Three-Dimensional Classification Structure-Activity Relationship Analysis Using Convolutional Neural Network.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Quantitative structure-activity relationship (QSAR) methods are vital in drug design.
• Three-dimensional QSAR techniques like comparative molecular field analysis (CoMFA) require precise compound alignment, posing challenges.
• Alignment-independent methods such as VolSurf, GRIND, and Anchor-GRIND have been developed to address these limitations.

Purpose of the Study:

• To develop a novel QSAR prediction model using molecular interaction field grid potentials as input for convolutional neural networks.
• To overcome the alignment limitations inherent in traditional 3D-QSAR methods.
• To provide a more accurate and versatile QSAR modeling approach for drug design.

Main Methods:

• Utilized molecular interaction field grid potentials as input data.
• Employed convolutional neural networks (CNNs) for model construction.
• Compared the proposed CNN-based QSAR model against conventional descriptor-based QSAR and Anchor-GRIND techniques.

Main Results:

• The proposed CNN-based QSAR model demonstrated superior accuracy compared to conventional descriptor-based QSAR models.
• The method achieved higher accuracy than existing Anchor-GRIND techniques.
• The model proved to be target-independent and provided insights into the importance of individual atoms within chemical datasets.

Conclusions:

• The developed technique offers a highly accurate and efficient QSAR modeling approach.
• Its target-independent nature and ability to identify key atomic contributions enhance its utility in drug design.
• This CNN-based method is poised for significant applications in future drug discovery operations.