Novel deep learning-based transcriptome data analysis for drug-drug interaction prediction with an application in...

Area of Science:

• Computational biology and bioinformatics
• Pharmacogenomics and drug discovery

Background:

• Drug-drug interactions (DDIs) pose a significant public health concern.
• The LINCS L1000 database offers extensive transcriptome data for thousands of compounds across multiple cell lines.
• The utility of this transcriptome data for developing advanced DDI prediction models remained unexplored.

Purpose of the Study:

• To develop and validate a novel deep learning model for predicting drug-drug interactions (DDIs).
• To leverage the LINCS L1000 transcriptome data for enhanced DDI prediction.
• To utilize known DDIs from DrugBank for model training and validation.

Main Methods:

• Developed a deep learning model integrating a graph convolutional autoencoder network (GCAN) for embedding L1000 transcriptome data.
• Employed a long short-term memory (LSTM) network for the DDI prediction task.
• Trained and validated the model using 89,970 known DDIs from the DrugBank database (version 5.1.4).

Main Results:

• The proposed deep learning model demonstrated superior performance in DDI prediction compared to other machine learning methods.
• A significant number of predicted DDIs were subsequently confirmed in the updated DrugBank database (version 5.1.7).
• Case studies successfully predicted interactions leading to hypoglycemia (sulfonylureas) and lactic acidosis (metformin), highlighting effects on metabolic proteins.

Conclusions:

• The developed deep learning model effectively accelerates the identification of novel drug-drug interactions.
• This computational approach holds potential for supporting clinical research and improving drug co-prescription safety.
• The model's ability to predict DDIs based on transcriptome data offers a promising avenue for future pharmaceutical research.