Unveiling the hidden power for drug-target interaction prediction: a game-changing approach

Accurately predicting the Drug-Protein Interaction (DPI) is crucial in virtual drug screening. However, current methodologies tend to allocate equal weighting to amino acids and atoms in encoding protein and drug sequences, thereby neglecting the varying contributions from distinct motifs.
To tackle this issue, a group of researchers headed by Juan Liu have recently published their pioneering research on the matter in Frontiers of Computer Science, jointly published by Higher Education Press and Springer Nature.
Their research introduced a revolutionary method, FragDPI, for the prediction of drug-protein binding affinity. This approach represents the initial endeavor to incorporate fragment coding and merge the sequence information of both drugs and proteins, hence preserving the primary features related to DPI interactions. Furthermore, this method employs transfer learning from significant DPI datasets to provide prospective DPI components.
Experimental results demonstrate that the FragDPI model yields commendable outcomes compared to the baselines, including deep neural networks. Intriguingly, the model accurately identified the specific interaction parts of the DTI pairs, thereby aiding in discovering new potential DTI pairs. FragDPI presents a novel approach for mining interacting fragments from DPI mechanism, thereby providing a fresh perspective towards drug discovery.