Applying machine learning to predict viral assembly for adeno-associated virus capsid libraries

Area of Science:

• Virology
• Gene Therapy
• Computational Biology

Background:

• Adeno-associated virus (AAV) capsid libraries are crucial for selecting gene therapy vectors.
• Next-generation sequencing (NGS) generates large datasets from these libraries, offering potential for advanced analysis.
• Current methods for library selection are primarily *in vivo*, limiting predictive capabilities.

Purpose of the Study:

• To investigate the use of machine learning (ML) for *in silico* analysis of AAV capsid libraries.
• To develop predictive models for identifying viable AAV capsid variants.
• To enhance the design of more robust and efficient AAV-based gene therapy vectors.

Main Methods:

• Utilized data from AAV capsid libraries before and after viral assembly to train ML algorithms.
• Employed artificial neural networks (ANNs) and support vector machines (SVMs) for predictive modeling.
• Simulated hypothetical mutation patterns based on ML model predictions.

Main Results:

• Developed ML models capable of predicting the assembly of unknown capsid variants into viable virus-like structures.
• Identified specific amino acid residues (N495, G546, I554) critical for AAV2 capsid assembly.
• Generated and validated comparative libraries using ML-derived data.

Conclusions:

• Machine learning significantly enhances the prediction of AAV capsid assembly and viability.
• ML-driven insights can guide the rational design of improved AAV capsid libraries for gene therapy.
• This study demonstrates the predictive power of ML in optimizing vector design for viral gene therapy applications.