IARAI founding director Sepp Hochreiter and his colleagues have used a deep neural network to find new drugs targeting the SARS-CoV-2 virus.
In our recent paper ‘Large-Scale Ligand-Based Virtual Screening for SARS-CoV-2 Inhibitors Using Deep Neural Networks’ we screened one billion molecules from a public drug-discovery database. We obtained a library of 30,000 top-ranked inhibitors of the two proteases critical for the replication of SARS-CoV-2 virus.
We performed ligand-based virtual screening run with ChemAI, a deep neural network trained to predict outcomes of bioassays. ChemAI has been trained on a large data set of diverse molecules from the ChEMBL, ZINC, and PubChem databases. It can predict a variety of bioactivities, and each bioactivity is represented by an output neuron. We used a set of output neurons associated with SARS-CoV inhibition and toxicity to rank the molecules. This is a highly efficient method to filter for drug candidates from very large molecular databases.
We subjected a set of one billion molecules from the ZINC database to screening. To evaluate the antiviral potency, we searched for compounds that can inhibit both 3CLpro and PLpro proteases critical for the replication of SARS-CoV. We ranked the molecules based on their inhibition potential and refined the ranking based on possible clinical toxicity. We thus obtained a custom-designed library of 30,000 antiviral drugs candidates for COVID-19, ranked and prioritized for validation. Check out the full paper for further details.
Artificial Intelligence is the new electricity which is an essential element of systems in this 4.0 industrial revolution which is transforming the way work is executed but the human factor remains key in steering the way technology is harnessed. This particular use case is ineed proving its versatility in the workflow process as the aspect of time is critical in finding the most suitable combination for a working vaccine.