Protein secondary structure from circular dichroism spectroscopy. Combining variable selection principle and cluster...

Area of Science:

• Biophysics
• Biochemistry
• Spectroscopy

Background:

• Circular dichroism (CD) spectroscopy is crucial for determining protein secondary structure.
• Accurate analysis of CD spectra is essential for understanding protein folding and function.
• Existing methods for CD spectral analysis have limitations in prediction accuracy.

Purpose of the Study:

• To compare different approaches for improving protein secondary structure analysis from CD spectra.
• To evaluate the effectiveness of variable selection, cluster analysis, and self-consistent methods.
• To assess the performance of neural networks, ridge regression, and singular value decomposition in CD spectral analysis.

Main Methods:

• Proteins were grouped using cluster analysis based on CD spectral similarity for variable selection.
• The performance of neural networks, ridge regression, and singular value decomposition was evaluated.
• Variable selection, cluster analysis, and the self-consistent method were applied to enhance prediction accuracy.

Main Results:

• Cluster analysis of basis set proteins yielded three distinct clusters, enabling a novel approach to variable selection.
• Neural networks with two hidden layers outperformed those with one hidden layer when combined with variable selection.
• Inclusion of variable selection improved the performance of all three basic analytical methods.

Conclusions:

• Variable selection, particularly through cluster analysis, is a valuable strategy for enhancing protein secondary structure prediction from CD spectra.
• All three evaluated methods (neural networks, ridge regression, SVD) showed improved performance with variable selection.
• The choice of analytical method becomes less critical when effective variable selection strategies are employed.