A structural machine learning approach for rapid prediction of thermodynamically destabilizing tyrosine phosphorylations.

Abstract

Tyrosine phosphorylations are a prominent characteristic of numerous diseases, yet it is challenging to identify potentially (dys)functional phosphorylations among thousands of phospho-proteins. Here, we propose a machine learning method to predict the thermodynamic stability change resulting from tyrosine phosphorylation. Our approach, based on the prediction of phosphomimetic stability (ΔΔG) from structural features, strongly correlates with experimental phosphorylation stability and mutational scanning cDNA proteolysis data (R = 0.55-0.67). We apply our approach to predict the potential destabilizing effects of all 384,858 tyrosine residues from the Alphafold2 database, the PhosphoSitePlus database, and on a pan-cancer phosphoproteomics dataset with 11 cancer subtypes. We predict destabilizing phosphorylations in both oncogenes and tumor suppressors, and ΔΔG values and local protein circuit topology features are able to distinguish phospho-proteins that are known to be dysregulated in cancer. Our approach can enable rapid screening of destabilizing phosphorylations and phosphomimetic mutations.