GPU Enhanced Virtual Screening for Novel DPP-4 Inhibitor Discovery
Type 2 diabetes mellitus (T2DM) is a major healthcare problem worldwide. Targeting dipeptidyl peptidase 4 (DPP-4) has become an important approach in the treatment of T2DM. Unfortunately many of the reported inhibitors have some undesirable side effects. Therefore identification of novel selective and safe scaffolds is highly desirable.
Fig. 1. DPP-4 cellular mechanism of action. (Vasquez-Martínez N.; et al. 2026)
Vasquez-Martínez N et al. employed a machine learning-guided computational workflow that performs GPU accelerated docking combined with long-term MD simulations can significantly accelerate lead finding from large chemical libraries.
GPU Acceleration Breaks Throughput Barriers
Performing static and dynamic computational assays for drug discovery against metabolic enzymes has traditionally become computationally expensive when trying to rapidly calculate accurate binding affinity information on upwards of tens of thousands of compounds. This research demonstrates how to leverage a high-performance workflow to work around this limitation.
Uni-Dock HTS against DPP-4
Using the GPU accelerated docking engine Uni-Dock, researchers were able to dock 30,000 bioactive ligands retrieved from PubChem against the crystal structure of human DPP-4. With GPUs capable of millions of parallel threads per drug candidate, screening speeds were orders of magnitude faster than conventional CPU implementations allowing the researchers to rapidly screen and obtain 32 high affinity hits for structural followup.
Molecular Dynamics
Docking scores give a good starting point, but what matters most is if the drug candidate can remain stable bound over time. Running 100 ns MD simulations on their hits allows researchers to simulate the real-world behaviors of drug candidates. The dynamic nature of the simulations allows visualization of flexibility in the protein-ligand complexes as well as confirming that the ligands remain bound in the active site of DPP-4.
Binding Free Energy Calculations
To go even one step further, researchers performed MM-GBSA calculations to acquire a more stringent estimation of their hit binding strength. Physics-based calculations like MM-GBSA allow for better refinement of docking scores. In this study, MM-GBSA calculations helped researchers identify ligands with strong favorable energetics and key salt bridges with Glu205 and Glu206.
Key Research Findings
The integrated AI and computational approach yielded several high-potential leads with superior binding characteristics.
Top Lead Identification
Lead selection compounds possessed better binding affinity compared to reference inhibitors.
Precise Pocket Targeting
Structural analysis confirmed that these compounds effectively occupy the essential pockets, forming critical hydrogen bonds with residues.
Optimized Pharmacokinetics
Virtual filters based on AI prediction models coupled with PAMPA predictions for membrane permeability indicate that the compounds possess good passive diffusibility.
Enhanced Safety Profiles
Pre-clinical toxicity predictions suggest that these novel scaffolds possess a more favorable safety profile compared to existing drugs, reducing the risk of off-target effects.
Fig. 2. Molecular docking interactions between DPP-4 and lead compounds. (Vasquez-Martínez N.; et al. 2026)
Empowering Your Research with Protheragen MedAI
At Protheragen MedAI, we provide the computational infrastructure and expertise required to execute the necessity of high-performance computing in modern drug development.
- AI-powered drug discovery and design
We integrate AI-accelerated docking and deep learning-based scoring to process millions of compounds.
- Molecular dynamics simulation services
Our high-fidelity MD simulations provide deep insights into protein-ligand stability and binding kinetics.
- AI-guided testing
We offer advanced ADMET and permeability prediction services to ensure your leads have optimal drug-like properties.
- Target identification
Utilizing structural bioinformatics to map active sites and identify druggable pockets in complex enzymes.
Looking to target T2DM or any other complex disease? Protheragen MedAI has the accuracy and speed you need to turn targets into clinical candidates. Contact us today and see how our AI solutions can enable your next endeavor.
Reference
- Vasquez-Martínez N.; et al. (2026). GPU-Accelerated Virtual Screening and Molecular Dynamics Simulations for Identification of Novel DPP-4 Inhibitors. ACS Omega. 2025.
Related Services:
