A compositional data model to predict the isotope distribution for average peptides using a compositional spline...

Area of Science:

• Proteomics
• Computational Chemistry
• Bioinformatics

Background:

• Accurate approximation of peptide isotope distributions is crucial for mass spectrometry-based proteomics.
• Existing methods may not fully account for factors like sulfur content and spectral variability.

Purpose of the Study:

• To propose an updated computational approach for approximating peptide isotope distributions.
• To develop methods for estimating sulfur atom counts from observed isotope distributions.
• To evaluate the performance of the proposed models against experimental data.

Main Methods:

• In-silico cleavage of the UNIPROT database to generate theoretical peptide datasets.
• Application of compositional data modeling with additive log-ratio transformation and penalized spline regression.
• Development of separate models for varying sulfur atom counts (0-5).
• Proposal of three methods for estimating sulfur atoms from observed isotope distributions.
• Evaluation using mean squared error and a modified Pearson's chi-squared goodness-of-fit on UPS2 data.

Main Results:

• The proposed spline models and sulfur prediction methods were evaluated.
• Variability in spectral accuracy (MS1 scans) was found to be a larger contributor to errors than the theoretical isotope distribution approximation.
• The accuracy of predicting sulfur atoms is limited by the measurement accuracy of the mass spectrometry data.

Conclusions:

• The developed computational approach provides a refined method for approximating peptide isotope distributions.
• Spectral accuracy is a critical factor influencing the reliability of peptide identification and quantification in mass spectrometry.
• Further improvements in mass spectrometry measurement accuracy are needed to enhance the prediction of peptide elemental composition, particularly sulfur content.