Machine learning guided structure function predictions enable in silico nanoparticle screening for polymeric gene...

Area of Science:

• Biomaterials Science
• Computational Biology
• Nanotechnology

Background:

• Developing efficient non-viral gene delivery systems remains challenging, particularly for hard-to-transfect cells.
• Current methods often rely on extensive, time-consuming experimental screening.

Purpose of the Study:

• To apply machine learning algorithms for predicting the transfection efficiency and cytotoxicity of synthetic biodegradable polymers, specifically poly(beta-amino ester)s (PBAEs).
• To accelerate the optimization and development of novel gene delivery reagents through in silico evaluation.

Main Methods:

• A dataset of PBAE polymers, including their properties, transfection performance, and toxicity, was compiled.
• Machine learning models, including random forest, were trained using an encoding scheme for polymer structures.
• In silico predictions were validated through de novo synthesis and experimental testing in RAW 264.7 and Hep3B cell lines.

Main Results:

• A random forest model successfully predicted in vitro DNA transfection based on PBAE chemical structure in a cell-dependent manner.
• Experimental validation showed strong correlations between predicted and experimental transfection (Spearman's R = 0.57 and 0.66).
• The computational approach demonstrated utility in predicting outcomes for de novo synthesized gene delivery nanoparticles.

Conclusions:

• In silico screening using machine learning can significantly accelerate the discovery and optimization of non-viral gene delivery materials.
• This computational strategy offers a cost-effective and rapid alternative to traditional brute-force screening methods.
• The developed pipeline for encoding polymer descriptors aids in understanding structure-function relationships for gene delivery applications.