IEV2Mol: Molecular Generative Model Considering Protein-Ligand Interaction Energy Vectors

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular modeling

Background:

• Structure-based drug design faces challenges in generating drug candidates with precise protein-ligand interactions.
• Accurately predicting and optimizing these interactions is crucial for developing effective therapeutics.

Purpose of the Study:

• To introduce IEV2Mol, a new generative model for designing drug candidates with desired protein-ligand interactions.
• To improve the accuracy of compound generation by incorporating quantitative interaction data.

Main Methods:

• Developed IEV2Mol, integrating interaction energy vectors (IEVs) from docking simulations into a variational autoencoder (VAE) framework.
• Trained the model using SMILES strings and minimized reconstruction error.
• Benchmarked IEV2Mol against random compounds, JT-VAE, and IFP-RNN models.

Main Results:

• IEV2Mol generated compounds that significantly retained more of the query structure's binding mode compared to other methods.
• The model successfully generated compounds with interactions similar to input compounds, irrespective of structural similarity.
• Demonstrated superior performance in generating targeted protein-ligand interactions.

Conclusions:

• IEV2Mol offers a powerful approach for structure-based drug design, enabling the generation of compounds with specific and desired protein-ligand interactions.
• The model's ability to preserve binding modes and generate interaction-specific compounds represents a significant advancement in drug discovery.