Phosphorylation is an essential component of cellular signaling, and mass spectrometrybased phosphoproteomics enables the global identification and quantification of phosphosites from biological samples. However, the full potential of this powerful technology is hindered by our limited ability to effectively analyse and interpret the resulting data. In fact, only a small fraction of the phosphosites in the PhosphoSitePlus database have been annotated with a function, giving rise to the notion of the "dark phosphoproteome”. We are engaged in several projects that utilize machine learning and deep learning methods to enhance the analysis and interpretation of phosphoproteomic data, with the ultimate goal of illuminating the dark phosphoproteome.
First, despite advancements in phosphoproteomic technologies, the low identification rate of phosphopeptides in data analysis restricts the potential of this technology. We have developed DeepRescore2, a computational workflow that leverages deep learningbased retention time and fragment ion intensity predictions to improve phosphopeptide identification and phosphosite localization [1].
Second, the interpretation of phosphoproteomic findings is impeded by our limited knowledge of the functions, phenotype associations, and regulating enzymes of the phosphosites. We have developed a computational pipeline IDPpub that uses BioBERT to extract phosphorylation sites from biomedical abstracts, facilitating the identification of regulating enzymes and biological functions of the phosphosites [2].
Finally, to elucidate the interplay between genetic variation and phosphorylation, we have devised a deep learning-based method for predicting the impact of coding variants on phosphorylation events. By applying our method to both pathogenic germline variants in human populations and somatic variants in cancer, we have uncovered numerous PTMaltering variants with high confidence. Furthermore, the interpretability of our model aids in connecting altered phosphorylation events to potential kinases, revealing new therapeutic possibilities.
Together, these efforts have improved our ability to analyse and interpret phosphoproteomics data, shedding light on the human phosphoproteome.
Literature:
[1] Yi et al., Mol Cell Proteomics. 2023, 23(2):100707.
[2] Savage et al., Mol Cell Proteomics. 2023, 23(1):100682.
