We strive to address one of the world’s most complex problems: drug discovery.
New drug development cost: $2.6 billion. Success rate: below 10%.
One of the first phases of drug discovery for a chosen target concerns high-throughput screening. Machine learning has the potential to substantially accelerate the drug discovery lifecycle by performing virtual screening of large databases of candidate drug molecules against a specific target a protein associated with some known disease. At the same time current approaches for structure-based virtual screening are slow.
Accurate and ultra-fast.
We leverage the power of DL, avoid physics-based simulations, and achieve state-of-the-art performance with orders of magnitude faster screening times: 10^4 times faster compared to competitors like VINA, GNINA. This renders our approach suitable for virtually screening of vast compound databases, scaling up to billions of molecules per day.
Methodology
DeNViS is based on state-of-the-art machine learning and geometric deep learning technologies:
- Geometric deep learning algorithmic backbone with Graph Neural Networks (GNNs)
- 3D protein-surface and molecular-graph modelling for the ligand with chemical & geometric features
- Multi-modal fusion of such multiple representations levels to predict a binding affinity score.
- Ensembling multiple models for the final binding affinity prediction.
We evaluated DeNViS in multiple benchmark datasets with impressive results.
More information to be published soon.
VALUE PROPOSITION
Our unique proposition is a Virtual Screening as a Service (VSaaS) solution.
We propose a service for fast virtual screening, suitable for massive databases with minimal requirements.
DeNViS can be employed for tasks within early-stage drug discovery: drug repurposing, repositioning and potentially other applications within the drug discovery pipeline.
“Deeplab’s tech and outputs ranked among the top-20 worldwide"
In JEDI Grand Challenge we virtually screened 1 Billion molecules against 3 proteins of Sars-Cov2. Currently our submission is evaluated in the wet-lab together with the top-20, out of ~138 teams.
Stand-by for upcoming results which are to be published soon !
Why us?
104x speed-up compared to state-of-the-art docking-based approaches such as gina, vina etc. DeNViS can be employed to screen billions of compounds in a matter of hours.
State-of-the-art performance employing algorithms in geometric deep learning, outperforming docking-based approaches in binding affinity prediction tasks evaluated in multiple benchmarks.
What our collaborators say about DeNViS
"Leveraging its extensive deep learning expertise, DeepLab's virtual screening approach is a novel, high throughput methodology in binding affinity elucidation. Fueled by state-of-the-art machine learning methods, this approach is propelling drug-repurposing into an exciting future."
Stefanos Leptidis
Postdoctoral Researcher in Cardiac Transcriptomics & Data Analysis
"DeepLab enables you to reimagine how healthcare is delivered. By using state of the art scientific approaches to drug discovery, we apply ideas that are novel and life-changing!"
Efthymios Vogiatzis
HealthTech Business & Market Analysis Consultant
Publication:
- A community effort to discover small molecule SARS-CoV-2 inhibitors, (under submission) 2023, [preprint]
- Α. Krasoulis, N. Antonopoulos, V. Pitsikalis, S. Theodorakis, "DENVIS: Scalable and high- throughput virtual screening using graph neural networks with atomic and surface protein pocket features", Journal of Chemical Information and Modeling (in press) 2022, [doi, preprint]