Exploring-DTI-Networks

Contrary to the once prevalent lock-and-key paradigm of substrate and enzyme interaction, scientists have found that the interactions between drugs and their macromolecular targets can often be quite promiscuous. For this reason, when designing drugs for a target receptor, it is important to consider how the that drug could interact with other structures. A drug-target interaction network can be useful for mining common interactions between drugs and their targets, understanding the systemic effect of drugs, and also for predicting new, undiscovered interactions. Several databases collect information about drug-target associations. Among the largest of these are NCBI’s PubChem, the European Bioinformatics Institute’s ChEMBL, Pharos, and BindingDB. There are also similar databases that record drug-drug interactions and protein-protein interactions. Comparing the structure of a new entity to those in the network can allow us to make inferences about its associations and activity. For proteins, a simple structural comparison can be done via sequence alignment, and for drugs, a common structural comparison metric is the Tanimoto similarity index. A possible approach for predicting new drug–target interactions is to use binary classification methods, taking drug–target pairs as an input for machine learning classifiers such as neural networks or support vector machines.