Classifying the Stoichiometry of Virus-like Particles with Interpretable Machine Learning.

Area of Science:

• Structural Biology
• Vaccine Development
• Computational Biology

Background:

• Virus-like particles (VLPs) are essential for vaccine development due to their immunogenicity.
• Determining VLP stoichiometry is critical for vaccine optimization but current methods are laborious.
• There is a need for efficient and data-driven approaches to classify VLP stoichiometry.

Purpose of the Study:

• To develop an efficient, data-driven pipeline for classifying VLP stoichiometry.
• To identify key protein sequence features that influence VLP assembly.
• To provide an interpretable machine learning model for stoichiometry classification.

Main Methods:

• Curated a new dataset for VLP stoichiometry classification.
• Developed an interpretable pipeline using linear machine learning models.
• Explored feature encoding and identified influential protein sequence features.

Main Results:

• The proposed pipeline accurately classifies VLP stoichiometry.
• The model successfully reveals protein features that potentially influence VLP assembly.
• The approach offers an efficient alternative to traditional experimental methods.

Conclusions:

• Accurate VLP stoichiometry classification streamlines vaccine design.
• This data-driven approach accelerates the development of novel vaccines.
• The developed pipeline and dataset are publicly available for further research.