May 20, 2022
Keio University
A research group at Keio University, including Kazuki Yasuda (a second-year master's student at the Graduate School of Science and Technology), Katsuhiro Endo (a second-year student in the Doctoral Programs at the time of the research), Project Associate Professor Hidenori Hirano, Senior Assistant Professor Eiji Yamamoto from the Faculty of Science and Technology, and Professor Kenji Yasuoka, has proposed a new method that uses machine learning to predict the binding affinity between drugs (ligands) and proteins from their structural fluctuations.
In recent years, there has been active research into predicting the binding affinity of ligands using computer simulations. However, conventional methods require long computation times and extensive computational resources. In this study, we have successfully evaluated the binding affinity of ligands with fewer computational resources by combining short-time molecular dynamics simulations with machine learning methods, including deep learning.
This study focused on the "structural fluctuations" of proteins. Proteins are flexible substances, and their structures fluctuate within cells due to interactions with various molecules such as water and ions. These structural fluctuations are also significantly influenced by the bound ligand. Therefore, our research group hypothesized that "information" regarding the affinity between a protein and a ligand might be reflected in the changes in the protein's structural fluctuations upon ligand binding. To verify this, we performed multiple molecular dynamics simulations for proteins bound to ligands with different binding affinities to obtain information on the protein's structural fluctuations and characterized these fluctuations using machine learning methods. The results revealed a strong correlation between the characteristics of the protein's structural fluctuations and the ligand's affinity. As the findings of this study have the potential to become an efficient method for evaluating ligand affinity, they are expected to contribute to future drug discovery research.
The results of this research were published in the international journal "Communications Biology" on May 19, 2022.
For the full press release, please see below.