Guided principal component analysis (GPCA): a simple method for improving detection of a known analyte

Area of Science:

• Medical Spectroscopy
• Biomedical Data Analysis
• Chemometrics

Background:

• Raman spectroscopy shows promise in clinical settings for real-time decisions and disease classification.
• Analyzing complex biological samples with Raman spectroscopy is challenging, limiting AI applications like Convolutional Neural Networks (CNNs) due to lack of ground truth.
• Principal Component Analysis (PCA) is a common but unsupervised method for spectroscopic data reduction, often yielding variable results.

Purpose of the Study:

• To introduce Guided Principal Component Analysis (GPCA) as a novel approach to enhance PCA for medical Raman spectroscopy.
• To improve the robustness and consistency of PCA analysis in complex biological samples.
• To enhance quantification, limit of detection, and reduce Root Mean Square Error (RMSE) in spectroscopic analysis.

Main Methods:

• Developed Guided Principal Component Analysis (GPCA) by incorporating a reference (guiding) spectrum into the PCA dataset.
• Applied GPCA to guide the PCA analysis towards a key target moiety, ensuring a consistent rank.
• Evaluated GPCA's performance in simplifying analysis and increasing the robustness of PCA for heterogeneous biological materials.

Main Results:

• GPCA simplifies the analysis of complex spectroscopic data from biological samples.
• The inclusion of a guiding spectrum ensures a consistent rank for target moieties, increasing PCA robustness.
• GPCA demonstrated improvements in quantification, lower limits of detection, and reduced RMSE compared to standard PCA.

Conclusions:

• GPCA offers a straightforward yet powerful method to guide PCA, overcoming limitations in analyzing complex medical samples.
• This approach enhances the reliability and accuracy of Raman spectroscopy for clinical applications.
• GPCA facilitates the deployment of advanced analytical techniques, including AI, by providing a more consistent and reliable data foundation.