An improved partial least-squares regression method for Raman spectroscopy.

Area of Science:

• Chemometrics
• Analytical Chemistry
• Biosensing

Background:

• Partial Least-Squares (PLS) regression is a common analytical technique.
• Backward Variable Selection (BVS) can improve PLS performance.
• Existing methods may have limitations in optimizing variable selection for complex datasets.

Purpose of the Study:

• To introduce an Improved Backward Variable Selection for Partial Least-Squares (IBVSPLS) method.
• To enhance the performance of PLS regression analysis, particularly for biosensing applications.
• To evaluate the effectiveness of the novel IBVSPLS method against existing techniques.

Main Methods:

• Developed a novel selection mechanism for BVSPLS based on sorting weighted regression coefficients.
• Utilized the Root Mean Square Errors of Prediction (RMSEP) criterion to evaluate variable importance iteratively.
• Applied the Improved BVSPLS (IBVSPLS) method to leukemia and heparin datasets.

Main Results:

• The IBVSPLS method demonstrated significant improvements in the limit of detection for Raman biosensing, ranging from 10% to 43% compared to standard PLS.
• IBVSPLS outperformed the jack-knifing method in all comparisons.
• IBVSPLS showed comparable or superior performance to the Genetic Algorithm method.

Conclusions:

• The proposed IBVSPLS method offers a robust and effective approach for variable selection in PLS regression.
• IBVSPLS provides enhanced accuracy and sensitivity in biosensing applications.
• This method represents a valuable advancement in chemometric analysis for complex data.