HypergraphSynergy

动机药物组合在治疗癌症方面具有较小的毒性和较少的不良反应。然而，由于组合爆炸，协同药物组合的体外筛选既耗时又费力。尽管已经开发了许多用于预测协同药物组合的计算方法，但是药物协同数据中存在的药物组合与细胞系之间的多向关系尚未得到很好的利用。结果我们提出了一种用于预测抗癌药物协同作用的多向关系增强超图表示学习方法，称为超图协同作用。HypergraphSynergy在癌细胞系上制定协同药物组合作为超图，其中药物和细胞系由节点表示，协同药物-细胞系三联体由超边缘表示，并利用药物和细胞系的生化特征作为节点属性。然后，设计了超图神经网络，以从超图中学习药物和细胞系的嵌入并预测药物协同作用。此外，重建药物和细胞系的相似性网络的辅助任务被认为是增强模型的泛化能力。在计算实验中，HypergraphSynergy在两个用于分类和回归任务的基准数据集上都优于其他最新的协同预测方法，并且适用于看不见的药物组合或细胞系。研究表明，超图公式使我们能够捕获和解释药物组合和细胞系之间复杂的多向关系，并且还提供了一个灵活的框架来充分利用各种信息。可用性和实现HypergraphSynergy的源数据和代码可以从https://github.com/liuxuan666/HypergraphSynergy免费下载。补充信息补充数据可在生物信息学在线上获得。

This work is motivated by the fact that anticancer drug combinations exhibit reduced toxicity and fewer adverse reactions during cancer treatment. However, due to the combinatorial explosion, in vitro screening of synergistic drug combinations is both time-consuming and labor-intensive. Although numerous computational methods have been developed for predicting synergistic drug combinations, the multi-relational relationships between drug combinations and cell lines inherent in drug synergy data have not been fully exploited. To address this gap, we propose a multi-relational relationship-enhanced hypergraph representation learning method for predicting anticancer drug synergy, termed HypergraphSynergy. HypergraphSynergy formulates synergistic drug combinations on cancer cell lines as a hypergraph, where drugs and cell lines are represented as nodes, and synergistic drug-cell line triplets are represented as hyperedges, and utilizes the biochemical profiles of drugs and cell lines as node attributes. Then, a hypergraph neural network is designed to learn embeddings of drugs and cell lines from the hypergraph and predict drug synergy. Furthermore, an auxiliary task of reconstructing the similarity networks of drugs and cell lines is incorporated to enhance the generalization ability of the model. In computational experiments, HypergraphSynergy outperforms other state-of-the-art synergy prediction methods on two benchmark datasets for both classification and regression tasks, and is applicable to unseen drug combinations or cell lines. Studies demonstrate that the hypergraph formulation enables us to capture and interpret the complex multi-relational relationships between drug combinations and cell lines, and also provides a flexible framework to fully leverage various types of information. Availability and implementation: The source data and code implementing HypergraphSynergy are freely available for download at https://github.com/liuxuan666/HypergraphSynergy. Supplementary information: Supplementary data are available at Bioinformatics online.

搜集汇总

背景与挑战

背景概述

HypergraphSynergy是一个用于预测抗癌药物协同作用的数据集，基于超图表示学习方法，将药物和细胞系建模为超图节点，利用生化特征学习嵌入以预测协同效果。该数据集在分类和回归任务中表现优异，能有效捕获药物组合与细胞系间的复杂多向关系，并提供了开源代码和数据供研究使用。