Protein classification by autofluorescence spectral shape analysis using machine learning.

Area of Science:

• Biophysics
• Biochemistry
• Computational Biology

Background:

• Proteins exhibit distinct autofluorescence (AF) spectral shapes based on Tryptophan (Trp) and Tyrosine (Tyr) residue composition and arrangement.
• Visual analysis of AF spectra for precise protein identification is challenging due to the vast diversity of proteins and spectral similarities between different proteins.

Purpose of the Study:

• To develop a practical machine learning (ML) based technology for rapid protein identification using AF spectra.
• To overcome the limitations of visual analysis in distinguishing proteins with similar AF spectral shapes.

Main Methods:

• Recorded AF spectra of fifteen diverse standard proteins excited at ~280 nm.
• Utilized the Minimum-Redundancy-Maximum-Relevance (mRMR) algorithm to select key spectral features.
• Developed and applied multiclass Support Vector Machine (SVM) models with Radial Basis Function (RBF), Polynomial, and Linear kernels for protein classification.

Main Results:

• Achieved high classification accuracies: 99.06% (RBF kernel), 99.03% (Polynomial kernel), and 98.29% (Linear kernel).
• Demonstrated the effectiveness of ML algorithms in differentiating proteins based on subtle AF spectral variations.

Conclusions:

• The proposed ML methodology provides a viable and accurate alternative to existing protein identification techniques.
• Accurate protein identification is crucial for understanding biological functions and disease diagnosis, and this method offers potential improvements.