DPSP: a multimodal deep learning framework for polypharmacy side effects prediction.

Area of Science:

• Pharmacology
• Computational Biology
• Artificial Intelligence

Background:

• Unanticipated drug-drug interactions (DDIs) pose significant health risks.
• Identifying polypharmacy's adverse effects is a critical challenge in human health.
• Computational methods for predicting polypharmacy side effects have advanced.

Purpose of the Study:

• To present DPSP, a novel framework for predicting polypharmacy side effects.
• To develop a deep neural network approach for DDI prediction.
• To generate novel drug features for improved DDI identification.

Main Methods:

• Drug information evaluation and feature extraction using Jaccard similarity.
• Generation of novel drug feature vectors by combining similarities.
• Application of a multimodal deep neural network framework for DDI prediction.

Main Results:

• DPSP demonstrated superior performance compared to established methods like GNN-DDI, MSTE, and DNN on benchmark datasets.
• The framework achieved high accuracy across various classification metrics.
• Diverse drug information integration proved effective for DDI adverse effect identification.

Conclusions:

• The DPSP framework offers an effective and efficient solution for predicting polypharmacy side effects.
• Leveraging diverse drug information enhances DDI prediction accuracy.
• The study highlights the potential of deep learning in mitigating DDI risks.