Bioprocess data mining using regularized regression and random forests.

Area of Science:

• Biotechnology
• Data Science

Background:

• Bioprocess development requires data analysis for overview, parameter identification, control direction, and experimental planning.
• Integrating multiple datasets challenges traditional regression models due to linearity assumptions.
• Regularized regression and random forests offer advantages in handling complex bioprocess data.

Purpose of the Study:

• To evaluate the applicability of regularized regression (Lasso) and random forests (RF) in bioprocess data mining.
• To benchmark their performance against multiple linear regression.

Main Methods:

• Applied Lasso and RF to a microbial hydrogen production dataset.
• Included linear, multiplicative, and quadratic terms of variables in modeling.
• Compared model performance with multiple linear regression.

Main Results:

• Multiple linear regression failed with non-linear terms.
• Lasso achieved a 0.69 correlation between observed and predicted yield.
• RF achieved a 0.91 correlation, demonstrating strong predictive power.

Conclusions:

• Both Lasso and RF successfully modeled bioprocess data, outperforming multiple linear regression.
• Random forests were particularly effective at capturing non-linear data patterns.
• These advanced methods are valuable for bioprocess data mining tasks.