Brand new drug design based on molecular generation takes advantages of advanced computing technology to generate potential candidate drug with brand new structure, so as to accelerate the development of new drugs. Recently, molecular generating approaches targeted at protein target structure has been getting more and more attention, whereas the approach tended to input the one-dimensional characterization layer of the target structure as a condition into the model. In light of the fact that the approach ignored the interaction between protein and small molecules, its feasibility and interpretability remained highly controversial, despite being able to direct molecules to generate towards a specific structure.
In October, 2023, Professor Tingjun Hou, Professor Changyu Xie, and Associate Professor Yu Kang from CPS-ZJU, jointly published a paper titled" Learning on topological surface and geometric structure for 3D molecular generation" in Nature Computational Science, which proposed a molecular generation method based on classical lock-key model called SurfGen.
In the method, by utilizing the technique in computer vision to construct triangular envelope surfaces, the author characterized the proteins as abstract surfaces embedded with relevant energy features to then perform topological learning through the framework Geodesic-GNN, while protein-small molecule interactions were learned through the framework Groattn-GNN. The calculation results indicated that SurfGen was superior than conventional and other methods on the protein-small molecule benchmark test set. Besides, the author also designed a series of experiments to demonstrate the performance of SurfGen as well as the limitation of present methods from the perspective of both geometric and energy matching. At last, the author proposed a method to address drug-resistant mutations through the use of SurfGen, which provided a new thought to resolve the drug-off-target issue.
More information: Haotian Zhang and Tianyue Wang, postgraduate students of CPS-ZJU, are the first authors. Professor Tingjun Hou, Professor Changyu Xie, and Associate Professor Yu Kang are the co-corresponding authors.
Translator: Zihao Liu
Editor: Yichen Zhu