‘Drugging the undruggable’ - discovery of unique peptide inhibitors of protein-protein interactions by machine learning and biophysical tools

A fundamental challenge in structural biology is the discovery of modulators binding a selected region on a target protein. This has immense importance for the discovery of protein-protein interaction (PPI) inhibitors. Indeed, the large number of PPIs and their relevance for a broad range of diseases, marks it as an important therapeutic target. However, physio-chemical characteristics and lack of structural data often prevent the application of traditional drug discovery methodologies. Discovery of new peptides targeting a defined protein interface is an attractive approach for the deciphering of PPI structural aspects. The main challenge stems from the exponential number of peptide sequences of length L which is 20^L. This makes the discovery of such peptides nearly impossible experimentally and a time-consuming challenge computationally.

Herein, we are developing synergistic machine learning algorithms integrated with real-time biophysical data. We applied our methodology on the discovery, characterization and optimization of peptides containing the motifs PxIxIT and LxVP known to bind the important protein calcineurin and inhibit its interaction with the transcription factor NFAT. The latter is considered as an imperative T-cell activation switch. We discovered and validated the biding and cellular activity of unique ‘out-of-the-box’ binding sequences, showing potent activity against calcineurin. These novel peptide sequences enable the development of new immune modulators and further study of this important PPI system. Moreover, our approach paves the way for the study of additional unexplored PPI interfaces.