Image Classification of Degraded Polysorbate, Protein and Silicone Oil Sub-Visible Particles Detected by Flow-Imaging...

Area of Science:

• Biopharmaceutical analysis
• Particle characterization
• Machine learning applications

Background:

• Polysorbate degradation in biopharmaceuticals generates sub-visible particles (SvPs), including free-fatty acids (FFAs) and protein aggregates.
• Flow-imaging microscopy (FIM) is crucial for SvP analysis but generates large datasets, hindering rapid manual characterization.
• Automated analysis is needed to overcome the limitations of manual FIM data interpretation.

Purpose of the Study:

• To develop and apply a custom convolutional neural network (CNN) for classifying SvP images obtained via FIM.
• To assess the CNN's ability to differentiate between FFAs, proteinaceous particles, and silicon oil droplets.
• To evaluate the CNN's performance in predicting the composition of unknown and labeled test samples.

Main Methods:

• Utilized a custom convolutional neural network (CNN) architecture for image classification.
• Trained the CNN on FIM-generated images of various sub-visible particles.
• Validated the CNN's predictive capability using pooled test samples with known and unknown compositions.

Main Results:

• The CNN demonstrated robust classification of SvP images, distinguishing FFAs, proteinaceous particles, and silicon oil droplets.
• Minor misclassifications between FFAs and proteinaceous particles were observed, deemed acceptable for pharmaceutical development.
• The CNN successfully predicted the composition of test samples with varying particle types.

Conclusions:

• The developed CNN provides a fast and reliable method for classifying common SvPs encountered in FIM analysis.
• This AI-driven approach enhances the efficiency and accuracy of biopharmaceutical quality control processes.
• The CNN is suitable for routine application in pharmaceutical development, aiding in the identification of particle-related issues.