Characterizing ssRNA and dsRNA electrophoretic behavior: empirical insights with neural network-aided predictions

Area of Science:

• Biophysics
• Molecular Biology
• Analytical Chemistry

Background:

• RNA-based therapeutics are gaining prominence due to their safety and rapid development.
• Microfluidic electrophoresis is a powerful tool for nucleic acid analysis, but RNA data, especially for modified and double-stranded forms, is limited.
• Chemically modified RNA, like that in mRNA vaccines, and double-stranded RNA by-products require detailed characterization.

Purpose of the Study:

• To empirically analyze the microfluidic electrophoresis of various RNA types, including modified and double-stranded forms.
• To compare experimental findings with existing electrophoretic mobility models.
• To develop and validate predictive models for RNA migration and length using neural networks.

Main Methods:

• Microfluidic electrophoresis was performed on single-stranded and double-stranded RNA, both non-modified and pseudouridine-modified, across different gel concentrations.
• Experimental data was compared against established electrophoretic mobility models.
• Data-driven and physics-informed neural networks were trained to predict RNA migration times and lengths.

Main Results:

• Experimental data on RNA electrophoretic behavior was collected and analyzed.
• A significant discrepancy was observed between experimental data and existing models for certain RNA types.
• Physics-informed neural networks achieved a highly accurate prediction of RNA migration and length with an average error of 0.77%, outperforming data-driven models (12.34% error).

Conclusions:

• Microfluidic electrophoresis is a viable method for characterizing diverse RNA molecules.
• Physics-informed neural networks offer a highly accurate, data-efficient approach for predicting electrophoretic behavior of RNA.
• This predictive capability extends beyond RNA, enabling characterization of other molecules with minimal experimental data.