Application of transformer model architecture in the new drugs design
Abstract
The article proposes to use the artificial intelligence method, based on the technology of natural language modeling, namely the transformer model, to design new molecules. Language modeling allows the generation of molecules that have specific properties similar to molecules existing in the training set. Furthermore, the transformer model allows determining the appropriate context between the various elements formed in the molecular sequence. In this study, the molecules will be presented as SMILES, and five non-canonical pairs of each molecule will be created and trained in a transformer. Finally, the derived model is used to generate new molecules. The source code and attachments needed to teach the Transformer model are available at https://github.com/Mr-Shoxrux/Transformer-modeli/blob/main/Transformers-model.ipynb