AI-ACCELERATED DRUG DISCOVERY
First strategy is based on the identification of small molecules, which bind independently to both of the target proteins. Matching the spatial arrangements of involved pockets from both proteins and aligning them according to the ligand binding poses allows to predict a protein-protein binding interface (PPI).
To find such dual binders we perform an AI pocket prediction for both proteins and run a virtual screening against all identified tentative binding pockets.
The stock chemical spaces or readily available fragment libraries are used at this stage. The molecules, which are predicted to bind to both protein partners are selected and their binding poses and the corresponding pockets are analyzed.
The second strategy involves direct prediction of the PPI by means of dedicated AI models and the protein-protein docking. Obtained structures of the protein-protein complexes are compared with the predictions of the first approach and the consensus structural model of the PPI is derived as well as the tentative sites for the glues.
The production virtual screening using the full power of the Receptor.AI ecosystem is then performed against the tentative glue locations on the PPI and the candidate compounds are selected from the large and diverse combinatorial chemical space.
The complexes of the best compounds with the proteins are subject to the stability assessment by means of AI-enabled MD simulations.
Those compounds, which effectively stabilize the complex are designated as the glue compound candidates. In the next stage the glue candidate compounds’ biological activity is evaluated. The complexes with the most active compounds are subject to structure determination by means of CryoEM to confirm their structure and location of the glues.
The compounds, which pass this stage, are designated as confirmed glues and are subject to subsequent lead discovery and lead optimization stages using the fit-for-targed Receptor.AI workflow.
