Applying deep learning and molecular dynamics to identify potential mTOR inhibitors

mTOR inhibition has been identified as a mechanism for extending lifespan in mice, and as a target for aging intervention in humans.

BrightCore is using ChEMBL database for deep learning model training on serine/threonine-protein kinase mTOR target. We compare the performance of deep learning models, including decision trees (e.g. XGBoost) and neural nets (e.g. ResNets), on high-dimensional molecular representations using molecular fingerprints.

We then run the pre-trained models on a test set of chemical spaces, and refine the results through molecular docking simulations to select top candidates for mTOR inhibition. To find novel, structurally different but pharmacologically similar drug candidates, we are utilizing Enamine REAL database, one of the largest chemical spaces with molecules that can be readily synthesized for further in vitro testing.