Deep Learning-Aided Drug Discovery via the Latent Space Visualization of Deep Neural Networks
Yasunobu YAMASHITA *1, Yuuki TAKI1, Wakabayashi YOSHINORI2, 3, Junichiro KANAZAWA2, Paddy R. A. MELSEN1, Yuri TAKADA1, Yukihiro ITOH1, Masanobu UCHIYAMA2, Takayashi SUZUKI1
1SANKEN, The University of Osaka
2Graduate School of Pharmaceutical Sciences, The University of Tokyo
3BY-HEX LLP
[Purpose]
Drug discovery using deep learning has attracted much attention. However, deep learning models remain unpolished and do not always provide drug leads. In this study, we propose a drug screening method based on the visualization of a hidden layer in the output process of a graph-convolutional network (GCN)-based deep learning.1) Using this method, we performed a proof-of-concept study to identify potential HDAC1 inhibitors as anticancer agents.
[Method]
The following procedure was used to identify a lead compound: (i) preparation of a chemical library that included a reported HDAC1 inhibitor as a probe molecule, (ii) screening of the library using the GCN-based deep learning model, (iii) visualization of the hidden layer as a chemical space, and (iv) experimental evaluation of compounds plotted near the probe molecule in the space.
[Results and Discussion]
Compound screening was performed as mentioned above, using a trained deep learning model. Specifically, 219,491 compounds in the library of the Drug Discovery Initiative (The University of Tokyo) were input into the deep learning model. Using this model, we screened 40 compounds plotted closer to a probe HDAC1 inhibitor, vorinostat, in the visualized chemical space and evaluated them using an HDAC1 inhibitory assay. As a result, four novel HDAC1 inhibitors were identified.
[Conclusions]
Our deep learning–based method enabled the efficient identification of new lead compounds as HDAC1 inhibitors. These findings support the usefulness of our deep-learning-aided screening method.
Reference
1) Y. Yamashita, Y. Taki, Y. Wakabayashi, J. Kanazawa, P. R. A. Melsen, Y. Takada, Y. Itoh, M. Uchiyama, T. Suzuki ACS Med Chem Lett. 2025, 16, 1299–1304.