SANE: A sequence combined attentive network embedding model for COVID-19 drug repositioning

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• COVID-19 presents a significant global health and economic challenge.
• Drug repositioning is a vital strategy to accelerate therapeutic development for COVID-19.
• Existing computational methods for drug repositioning are limited by their focus on single data types.

Purpose of the Study:

• To propose a novel computational model, SANE (sequence combined attentive network embedding), for effective drug repositioning.
• To integrate sequence and network features for improved drug-target identification.
• To enhance the prediction accuracy of potential therapeutic agents for COVID-19.

Main Methods:

• SANE utilizes an encoder-decoder architecture to extract initial node embeddings from drug SMILES and virus sequences.
• Attention-based Depth-First-Search (DFS) is employed to select relevant node features, reducing noise and improving representation learning.
• A bottom-up aggregation strategy and a forward neural network are used for classification and prediction.

Main Results:

• SANE achieved 81.98% accuracy and 0.8961 AUC, outperforming existing baseline methods.
• A case study on COVID-19 demonstrated SANE's predictive power, with 62.5% of top predicted drugs validated.
• The model effectively integrates diverse data sources for robust drug repositioning predictions.

Conclusions:

• SANE is a powerful and effective tool for drug repositioning in the context of COVID-19.
• The model offers a novel perspective for identifying therapeutic candidates by leveraging combined sequence and network information.
• SANE's approach can accelerate the development of treatments for emerging infectious diseases.