Multiblock variance partitioning: a new approach for comparing variation in multiple data blocks.

Area of Science:

• Chemometrics
• Multivariate Data Analysis
• Analytical Chemistry

Background:

• Multiple analytical techniques are often used for sample characterization.
• Data from different techniques can contain redundant information.
• Existing methods may not efficiently partition information across datasets.

Purpose of the Study:

• To present a novel method, multiblock variance partitioning (MVP), for comparing information across data blocks.
• To introduce quantitative measures for unique and common variation.
• To demonstrate the utility of MVP in diverse scientific fields.

Main Methods:

• Utilizing partial least squares (PLS) models to relate predictor blocks to a common response.
• Calculating unique variation present in a single data block.
• Calculating common variation shared across multiple data blocks.

Main Results:

• MVP provides a distinct perspective on data block information compared to traditional multiblock analysis.
• The method effectively quantifies unique and common variance components.
• Demonstrated applications in process control, sensory analysis, and preprocessing evaluation.

Conclusions:

• MVP is a valuable tool for understanding information distribution across multiple analytical datasets.
• The method offers complementary insights, particularly when used alongside interval PLS analysis.
• MVP has broad applicability in analyzing complex chemical and sensory data.