In the pharmaceutical industry, the average new drug development is approximately $2.6 billion, while the success rates are typically 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 (e.g., a protein) associated with some known disease.
We developed a deep learning solution based on state-of-the-art geometric deep learning technology for 3D protein-surface [Gainza 2020] and molecular-graph modelling for the ligand [Hu 2019] and subsequently fused the two modalities to predict a binding affinity score. We evaluated the above solution in various benchmark datasets with impressive results.
We virtually screened billions of ligands as part of the JEDI worldwide Grand Challenge
. Our team currently is among the top-20 finalist teams by which our selected molecules will be experimentally validated in the laboratory for potential drug-treatment for COVID19.
- Gainza, P., Sverrisson, F., Monti, F., Rodolà, E., Boscaini, D., Bronstein, M. M., & Correia, B. E. (2020). Deciphering interaction fingerprints from protein molecular surfaces using geometric deep learning. Nature Methods, 17(2), 184–192.
- Hu, W., Liu, B., Gomes, J., Zitnik, M., Liang, P., Pande, V., & Leskovec, J. (2019). Strategies for Pre-training Graph Neural Networks. 1, 1–5.