Abstract
Identifying unexpected drug-protein interactions is crucial for drug repurposing. We develop a comprehensive proteome scale approach that predicts human protein targets and side effects of drugs. For drug-protein interaction prediction, FINDSITEcomb, whose average precision is ~30% and recall ~27%, is employed. For side effect prediction, a new method is developed with a precision of ~57% and a recall of ~24%. Our predictions show that drugs are quite promiscuous, with the average (median) number of human targets per drug of 329 (38), while a given protein interacts with 57 drugs. The result implies that drug side effects are inevitable and existing drugs may be useful for repurposing, with only ~1,000 human proteins likely causing serious side effects. A killing index derived from serious side effects has a strong correlation with FDA approved drugs being withdrawn. Therefore, it provides a pre-filter for new drug development. The methodology is free to the academic community on the DR. PRODIS (DRugome, PROteome, and DISeasome) webserver at http://cssb.biology.gatech.edu/dr.prodis/. DR. PRODIS provides protein targets of drugs, drugs for a given protein target, associated diseases and side effects of drugs, as well as an interface for the virtual target screening of new compounds.
Highlights
Identifying unexpected drug-protein interactions is crucial for drug repurposing
The intrinsic promiscuity of a drug is partly responsible for its unintended side effects[5], but this suggests that FDA approved drugs could be utilized for large scale repurposing
We present results for the virtual target screening of DrugBank drugs against the human proteome and describe promising examples of drug repurposing to treat a variety of diseases
Summary
Identifying unexpected drug-protein interactions is crucial for drug repurposing. We develop a comprehensive proteome scale approach that predicts human protein targets and side effects of drugs. For a given protein target, the chemical similarity of a drug to its known binding ligands was employed to predict possible association to a given protein target[17,25] These methods require prior knowledge of a target protein’s or drug’s binding partners, side effects, interaction network, etc. To address the limitation imposed by the requirement of prior knowledge of small molecule ligand-protein interactions, recent developments infer interactions from neighbors (evolutionarily related proteins)[27,28] where such interactions are known They have not yet been tested on a large scale, e.g. on DrugBank drugs[8] when there are no known interactions for a given drug or a target of interest. This approach is still limited by the requirement of having high-resolution target protein structures (available for at most only 1/3 of the human proteome9), and a lack of accurate scoring functions to rank docked ligands[29,30]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
