Machine learning-driven exosome-mimetic lipid nanoparticles for tumor-specific targeting.

Area of Science:

• Biomedical Engineering
• Nanotechnology
• Artificial Intelligence in Medicine

Background:

• Exosome-mimetic lipid nanoparticles (ENPs) show promise for cancer therapy, offering advantages over traditional PEGylated lipid nanoparticles (LNPs).
• Designing effective ENPs is complex due to intricate lipid composition requirements.
• There is a need for advanced computational methods to guide the rational design of ENPs.

Purpose of the Study:

• To develop and validate a hybrid algorithm for optimizing exosome-mimetic nanoparticle formulations.
• To predict key nanoparticle properties like size, zeta potential, and polydispersity index.
• To leverage AI for designing safer and more effective cancer nanomedicines.

Main Methods:

• A hybrid algorithm was developed, integrating physicochemical modeling and feature extraction.
• The algorithm was trained on a large dataset (17,800 compositions) augmented by the LipidGAN generative model.
• Optimization focused on predicting nanoparticle properties for exosome-mimetic formulations.

Main Results:

• The AI algorithm successfully identified optimal lipid formulations for ENPs.
• In vitro validation in HeLa, H1975, and MCF-7 cancer cell lines showed minimal toxicity (>90% cell viability).
• Efficient and cell-type-specific cellular uptake was observed (91-95%).

Conclusions:

• AI-driven lipid design can effectively emulate natural exosome functionality.
• This approach facilitates the development of safe, effective, and personalized cancer nanomedicines.
• The optimized ENPs demonstrate significant potential for targeted cancer therapy.