Authorship：Last author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)  

The design of drug molecules is a critical stage in the drug discovery process. The structure-based drug design has long played an important role in efficient development. Significant progress has been made in recent years in the generation of 3D molecules via deep generation models. However, while many existing models have succeeded in incorporating structural information on target proteins, they have not been able to address important interactions between protein and drug molecules, especially hydrogen bonds. In this study, we propose DiffInt as a novel structure-based approach that explicitly addresses interactions. The model naturally incorporates hydrogen bonds between protein and ligand molecules by treating them as pseudoparticles. The experimental results show that DiffInt reproduces hydrogen bonds, and the hydrogen binding energies significantly outperform those of existing models. To facilitate the use of our tool for generating new drug molecules based on any protein's three-dimensional structure, we have made the source code and trained model available on GitHub (https://github.com/sekijima-lab/DiffInt) under the MIT license, with the execution environment provided on Google Colab.

File： sako-et-al-2024-diffint-a-diffusion-model-for-structure-based-drug-design-with-explicit-hydrogen-bond-interaction (1).pdf

DOI： 10.1021/acs.jcim.4c01385

PubMed

researchmap

Authorship：Last author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：American Chemical Society (ACS)  

File： suzuki-et-al-2024-mothra-multiobjective-de-novo-molecular-generation-using-monte-carlo-tree-search (1).pdf

DOI： 10.1021/acs.jcim.4c00759

researchmap

Authorship：Last author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)  

Generating drug candidates with desired protein-ligand interactions is a significant challenge in structure-based drug design. In this study, a new generative model, IEV2Mol, is proposed that incorporates interaction energy vectors (IEVs) between proteins and ligands obtained from docking simulations, which quantitatively capture the strength of each interaction type, such as hydrogen bonds, electrostatic interactions, and van der Waals forces. By integrating this IEV into an end-to-end variational autoencoder (VAE) framework that learns the chemical space from SMILES and minimizes the reconstruction error of the SMILES, the model can more accurately generate compounds with the desired interactions. To evaluate the effectiveness of IEV2Mol, we performed benchmark comparisons with randomly selected compounds, unconstrained VAE models (JT-VAE), and compounds generated by RNN models based on interaction fingerprints (IFP-RNN). The results show that the compounds generated by IEV2Mol retain a significantly greater percentage of the binding mode of the query structure than those of the other methods. Furthermore, IEV2Mol was able to generate compounds with interactions similar to those of the input compounds, regardless of structural similarity. The source code and trained models for IEV2Mol, JT-VAE, and IFP-RNN designed for generating compounds active against the DRD2, AA2AR, and AKT1, as well as the data sets (DM-QP-1M, active compounds to each protein, and ChEMBL33) utilized in this study, are released under the MIT License and available at https://github.com/sekijima-lab/IEV2Mol.

File： ozawa-et-al-2024-iev2mol-molecular-generative-model-considering-protein-ligand-interaction-energy-vectors.pdf

DOI： 10.1021/acs.jcim.4c00842

PubMed

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1016/j.ijbiomac.2024.129490

Scopus

PubMed

researchmap

Event date： 2025.10

Language：English   Presentation type：Poster presentation  

researchmap

Language：English   Presentation type：Poster presentation  

researchmap

Language：English   Presentation type：Poster presentation  

researchmap

Language：English   Presentation type：Poster presentation  

researchmap

Language：English   Presentation type：Poster presentation  

researchmap

Language：English   Presentation type：Poster presentation  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Work type：Software  

Drug-target interaction (DTI) prediction is a problem that identifies novel protein-ligand interactions from previous information. DTI plays an important role in computer-aided drug discovery because it is related to many aspects of drug discovery, such as virtual screening, target prediction, side effect prediction, and drug repositioning. Previous methods can be divided into two types: content-based methods and collaborative filtering. However, both types have problems, namely, a lack of diversity and ”cold-start” problems. In this study, we developed a new method named CoDe-DTI (COllaborative DEep learning-based Drug Target Interaction predictor) that combines both methods to avoid these problems. CoDe-DTI is based on collaborative deep learning, which introduces the information of chemical structures into the latent variables by combining probabilistic matrix factorization with a denoising autoencoder. Fivefold cross validation showed that CoDe-DTI significantly outperformed other machine learning-based methods regarding hit rate (top 5%). Comparing
between drugwise cross validation and interactionwise cross validation, CoDe-DTI still works even when there is no interaction
information of the input ligand exists. The source code for CoDeDTI is available at: https://github.com/sekijima-lab/CoDe-DTI .

researchmap

Work type：Software  

researchmap