Predicting Peptide Ionization Efficiencies for Electrospray Ionization Mass Spectrometry Using Machine Learning.

Area of Science:

• Proteomics
• Analytical Chemistry
• Biotechnology

Background:

• Label-free mass spectrometry (MS) is crucial for analyzing complex biological systems.
• Accurate quantification in MS relies on predicting ionization efficiency (IE) from peptide physicochemical properties.
• Limited research exists on modeling peptide IE compared to small molecules.

Purpose of the Study:

• To develop predictive models for peptide ionization efficiency (IE) using machine learning.
• To establish accurate label-free quantification workflows for peptides and proteins.
• To improve the quantification of peptide post-translational modifications (PTMs).

Main Methods:

• Compiled a dataset of relative ionization efficiencies (RIEs) for 241 peptides from trypsin digests of monoclonal antibodies (mAbs).
• Computed peptide physiochemical descriptors from their sequences.
• Trained machine learning models, including random forest (RF) and multilayer perceptron (MLP), to predict RIEs.

Main Results:

• For peptides < 20 amino acids, RIEs strongly correlated with molecular weight.
• A random forest (RF) model achieved a 23.9% mean relative error in predicting RIEs.
• A multilayer perceptron (MLP) model reduced mean relative error from 60.5% to 32.0% for larger peptides, outperforming existing methods.

Conclusions:

• Predicting peptide IE from sequence enables accurate label-free quantification.
• The developed models enhance relative protein quantification and PTM analysis.
• This methodology supports the advancement of high-throughput proteomics and biomarker discovery.