Enhancing ADMET property predictions using cross-aligned multimodal attention mechanisms

Area of Science:

• Computational chemistry
• Drug discovery
• Pharmacokinetics

Background:

• Accurate prediction of Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET) properties is vital for successful drug development.
• Existing methods often struggle with the complexity and heterogeneity of ADMET data.

Purpose of the Study:

• To develop and present a novel approach for enhancing ADMET property predictions.
• To improve the accuracy and efficiency of predicting drug metabolism and pharmacokinetic properties.

Main Methods:

• Utilized Cross-Aligned Multimodal Attention (CMA) mechanisms with pretrained models (GROVER, ResNet) and multimodal techniques.
• Processed ADMET data using image processing, graph neural networks, and chemical fingerprinting.
• Employed Grad-CAM for model interpretation, visualizing compound property-fragment relationships.

Main Results:

• Successfully integrated multimodal data sources through cross-modal alignment.
• Demonstrated improved efficiency and accuracy in ADMET property predictions.
• Developed an ADMET property prediction server implementing the CMA-based model.

Conclusions:

• The novel CMA-based approach significantly enhances ADMET property prediction accuracy.
• This integration of multimodal data and pretrained models opens new research avenues in molecular science for drug design.
• The developed server provides a valuable tool for drug design and evaluation.