Neural Network Guided Global Least Squares for Quantitative Biochemical Component Analysis in Raman Spectroscopy

Area of Science:

• Analytical Chemistry
• Biochemistry
• Spectroscopy

Background:

• Raman spectroscopy is a label-free technique valuable in pharmaceuticals, biology, and food analysis.
• Least squares regression and neural networks are common for spectral analysis but have limitations with many parameters or small datasets.
• Quantitative biochemical component analysis in Raman spectroscopy faces challenges with convergence and limited training data.

Purpose of the Study:

• To propose a hybrid algorithm (NN-GMLS) addressing limitations of existing methods for Raman spectral analysis.
• To enhance quantitative biochemical component analysis in Raman spectroscopy, particularly for small or variable datasets.
• To improve the classification accuracy of leukemia cell states using Raman spectroscopy.

Main Methods:

• Developed a hybrid algorithm (NN-GMLS) combining neural networks and global modified least-squares.
• Utilized simulated data from modified experimental spectra for neural network training.
• Validated the NN-GMLS method on surface-enhanced Raman spectroscopy (SERS) and spontaneous Raman spectroscopy datasets.

Main Results:

• NN-GMLS demonstrated superior accuracy in quantitative biochemical analysis compared to standalone methods on K562 leukemia cell data.
• The method showed effectiveness even with potential variations or inaccuracies in reference spectra.
• A 1D ResNet-10 trained on NN-GMLS synthesized data achieved 87.5% accuracy in classifying leukemia cell states.

Conclusions:

• The NN-GMLS hybrid algorithm offers enhanced accuracy for quantitative biochemical analysis in Raman spectroscopy.
• This method is particularly beneficial in scenarios with limited training data, such as microbial and cellular sample analysis.
• NN-GMLS shows promise for advancing spectral analysis and cell classification applications.