Transformer-based deep learning for next generation mass spectrometry-based phosphoproteomics

Multiple oncogenes encode protein kinases that are involved in aberrant cellular signaling by phosphorylation and represent targets of a wealth of FDA-approved drugs. Phosphoproteomics by mass spectrometry (MS) provides a global view of cellular protein phosphorylation, making it highly relevant to cancer research. Important for clinical proteomics, a recent MS advancement called data-independent acquisition (DIA) enables the high-throughput generation of quantitative proteomics data in a more comprehensive fashion.

Standard DIA requires a pre-existent library of high-quality MS spectra for molecules to be considered. A challenge for DIA application in phosphoproteomics is the lack of a spectral library for the plethora of possible phosphoprotein modifications. Thus far, project-specific libraries are created, which have limited depth, and require considerable instrument time.

Excitingly, recently developed AI methods can utilize prior MS sequence assignments to predict spectra for previously unobserved molecules, obviating project-specific library generation. Our hypothesis is that this can be applied to phosphoproteomics.

In this project, we will apply state-of-the-art AI technologies to our unique collection of phosphoproteomics data. This combination will change current practice in DIA-MS analysis, and catalyze both cancer signaling research and biomarker and target discovery enterprises to ultimately improve cancer diagnosis and treatment.