What is drug repurposing, virtual screening and drug-target interaction prediction?¶
Drug repurposing aims to repivot an existing drug to a new therapy.
- Virtual screening means to use computer software to automatically screen
- a huge space of potential drug-target pairs to obtain a predicted binding score.
Both of these tasks are able to save cost, time, and facilitate drug discovery. Deep learning has shown strong performance in repurposing and screening. It relies on the accurate and fast prediction of a fundamental task: drug-target interaction prediction. DTI prediction task aims to predict the input drug target pair’s interaction probability or binding score. Given a powerful DTI model that is able to generalize over a new unseen dataset, we can then extend to repurposing/screening. For repurposing, given a new target of interest, we can first pair it to a repurposing drug library. Then this list of input drug-target pairs is fed into the trained DTI model, which will output the predicted binding score. Similarly, for virtual screening, given a list of screening drug-target pairs we want, the DTI model can output the predicted interaction binding scores. We can then rank the predicted outcome based on their binding scores and test the top-k options in the wet lab after manual inspection. DeepPurpose automates this process. By only requiring one line of code, it aggregates five pretrained deep learning models and retrieves a list of ranked potential outcomes.
Identifying Drug-Target Interactions (DTI) will greatly narrow down the scope of search of candidate medications, and thus can plays a pivotal role in drug discovery. Drugs usually interact with one or more proteins to achieve their functions. However, discovering novel interactions between drugs and target proteins is crucial for the development of new drugs, since the aberrant expression of proteins may cause side effects of drugs.
Considering that in vitro experiments are extremely costly and time-consuming, high efficiency computational prediction methods could serve as promising strategies for drug-target interaction (DTI) prediction. In this project, our goal is to focus on deep learning approaches for drug-target interaction (DTI) prediction.