Construction of Sonosensitizer-Drug Co-Assembly Based on Deep Learning Method

Area of Science:

• Drug delivery systems
• Nanomedicine
• Computational chemistry

Background:

• Drug co-assemblies offer advantages like controlled drug release and combined therapeutic effects.
• Developing effective co-assembly strategies is crucial for their clinical application.
• Predicting particle size is key to controlling co-assembly properties.

Purpose of the Study:

• To develop a deep learning model for predicting drug mixture particle size.
• To analyze molecular structural factors influencing drug co-assembly.
• To validate the model's prediction of a novel nanomedicine for cancer treatment.

Main Methods:

• Utilized a graph neural network to extract molecular features (atomic, bond, structural).
• Implemented a multi-scale cross-attention mechanism for integrating drug substructure information.
• Employed ablation experiments to assess the influence of molecular properties.

Main Results:

• Achieved high prediction accuracy: 90.00% precision, 96.00% recall, and 91.67% F1-score.
• The model successfully predicted the co-assembly of methotrexate and emodin into NanoME.
• Experimental validation confirmed NanoME's utility in liver cancer imaging and therapy.

Conclusions:

• The deep learning-based sonosensitizer-drug interaction (SDI) model accurately predicts particle size in drug mixtures.
• Molecular structure significantly impacts drug co-assembly, as revealed by the model.
• Validated nanomedicine NanoME shows promise for dual-mode cancer therapy and diagnostics.