Using isotopic envelopes and neural decision tree-based in silico fractionation for biomolecule classification

Area of Science:

• Biochemistry
• Computational Biology
• Analytical Chemistry

Background:

• Untargeted mass spectrometry (MS) is crucial for high-throughput biological sample analysis.
• Current untargeted MS workflows struggle with complex samples containing multiple compound classes, requiring tailored data processing.
• Effective analysis of multi-class/component mixtures is essential for comprehensive biological characterization.

Purpose of the Study:

• To develop a novel computational approach for classifying complex biological samples analyzed by MS.
• To enhance the feasibility of analyzing MS data from mixtures containing multiple biomolecule classes.
• To improve the accuracy and efficiency of untargeted MS data processing for multi-class samples.

Main Methods:

• Developed an in silico fractionation (iSF) approach utilizing a neural decision tree for MS data classification.
• Employed supervised binary classifiers to identify polypeptides and lipids.
• Utilized a third supervised network for classifying lipids into eight main sub-categories based on MS isotopic patterns.

Main Results:

• The neural decision tree achieved 100% sensitivity and 100% specificity in classifying polypeptides and lipids.
• The lipid sub-classification network demonstrated 95% sensitivity and 99% specificity.
• Established relationships between chemical properties, MS isotopic envelopes, and analyte classification.

Conclusions:

• The in silico fractionation (iSF) approach effectively classifies biomolecules in complex mixtures using MS data.
• This neural network-based method significantly improves the analysis of multi-class biological samples.
• iSF offers a robust solution for enhancing untargeted mass spectrometry data interpretation.