (nearly two-thirds).

For prediction of off-target effects, the performance of 3D-similarity over either 2D or clinical effects similarity alone was substantial, and there was added benefit from combining all of the methods. In addition to assessing predictive accuracy of the different similarity methods, the relationship between chemical similarity and pharmacological novelty was studied with regards to protein target modulation and clinical effects. Drug pairs that shared high 3D similarity but low 2D similarity (i.e. having different underlying scaffolds) were shown to be much more likely to exhibit pharmacologically relevant differences in terms of specific target modulation and differences in clinical effects.