No abstract

No abstract

Accurate and absolute quantification of individual peptides in complex mixtures is a challenge not easily overcome. A potential solution is the use of quantitative mass spectrometry (MS) based methods, however, current state of the art requires foreground knowledge and isotopically labeled standards for each peptide to be accurately quantified. This increases analytical expenses, time consumption, and labor, limiting the number of peptides that can be quantified. A key step in developing less restrictive label-free quantitative peptidomics methods is understanding of the physicochemical properties of peptides that influence the MS response. In this work, a deep learning model was developed to identify the most relevant physicochemical properties based on repository MS data from equimolar peptide pools. Using an autoencoder with attention mechanism and correlating attention weights with corresponding physicochemical property indices from AAindex1, we were able to obtain insight on the properties governing the peptide-level MS1 response. These properties can be grouped in three main categories related to peptide hydrophobicity, charge, and structural propensities. Moreover, we present a model for predicting the MS1 intensity output based solely on peptide sequence input. Using a refined training dataset, the model predicted log-transformed peptide MS1 intensities with an average error of 11%.