O06_05
Design of Novel Compounds Through Protein-Ligand Interaction-Based Generative Methods
Mami OZAWA1, Shogo NAKAMURA1, Nobuaki YASUO2, MASAKAZU SEKIJIMA *1
1Department of Computer Science, Tokyo Institute of Technology
2TAC-MI, Tokyo Institute of Technology
( * E-mail: sekijima@c.titech.ac.jp)
The generation of new compounds with protein-ligand interactions is an important issue in structure-based drug design. In this study, we propose a model for generating new compounds called “IEV2Mol” that incorporates the interaction energy vector (IEV) between the protein and ligand obtained from docking simulations. This IEV quantitatively captures the strength of each interaction type, such as hydrogen bonds, electrostatic interactions, and van der Waals forces, and unlike the conventional interaction fingerprint (IFP), it reflects the strength of the interaction. IEV2Mol , by integrating IEV into an end-to-end variational autoencoder (VAE) framework that learns chemical space from SMILES representations and minimizes SMILES reconstruction error, can generate compounds with the desired interactions more accurately.
To evaluate the effectiveness of this model, we conducted benchmark tests comparing it with randomly selected compounds, an unconstrained VAE model (JT-VAE), and an RNN model based on interaction fingerprints (IFP-RNN). The results showed that the compounds generated by IEV2Mol had a significantly higher rate of retaining the binding mode of the query structure than the other methods.
The IEV2Mol proposed in this study is expected to contribute to the efficiency of compound generation based on interaction energy in the design of new compounds for target proteins. In addition, the source code and trained models for IEV2Mol, JT-VAE, and IFP-RNN used in this study are available under the MIT license.