Joint and Unique Multiblock Analysis for Integration and Calibration Transfer of NIR Instruments

Area of Science:

• Chemometrics
• Analytical Chemistry
• Spectroscopy

Background:

• Near-infrared (NIR) spectroscopy utilizes calibration models to replace traditional wet-chemistry methods.
• NIR calibration models are often instrument-specific, necessitating calibration-transfer techniques for cross-instrument standardization.
• Understanding similarities and differences between NIR instruments is crucial for effective model transfer.

Purpose of the Study:

• Introduce Joint and Unique Multiblock Analysis (JUMBA) as an end-to-end workflow for simultaneous multi-source data analysis.
• Investigate similarities and differences among various NIR instruments using JUMBA.
• Develop and evaluate a novel JUMBA-based calibration transfer method.

Main Methods:

• Implemented JUMBA for simultaneous analysis of data from multiple NIR instruments.
• Applied JUMBA to analyze NIR spectra of mushroom substrate from five instruments.
• Utilized a publicly available dataset of corn samples measured on multiple NIR instruments.
• Developed a JUMBA transfer method leveraging shared spectral structures across instruments.

Main Results:

• NIR spectra from different instruments share significant underlying structures.
• The JUMBA transfer method demonstrated excellent performance in calibration transfer across datasets.
• JUMBA transfer outperformed existing methods in prediction accuracy and stability.
• JUMBA facilitates the integration of multiple instruments into a single model, enhancing consistency.

Conclusions:

• JUMBA provides a powerful framework for analyzing complementary data sources simultaneously.
• JUMBA-based calibration transfer offers a robust and superior alternative to existing methods for NIR spectroscopy.
• The JUMBA approach enhances consistency and accuracy in multi-instrument NIR calibration models.