An Automated Workflow Leveraging Machine Learning for Physical Titer Determination from Cryo-TEM Images of...

Area of Science:

• Biotechnology and Bioprocessing
• Molecular and Cellular Biology
• Computational Biology and Bioinformatics

Background:

• Gene therapies offer transformative potential for untreatable diseases but face challenges in production quality control.
• Current methods for monitoring viral vector impurities, like partially filled or empty capsids, are costly, time-consuming, and subjective.
• Transmission electron microscopy (TEM) is crucial for visualizing nanometer-sized structures but has limitations in sample preparation and manual analysis.

Purpose of the Study:

• To develop and validate a computer vision (CV) model for automating the classification of viral capsids in TEM images.
• To improve the accuracy and efficiency of impurity measurements in gene therapy production.
• To create a user-friendly tool for standardized labeling of capsid physical titers.

Main Methods:

• A machine-learned CV model was trained to classify individual viral capsid images into three categories: full, partially full, or empty, based on DNA content.
• The trained model was applied to predict capsid category distributions in production samples.
• A graphical user interface, named Capsidize, was developed for accessible and standardized labeling of cryo-TEM images.

Main Results:

• The CV model achieved 72% accuracy in classifying capsid fullness.
• The Capsidize software reduced the active working time for labeling capsid titers by 95%.
• Automated CV analysis enhances the precision of impurity measurements compared to manual screening.

Conclusions:

• Computer vision offers a viable solution to automate and enhance the quality control of viral vectors in gene therapy manufacturing.
• The developed CV model and Capsidize tool significantly improve efficiency and standardization in analyzing TEM images.
• This approach addresses critical bottlenecks in gene therapy production, paving the way for more accessible and cost-effective treatments.