Proteome-Wide Profiling of the Covalent-Druggable Cysteines with a Structure-Based Deep Graph Learning Network

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Covalent ligands offer advantages like prolonged target engagement and high specificity.
• Identifying suitable covalent binding sites remains a significant bottleneck in drug discovery.
• Previous attempts to find non-covalent inhibitors have often been unsuccessful for certain protein targets.

Purpose of the Study:

• To develop an in silico method for identifying ligandable covalent binding sites in proteins.
• To introduce DeepCoSI, the first structure-based deep graph learning model for this purpose.
• To facilitate the design of novel covalent ligands by predicting cysteine reactivity.

Main Methods:

• Developed DeepCoSI, a deep graph learning model utilizing protein structure information.
• Integrated pocket characterization and cysteine-environment interactions into the model.
• Validated DeepCoSI on external datasets simulating real-world drug discovery scenarios.

Main Results:

• DeepCoSI achieved state-of-the-art performance in predicting ligandable covalent sites.
• The model demonstrated a strong capability to differentiate between suitable and unsuitable binding sites.
• A comprehensive profiling of all cysteine residues in the Protein Data Bank (PDB) was performed.

Conclusions:

• DeepCoSI is a powerful tool for identifying covalent binding sites, advancing covalent ligand discovery.
• The model's predictions are publicly available, supporting future research and development.
• This work addresses a critical need for in silico methods in covalent drug design.