Advanced spectral modeling for bacterial strains: A MARS-PLS2 approach with Lasso regularization and baseline...

Area of Science:

• Chemometrics and Spectroscopy
• Bioanalytical Spectroscopy
• Machine Learning for Biochemical Analysis

Background:

• Infrared spectroscopic data presents challenges like baseline distortion, high dimensionality, and non-linear correlations, hindering accurate biochemical property prediction.
• Existing methods struggle to effectively address these inherent data characteristics for reliable bacterial analysis.

Purpose of the Study:

• To develop and validate an integrated framework (MARS-PLS2-Lasso) for enhanced chemometric modeling of infrared spectral data.
• To improve the accuracy and interpretability of predicting biochemical properties in bacteria using spectral analysis.
• To identify key spectral regions and biochemical components crucial for bacterial differentiation.

Main Methods:

• Implemented an integrated MARS-PLS2-Lasso framework combining Multivariate Adaptive Regression Splines (MARS), Partial Least Squares 2 (PLS2), and Lasso regularization.
• Evaluated four baseline correction methods, identifying the Wavelet method (sym8, Level 5) as optimal for spectral variation and noise reduction.
• Utilized MARS for non-linear relationships, PLS2 for latent variable extraction, and Lasso for sparsity and model complexity reduction.

Main Results:

• The Wavelet baseline correction achieved high predictive accuracy (RMSE = 0.2846-0.6857, MAE = 0.2371-0.5445, MSE = 0.0810-0.4705) for bacterial spectra.
• Six key functional regions were identified, corresponding to C-Cl, C-O, C=C, C=O, C-H, and O-H/N-H stretching vibrations.
• The MARS-PLS2-Lasso model demonstrated high prediction accuracy, reduced noise, and enhanced interpretability of spectral-biochemical interactions.

Conclusions:

• The integrated MARS-PLS2-Lasso framework offers a robust, interpretable, and chemically consistent approach for high-dimensional infrared spectral modeling.
• Wavelet baseline correction is highly effective for preprocessing infrared spectra in bacterial analysis.
• The study provides a promising method for bacteria modeling, spectral diagnostics, and bio-analytical applications.