DeepTTA: a transformer-based model for predicting cancer drug response.

Area of Science:

• Computational biology
• Pharmacogenomics
• Drug discovery

Background:

• Accurate prediction of anti-cancer drug response is crucial for drug design and precision medicine.
• Traditional in vitro validation is time-consuming and costly.
• Pharmacogenomics and computational models offer promising avenues for drug response prediction.

Purpose of the Study:

• To develop a novel deep learning model, DeepTTA, for predicting anti-cancer drug responses.
• To leverage transcriptomic data and chemical drug features for enhanced prediction accuracy.
• To identify potential therapeutic applications for existing anti-cancer drugs.

Main Methods:

• Developed DeepTTA, an end-to-end deep learning model incorporating transformer networks for drug representation and multilayer neural networks for prediction.
• Utilized transcriptomic gene expression data and drug chemical substructures as input features.
• Evaluated model performance using root mean square error, Pearson correlation coefficient, and Spearman's rank correlation coefficient on multiple test sets.

Main Results:

• DeepTTA demonstrated superior performance compared to existing methods on multiple test sets.
• Achieved high accuracy in predicting anti-cancer drug responses.
• Identified bortezomib and dactinomycin as potential therapeutic options for multiple clinical indications.

Conclusions:

• DeepTTA is an effective computational method for predicting anti-cancer drug responses.
• The model shows promise for accelerating cancer drug design and facilitating precision medicine.
• Further research can explore DeepTTA's application in identifying novel drug candidates and repurposing existing drugs.