One of the greatest challenges in developing new drugs for neurological disorders is the slow, trial-and-error process of testing small molecules. Current drug development takes about 9 years, with costs often exceeding $1.5 billion per approved drug. With an aging global demographic, there is a need to accelerate drug development times to navigate public health challenges brought on by neurodegenerative diseases.
Most drugs are found in nature and then modified rather than being fully developed in a lab, which is why we are using and refining the NP-VAE (Natural Product-oriented Variational Autoencoder) probabilistic machine-learning model developed by Toshiki Ochiai’s team, which is trained on 30,000 compounds from DrugBank and a dataset with natural biologically-active compounds that existing VAE models couldn’t handle (Ochiai et al., 2023).
This model works in two stages: an encoder compresses a molecule from a simplified molecular input line entry system format (SMILES) into an even simpler latent representation. A decoder then uses this latent representation to reconstruct or generate new molecular structures in the same SMILES format as the input.
To demonstrate the model's viability, we focus on Alzheimer’s Disease (AD). This neurodegenerative disorder influences memory, thinking, and behavior and is the 7th leading cause of death in the world. The novelty of NeuroMedix is that we have refined this model and trained it on FDA-approved drugs that target AD. We have created a fully working website where scientists from pharmaceutical companies can enter a lead FDA-approved drug into the search bar to generate new drug candidates that are ready to be synthesized. NeuroMedix is transformative as we can scale it to target other diseases and offer new drug candidates for small pharmaceutical companies to synthesize.
Log in or sign up for Devpost to join the conversation.