BiBLDR: Bidirectional Behavior Learning for Drug Repositioning

TL;DR

BiBLDR introduces a bidirectional behavior learning framework for drug repositioning, effectively capturing drug-disease interaction patterns and outperforming existing methods, especially in cold-start scenarios.

Contribution

This work presents a novel bidirectional behavior learning strategy that redefines drug repositioning as a sequential learning task, overcoming limitations of graph-based methods.

Findings

Achieves state-of-the-art performance on benchmark datasets.

Demonstrates superior cold-start scenario performance.

Effectively captures drug-disease interaction patterns.

Abstract

Drug repositioning aims to identify potential new indications for existing drugs to reduce the time and financial costs associated with developing new drugs. Most existing deep learning-based drug repositioning methods predominantly utilize graph-based representations. However, graph-based drug repositioning methods struggle to perform effective inference in cold-start scenarios involving novel drugs because of the lack of association information with the diseases. Unlike traditional graph-based approaches, we propose a bidirectional behavior learning strategy for drug repositioning, known as BiBLDR. This innovative framework redefines drug repositioning as a behavior sequential learning task to capture drug-disease interaction patterns. First, we construct bidirectional behavioral sequences based on drug and disease sides. The consideration of bidirectional information ensures a more meticulous and rigorous characterization of the behavioral sequences. Subsequently, we propose a two-stage strategy for drug repositioning. In the first stage, we construct prototype spaces to characterize the representational attributes of drugs and diseases. In the second stage, these refined prototypes and bidirectional behavior sequence data are leveraged to predict potential drug-disease associations. Based on this learning approach, the model can more robustly and precisely capture the interactive relationships between drug and disease features from bidirectional behavioral sequences. Extensive experiments demonstrate that our method achieves state-of-the-art performance on benchmark datasets. Meanwhile, BiBLDR demonstrates significantly superior performance compared to previous methods in cold-start scenarios. Our code is published in https://github.com/Renyeeah/BiBLDR.