Predicting Peptide\nIonization Efficiencies for Electrospray\nIonization Mass Spectrometry Using Machine Learning

Mass spectrometry (MS) is inherently an information-rich\ntechnique.\nIn this era of big data, label-free MS quantification for nontargeted\nstudies has gained increasing popularity, especially for complex systems.\nOne of the cornerstones of successful label-free quantification is\nthe predictive modeling of ionization efficiency (IE) based on solutes’\nphysicochemical properties. While many have studied IE modeling for\nsmall molecules, there are limited reports on peptide IEs. In this\nstudy, we leverage the stoichiometric relationship in trypsin digests\nof well-characterized monoclonal antibodies (mAbs) to compile a data\nset of relative ionization efficiencies (RIEs) for 241 peptides. From\neach peptide’s sequence, we computed a set of physiochemical\ndescriptors, which were then used to train machine learning regression\nmodels to predict RIEs. Peptides shorter than 20 amino acids had RIEs\nthat were highly correlated to their molecular weight. A random forest\n(RF) model was able to best predict the RIEs of a test data set with\na mean relative error of 23.9%. For larger peptides, a multilayer\nperceptron (MLP) model improved RIE prediction compared to current\nbest practices, reducing mean relative error from 60.5% to 32.0%.\nFinally, we also show the application of the RF model in label-free\nrelative protein quantification and improving the quantification of\npeptide post-translational modifications (PTMs). This approach to\npredicting peptide IEs from their sequences enables the development\nof accurate label-free quantification workflows for peptide and protein\nanalysis.