Augmentation Models of Stem Cell Culture Data for the Application of Machine Learning

Area of Science:

• Biotechnology
• Cell Biology
• Computational Biology

Background:

• Stem cell bioprocesses for therapeutics are complex and data-demanding.
• Limited data hinders machine learning applications in biomanufacturing.
• Understanding metabolite-CQA relationships is crucial for process optimization.

Purpose of the Study:

• To develop a framework for investigating metabolite-CQA relationships in stem cell cultures.
• To compare different modeling approaches (FP, HD, NODEAM) using data augmentation.
• To identify key metabolites and time points influencing critical quality attributes (CQAs).

Main Methods:

• Implemented a framework combining data augmentation, multivariate regression, and feature importance analysis.
• Utilized first-principles (FP), hybrid (HD), and neural ODE (NODEAM) models.
• Employed temporal feature importance and functional data analysis of variance.

Main Results:

• Data augmentation enhanced biological information across all models.
• Feature importance convergence followed a power law with augmented runs.
• Glucose and lactate showed strong correlations with CQAs during specific time windows.
• The hybrid (HD) model demonstrated superior fidelity and accuracy.

Conclusions:

• The developed framework provides insights into cultivated cell physiology.
• Combining mechanistic and statistical modeling (HD model) enhances interpretability and accuracy.
• This approach is adaptable for various biomanufacturing systems.