BiBLDR: Bidirectional Behavior Learning for Drug Repositioning.

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence

Background:

• Graph-based deep learning methods have advanced drug repositioning.
• These methods struggle with cold-start scenarios due to reliance on known associations.
• A need exists for robust drug repositioning strategies that handle limited data.

Purpose of the Study:

• To propose BiBLDR, a novel framework for drug repositioning.
• To address the cold-start problem in drug repositioning using behavior sequence learning.
• To improve the prediction of drug-disease associations.

Main Methods:

• Reformulated drug repositioning as a behavior sequence learning task.
• Constructed bidirectional behavioral sequences for drugs and diseases.
• Employed a two-stage strategy involving prototype spaces and sequence data for association prediction.

Main Results:

• BiBLDR effectively handles both drug and disease cold-start scenarios.
• The framework captures hidden pharmacological relationships from bidirectional sequences.
• Achieved state-of-the-art performance on benchmark datasets, outperforming existing methods in cold-start situations.

Conclusions:

• BiBLDR offers a robust and effective solution for drug repositioning, particularly in cold-start scenarios.
• The bidirectional behavior learning strategy enhances feature representation and prediction accuracy.
• This approach significantly advances the field of computational drug discovery.