Advancing Drug Discovery: Receptor.AI's AI-Augmented Structure-Based Prediction Methods
Improving virtual screening accuracy with smart consensus function
Advancing Drug Discovery: Receptor.AI's AI-Augmented Structure-Based Prediction Methods
Improving virtual screening accuracy with smart consensus function
PARTNERSHIP
Announcement
Summary
Receptor.AI introduces AI-augmented structure-based prediction methods to support compound evaluation during secondary screening in virtual drug discovery workflows. The approach combines a drug-target interaction model (3DProtDTA), a fragment-based model (FB-DTI), and docking with AI rescoring to improve the accuracy of binding predictions. A smart consensus function integrates these models, with automated optimization selecting parameters for each case, improving precision in compound ranking and selection.
Full Text
Receptor.AI presents AI-augmented structure-based prediction methods designed to enhance compound evaluation in structure-guided drug discovery workflows. These methods support compound prioritization following the initial narrowing of chemical space. By incorporating structural information from target protein binding pockets, the approach integrates multiple predictive models to support more accurate assessments of compound-target interactions.
Central to this approach are the drug-target interaction model (DTI), the fragment-based drug-target interaction model (FB-DTI), and custom docking with AI rescoring. These methods elevate the specificity and accuracy of evaluating compounds based on docking poses.
Figure 1 highlights the DTI model (3DProtDTA), demonstrating its crucial role in understanding drug and protein interactions.
The FB-DTI model, depicted in Figure 2, evaluates molecular fragments against protein subpockets, enriching the drug discovery process with its fragment-based insights.
Figure 3 illustrates the AI-based rescoring model, used to refine traditional docking results and improve ranking accuracy. As shown in Figure 4, benchmark results indicate that Receptor.AI’s integrated methods achieve higher predictive performance compared to standard docking methods across a range of tasks.
A key component of this system is the smart consensus function, which dynamically integrates outputs from each model. Automated optimization fine-tunes this process, selecting the optimal consensus function and parameters for each specific scenario.
This model integration approach provides a scalable solution for enhancing precision in structure-based virtual screening and supports the identification of compounds with higher therapeutic potential.
