Unlike traditional pharmacological inhibition, targeted protein degradation (TPD) induces ubiquitin-dependent degradation of proteins. However, it’s poorly understood which/why certain proteins are susceptible to TPD. We developed a machine learning model to answer this question.
Donovan et al. previously mapped kinase degradability based on a library of multi-kinase degraders. Here, by modeling features intrinsic to protein targets, we classified highly- and lowly- degradable kinases, which show similar drug-target engagement, but different degradation.
We found that degradability of the kinome can be well predicted by protein-intrinsic features. The ubiquitination potential is the most informative feature, which is determined by the fraction of lysine residues that have reported ubiquitination events in the PhosphoSitePlus.
The ubiquitination potential likely reflects a protein’s endogenous capacity to be ubiquitinated since the ubiquitination events are from cell lines in the absence of degrader treatment.
Our model (MAPD short for Model-based Analysis of Protein Degradability) shows promising performance in predicting degradable kinases by multi-kinase degraders and previously reported kinase targets of PROTAC compounds.
Furthermore, our model (MAPD) could also identify non-kinase PROTAC targets and degradable transcription factors, suggesting that it is generalizable to proteins-of-interest (POI) outside of the kinome.
Given the robust performance of MAPD, we extended MAPD predictions to the entire proteome to find 964 disease-causing proteins, including 278 cancer genes, that may be tractable to TPD drug development. Our results are available at mapd.cistrome.org.
Since the ubiquitination potential was the most important feature in MAPD, we reasoned that the accessibility of Ub sites to the E2 enzyme might influence the transfer of ubiquitin to the protein target in degrader-mediated ternary complexes.
In comparison to the total number of Ub sites in the structure of the POI, the E2-accessible Ub sites showed a more significant positive association with protein degradability. These results suggest that lysines with detected ubiquitination events are more amenable to TPD.
In Summary, our study demonstrates that ubiquitination potential is highly predictive of protein degradability, and provides the first quantitative model (MAPD) to prioritize tractable targets proteome-wide for developing TPD compounds.
• • •
Missing some Tweet in this thread? You can try to
force a refresh