DTIP

An integrated tool for predicting drug-target interactions

When designing, a drug is specified to particular therapeutic target proteins to treat diseases. However, due to complex interactions among proteins in a cell (also known as protein interaction network), a drug can target to other proteins. Therefore, prediction of novel drug-target interaction is an important issue in drug discovery. Here, we develop a tool DTIP (Drug-Target Interaction Prediction) for such the task.

Previous studies show that the prediction performance is better when integrating similarity networks/matrices of drugs, targets and bipartite network of known drug-target interactions. Therefore, DTIP is designed based on state-of-the-art algorithms including both network- and machine learning-based methods, which make use those networks.

In particular, we implemented two network-based algorithms RWRH (Random Walk with Restart on Heterogeneous Network), HGBI (Heterogeneous Graph Based Inference), and one semi-supervised learning algorithm RLS (Regularized Least Square). The rankings of candidate targets generated by these algorithms then are aggregated by different methods.

Highly ranked candidate targets can be visualized on the similarity target network as well as supported with annotations and evidence about the predicted interactions.