Data-balanced transformer for accelerated ionizable lipid nanoparticles screening in mRNA delivery

Area of Science:

• Biotechnology
• Computational Chemistry
• Drug Delivery Systems

Background:

• Ionizable lipid nanoparticles (LNPs) are crucial for mRNA delivery but screening is time-consuming and costly.
• Efficient selection of LNPs is a major challenge in mRNA drug development.

Purpose of the Study:

• To accelerate the early development of LNPs for mRNA delivery systems.
• To introduce TransLNP, a transformer-based model for predicting LNP transfection efficiency.

Main Methods:

• TransLNP utilizes coarse-grained atomic sequence and fine-grained spatial information.
• Molecular pretraining via 3D coordinate reconstruction and atom prediction improves property prediction.
• The BalMol block addresses data imbalance by smoothing label and feature distributions.

Main Results:

• TransLNP outperforms state-of-the-art methods in transfection property prediction.
• Identified 4267 molecular transfection cliffs, linking structural similarity to transfection efficiency differences.
• Revealed key sources of prediction errors by analyzing molecular transfection cliffs.

Conclusions:

• TransLNP significantly accelerates LNP screening for mRNA delivery.
• The model provides insights into structure-transfection relationships.
• Publicly available code, model, and data facilitate further research.