Integrating deep learning for post-translational modifications crosstalk on Hsp90 and drug binding

Post-translational modification (PTM) of proteins regulates cellular proteostasis by expanding protein functional diversity. This naturally leads to increased proteome complexity as a result of PTM crosstalk. Here, we used the molecular chaperone protein, Heat shock protein-90 (Hsp90), which is subject to a plethora of PTMs, to investigate this concept. Hsp90 is at the hub of proteostasis and cellular signaling networks in cancer and is, therefore, an attractive therapeutic target in cancer. We demonstrated that deletion of histone deacetylase 3 (HDAC3) and histone deacetylase 8 (HDAC8) in human cells led to increased binding of Hsp90 to both ATP and its ATP-competitive inhibitor, Ganetespib. When bound to this inhibitor, Hsp90 from both HDAC3 and HDAC8 knock-out human cells exhibited similar PTMs, mainly phosphorylation and acetylation, and created a common proteomic network signature. We used both a deep-learning artificial intelligence (AI) prediction model and data based on mass spectrometry analysis of Hsp90 isolated from the mammalian cells bound to its drugs to decipher PTM crosstalk. The alignment of data from both methods demonstrates that the deep-learning prediction model offers a highly efficient and rapid approach for deciphering PTM crosstalk on complex proteins such as Hsp90.