Detection of polypharmacy side effects by integrating multiple data sources and convolutional neural networks

Area of Science:

• Pharmacology
• Computational Biology
• Artificial Intelligence

Background:

• Polypharmacy, the use of multiple drugs, is common for complex conditions but increases the risk of harmful drug-drug interactions and side effects.
• Identifying these polypharmacy side effects is a significant challenge in drug development and patient safety.

Purpose of the Study:

• To introduce a novel deep learning method, PSECNN, for predicting polypharmacy side effects.
• To enhance the accuracy of predicting adverse drug reactions arising from drug combinations.

Main Methods:

• PSECNN utilizes a deep learning approach, combining five fundamental drug features: individual side effects, drug-protein interactions, chemical substructures, targets, and enzymes.
• A feature extraction module generates uniform-dimension feature vectors using the Jaccard similarity index, creating unique drug representations.
• Paired drug representative vectors are input into a deep neural network for predicting the probability of side effects.

Main Results:

• The PSECNN method demonstrated superior performance compared to state-of-the-art approaches, achieving up to a 74% improvement in polypharmacy side effect prediction.
• The novel combination of drug features in PSECNN contributes to its enhanced accuracy, particularly for side effects with a molecular basis.

Conclusions:

• PSECNN offers a powerful and accurate deep learning solution for predicting polypharmacy side effects.
• The method's ability to integrate diverse drug features provides a more comprehensive understanding of drug-drug interactions and their consequences.