Related Experiment Video
Updated: Jun 9, 2026
08:49
Incorporating Target Protein Structure Flexibility and Dynamics in Computational Drug Discovery Using Ensemble-Based Docking Analysis
Published on: June 20, 2025
Deep docking, part 2: an amplified DDU platform for ultra-large virtual screening.
Mohit Pandey1,2, Tanvir Sajed1,2, Ivan Semenov1,2
1Vancouver Prostate Centre, University of British Columbia Vancouver British Columbia Canada acherkasov@prostatecentre.com.
Chemical Science
|June 8, 2026
Summary
Deep Docking Ultra (DDU) accelerates drug discovery by integrating advanced acquisition functions with a molecular large language model (MLLM). This approach significantly reduces computational costs for virtual screening without sacrificing accuracy.
Area of Science:
- Computational chemistry
- Drug discovery
- Artificial intelligence in pharmacology
Background:
- Virtual screening of large chemical libraries presents significant computational challenges.
- Existing methods like Deep Docking struggle with billion-entry scale libraries.
- Need for more efficient computational approaches in structure-based drug design.
Purpose of the Study:
- To introduce Deep Docking Ultra (DDU), an enhanced approach for accelerating virtual screening.
- To improve the accuracy and reduce computational costs of docking score emulations.
- To benchmark DDU performance against existing methods and brute-force docking.
Main Methods:
- Integration of advanced acquisition functions with a pre-trained molecular large language model (MLLM).
- Systematic benchmarking through 384 virtual screening experiments across 12 diverse protein targets.
- Optimization of DDU configurations for maximum computational efficiency.
Main Results:
- DDU significantly reduces computational costs, up to 45-fold compared to Deep Docking and 28,500-fold compared to brute-force docking.
- Maintained predictive accuracy while achieving substantial computational savings.
- Successfully screened 10.1 billion ligands against phosphoglycerate kinase 2 in 10 days with a high enrichment factor (12,000).
Conclusions:
- DDU offers a highly efficient and accurate solution for large-scale virtual screening.
- The integration of MLLMs and advanced acquisition functions represents a significant advancement in computational drug discovery.
- DDU enables rapid screening of massive chemical spaces, facilitating faster identification of potential drug candidates.
