Advancing Reversed-Phase Chromatography Analytics of Influenza Vaccines Using Machine Learning Approaches on a Diverse Range of Antigens and Formulations

Introduction

• Reversed-phase chromatography is a vital tool in influenza vaccine analysis, yet optimising methods for difficult separations – like strain co-elution or excipient interference – has traditionally depended on manual interpretation and analyst expertise. This reliance creates bottlenecks in swift vaccine development and emergency responses.

• Although many chromatographic databases exist across laboratories, the large volume of data complicates identifying relevant patterns and optimising methods without proper annotation and automated classification systems.

• This pioneering study shows how machine learning can revolutionise vaccine quality assessment by automatically classifying chromatographic profiles, differentiating high-quality separations from problematic ones such as low resolution, co-elution, or signal absence.

• By employing data augmentation and multiple supervised learning algorithms, the research achieved over 95% accuracy in chromatogram classification, offering a scalable, objective approach that reduces manual interpretation and speeds up method development – especially vital for pandemic preparedness.

Key Learning Outcomes

• Discover how reversed-phase chromatography can be quickly tailored to various influenza vaccine formulations, modalities, and strains through strategic adjustments like choosing the right column, preparing samples, and optimising gradients.

• Explore practical machine learning workflows for managing imbalanced chromatographic datasets, highlighting why data augmentation techniques- such as Gaussian noise, intensity scaling, time shifting, and peak spiking- perform better than traditional oversampling or undersampling in real-world analytical settings.

• Understand how Principal Component Analysis (PCA) can reduce dimensionality by extracting significant features from complete chromatograms, preserving 95% of variance and facilitating effective classifier training.

• Recognise the potential for computer-assisted methods in vaccine quality control and how ML-based classification can be expanded beyond influenza to detect contaminants, identify baseline drift, and recognise ghost peaks.

About

Dr. Huixin Lu (Lulu) works as a Research Scientist in the Regulatory Research Division at Health Canada. She focuses on developing innovative methods to assess the physicochemical quality of complex biotherapeutics, including peptides, vaccines, and gene therapies. Her recent work involves using machine learning for reversed-phase chromatography analysis, marking a notable progress in vaccine quality assessment and providing scalable solutions for quick method optimisation during pandemics. Dr. Lu actively collaborates with international organisations to enhance the safety and effectiveness of biotherapeutics through global standardisation of analytical techniques and reference materials. Prior to her role at Health Canada, she gained industry experience as a Research Scientist at a CDMO, working on biomanufacturing processes and analytical characterisation of various biotherapeutic products. Her research integrates advanced data science with practical regulatory science, preparing analytical laboratories to face future quality control challenges.