Generative deep learning for drug discovery
Rugved Palodkar, Ahaan Kotian, Smit Malde, Prasanna Shete · 2026
The drug discovery process for complex neurological conditions like Alzheimer’s is extremely complex and time consuming. One of the main reasons for this is the size of the number of drugs like molecules which are synthetically accessible. This research aims to accelerate de novo drug design by leveraging generative deep learning models, like LSTMs and RNNs, to generate novel molecular structures inhibiting specific biomarkers for diseases, such as BACE-1 for Alzheimer’s disease. Trained on the dataset sourced from ChEMBL, the model employs transfer learning to generate novel molecular structures with properties akin to biomarker inhibitor. The improved LSTM model achieved an accuracy of 85.3% from which six valid molecules were sampled. These are validated with RDKit, a cheminformatics toolkit, supporting the discovery of disease specific drug compounds.