Putting the AI in DIA - building an ML-driven multi-modification engine

01 October 2023 → 30 September 2026
Research Foundation - Flanders (FWO)
Research disciplines
  • Natural sciences
    • Computational biomodelling and machine learning
    • Development of bioinformatics software, tools and databases
    • Structural bioinformatics and computational proteomics
    • Proteomics
  • Medical and health sciences
    • Medical metabolomics
Data independant acquisition mass spectrometry Deep Learning Post-translational modifications
Project description

Proteins and the modifications they carry are essential in all biological processes of a cell. The high-throughput method to study the protein content of a cell is using liquid chromatography coupled to mass spectrometry and is acquired in a data dependent acquisition (DDA) mode. This DDA approach is limited to the identification and quantification of a small subset of generally selecting the most abundant peptides. Data independent acquisition (DIA) allows for a more comprehensive view of the proteome because it does not put selection criteria on peptides and measures a wider variety of peptides at the same time. However, this makes the data analysis substantially more complex as it needs to be deconvolved. In this project proposal, I will describe a new, library-free DIA search engine that can nevertheless tackle the identification ambiguity problem. I will achieve this reduction of identification ambiguity by predicting even more aspects of peptide behavior than existing tools. In addition to leveraging these new predictive models, I will also optimize the scoring features extracted from these predictions. Moreover, the resulting identification ambiguity reduction should even allow for a more open modification search in DIA data, promising substantial impact on downstream biological analysis.