Data augmentation and machine learning techniques for control strategy development in bio-polymerization process.

Area of Science:

• Biochemistry
• Organic Chemistry
• Polymer Science

Background:

• Machine learning is increasingly applied in biochemistry, but experiment-based fields like organic chemistry suffer from inadequate data.
• The COVID-19 pandemic exacerbated data scarcity, hindering model performance and control strategy development.
• Small sample sizes pose a significant challenge for developing robust machine learning models in scientific research.

Purpose of the Study:

• To propose a machine learning solution addressing the challenge of small sample sizes in biopolymerization.
• To enhance data availability through augmentation techniques for improved model performance.
• To compare the effectiveness of different machine learning algorithms and data augmentation strategies.

Main Methods:

• Employed variational auto-encoder and generative adversarial network algorithms for data augmentation to prevent overfitting.
• Implemented random forest and artificial neural network algorithms for the regression modeling process.
• Compared the performance of various machine learning models on predicting molecular weight in biopolymerization.

Main Results:

• Data augmentation techniques were proven to effectively enhance the performance of regression models.
• The random forest model, augmented by generative adversarial networks, demonstrated superior predictive performance.
• The best-performing model achieved an R-squared of 0.94 on the training set and 0.74 on the test set for molecular weight prediction.

Conclusions:

• Data augmentation is a crucial strategy for improving machine learning model performance with limited experimental data.
• Generative adversarial networks offer a powerful method for augmenting data in scientific modeling.
• The random forest algorithm, when combined with GAN-based data augmentation, provides an effective solution for biopolymerization molecular weight prediction.