Abstract

The research presented in this thesis aims to improve the understanding and predictability of peptide responses in mass spectrometry (MS), specifically in bottom-up proteomics, serving as a foundation for the future development of a novel method for absolute peptide quantification using deep learning (DL). MS is a central technology for protein and peptide identification and quantification, widely used across various fields due to its flexibility and high-throughput capabilities. However, existing absolute quantification methods, whether label-based or label-free, encounter significant challenges related to cost, accuracy, and flexibility. This thesis explores these challenges by investigating how peptide composition—specifically single amino acids (AAs), their combinations (dimers), and peptide physicochemical properties—affects MS response variability, and by leveraging DL models to predict peptide detectability and abundance.

The thesis is structured around three key studies:

Article I investigates how individual AAs and their physicochemical properties influence MS response. To explore this, DL models with an encoder-decoder architecture and attention mechanism were developed to extract insights from peptide sequences, shedding light on the impact of specific AAs on response variability.

Article II builds on this by analyzing how combinations of two AAs within peptide sequences influence MS response, examining potential synergistic or antagonistic interactions between adjacent AAs. This study provides a deeper understanding of the factors affecting peptide response.

Article III presents a model for predicting tryptic peptide detectability in MS experiments. This model accurately identifies peptides likely to be observed during analysis, making it a valuable tool for designing targeted MS experiments and other potential applications.

In addition to these studies, the thesis also lays the groundwork for a new absolute peptide quantification method using DL. This method shows promise for eventually overcoming the limitations of existing approaches by offering high accuracy and flexibility without the need for heavy isotopic labels. The findings from these studies advance the field of proteomics by contributing to a deeper understanding of peptide behavior in MS, and by introducing innovative DL-based methods for peptide detectability and quantification.

For further details, please contact Naim Abdul Khalek (nakg@bio.aau.dk).

Supervisors

• Professor Reinhard Wimmer, Department of Chemistry and Bioscience, Aalborg University, Denmark.
• Associate Professor Simon Gregersen Echers, Department of Chemistry and Bioscience, Aalborg University, Denmark.

Assessment committee

• Professor Lukas Käll, Department of Gene Technology, KTH Royal Institute of Technology, Sweden.
• Professor Lennart Martens, Department of Biomolecular Medicine, Ghent University, Belgium.
• Associate Professor Morten Kam Dahl Dueholm (chair), Department of Chemistry and Bioscience, Aalborg University, Denmark.