Mining of biological data I: identifying discriminating features via mean hypothesis testing

Area of Science:

• Biotechnology and Bioprocess Engineering
• Data Science in Biology
• Genomics and Systems Biology

Background:

• Bioprocess operations and genomics generate large datasets.
• These datasets often lack detailed information for mechanism identification.
• Underlying biological mechanisms may be reflected in sensor log patterns.

Purpose of the Study:

• To develop a systematic approach for identifying and modeling patterns in historical bioprocess data.
• To enable process or phenotype classification using identified data patterns.
• To extract potentially relevant features prior to model creation by analyzing data structure.

Main Methods:

• Feature identification via mean hypothesis testing on measurements and time windows.
• Exploration of multivariate data homogeneity using PC1 Time Series Clustering (to be detailed in a future paper).
• Focus on analyzing data structure to extract features before model building.

Main Results:

• A method for identifying discriminating features in data using mean hypothesis testing is presented.
• Case studies from industrial fermentations demonstrate the application of the method.
• The approach successfully identifies features for discriminating different classes of process behavior.

Conclusions:

• The developed method provides a systematic approach to pattern identification and feature extraction in bioprocess data.
• This facilitates the classification of process behavior and outcomes.
• The findings support the use of data-driven approaches for enhanced bioprocess monitoring and control.