Potential SARS-CoV-2 protease Mpro inhibitors: repurposing FDA-approved drugs.

Using as a template the crystal structure of the SARS-CoV-2 main protease, we developed a pharmacophore model of functional centers of the protease inhibitor-binding pocket. With this model, we conducted data mining of the conformational database of FDA-approved drugs. This search brought 64 compounds that can be potential inhibitors of the SARS-CoV-2 protease. The conformations of these compounds undergone 3D fingerprint similarity clusterization. Then we conducted docking of possible conformers of these drugs to the binding pocket of the protease. We also conducted the same docking of random compounds. Free energies of the docking interaction for the selected compounds were clearly lower than random compounds. Three of the selected compounds were carfilzomib, cyclosporine A, and azithromycin-the drugs that already are tested for COVID-19 treatment. Among the selected compounds are two HIV protease inhibitors and two hepatitis C protease inhibitors. We recommend testing of the selected compounds for treatment of COVID-19.

Deep-learning- and pharmacophore-based prediction of RAGE inhibitors.

The receptor for advanced glycation end products (RAGE) has been identified as a therapeutic target in a host of pathological diseases, including Alzheimer's disease. RAGE is a target with no crystallographic data on inhibitors in complex with RAGE, multiple hypothesized binding modes, and small amounts of activity data. The main objective of this study was to demonstrate the efficacy of deep-learning (DL) techniques on small bioactivity datasets, and to identify candidate inhibitors of RAGE. We applied transfer learning in the form of a semi-supervised molecular representation in order to address small dataset problems. To validate the candidate inhibitors, we examined them using more computationally expensive pharmacophore-modeling and docking techniques. We created a strong classifier of RAGE activity, producing 79 candidate inhibitors. These candidates agreed with docking models and were shown to have no significant statistical difference from pharmacophore-based results. The transfer-learning techniques used allow DL to generalize chemical features from small bioactivity datasets to a broader library of compounds with high accuracy. Furthermore, the DL model is able to handle multiple binding modes without explicit instructions. Our results demonstrate the potential of a broad family of DL techniques on bioactivity predictions.

The challenges of generalizability in artificial intelligence for ADME/Tox endpoint and activity prediction.

INTRODUCTION: Artificial intelligence (AI) has seen a massive resurgence in recent years with wide successes in computer vision, natural language processing, and games. The similar creation of robust and accurate AI models for ADME/Tox endpoint and activity prediction would be revolutionary to drug discovery pipelines. There have been numerous demonstrations of successful applications, but a key challenge remains: how generalizable are these predictive models? AREAS COVERED: The authors present a summary of current promising components of AI models in the context of early drug discovery where ADME/Tox endpoint and activity prediction is the main driver of the iterative drug design process. Following that is a review of applicability domains and dataset construction considerations which determine generalizability bottlenecks for AI deployment. Further reviewed is the role of promising learning frameworks - multitask, transfer, and meta learning - which leverage auxiliary data to overcome issues of generalizability. EXPERT OPINION: The authors conclude that the most promising direction toward integrating reliable and informative AI models into the drug discovery pipeline is a conjunction of learned feature representations, deep learning, and novel learning frameworks. Such a solution would address the sparse and incomplete datasets that are available for key endpoints related to drug discovery.