Transformer-Based Model for Drug Design and Therapy Response Prediction

The use of transformer models in drug design is gradually emerging. The models are applied in chemical structure prediction and target discovery. Textual sequences and graphs are commonly used to describe molecules. In this review, we discussed the application of transformers and their advantages in different prediction tasks. We further elaborated on their applications in various aspects of drug discovery, including the SMILES principle, spectrum prediction, drug response prediction, chemical structure prediction, drug-drug interaction and activity prediction, drug target interaction, and protein prediction. We compared the differences between transformer and traditional drug discovery and then presented our proposed resolutions to this challenge. In the future, advancements will lead to the development of more efficient models with superior parameters. Transformer models can incorporate any input. Combining the transformer with different models and algorithms can also enhance the operational performance for multimodal and multitask prediction.