一种基于多头注意力机制的关键蛋白质靶点识别方法

A Method for Essential Protein Target Recognition Based on a Multi-Head Attention Mechanism

摘要: 蛋白质是机体功能的重要执行者。特定的一些蛋白质对生物的生存、繁殖和生理调节尤为关键，它们往往成为疾病发生、发展中的重要参与者，这类蛋白质被称为关键蛋白质。因此，在疾病预防和治疗过程中快速找到蛋白质关键靶点是尤为重要的。本文提出了一种借助多头注意力机制来解决这一难题的方法。该方法将蛋白质相互作用(PPI)网络的拓扑特征、基因表达谱特征以及同源特融合，从而构建一个融合PPI网络。进而，我们采用图注意力神经网络(GAT)模型，学习融合PPI网络中节点的特征表示，为了更好的地捕获蛋白质之间的关联关系，我们引入多头注意力机制增强模型的学习效果。最终，通过在DIP酵母蛋白数据集上的训练和测试，实验结果证明了我们的方法相较于传统的基于拓扑的策略具有更高的识别精度。

Abstract: Proteins are important performers of organismal functions. Certain proteins are especially critical for organisms’ survival, reproduction, and physiological regulation. They often play a significant role in the onset and progression of the disease, which are called essential proteins. Therefore, rapidly identifying the essential protein targets is of paramount importance in the prevention and treat-ment of diseases. In this paper, we propose an approach to address this challenge with the help of a multi-head attention mechanism. Our approach uses topological features of protein-protein inter-action (PPI) networks, gene expression profile features, and homology features to construct a fusion PPI network. Further, we employ a graph attention neural network (GAT) model to learn the feature representations of the nodes in the fusion PPI network. To better capture the association relation-ships among proteins, we introduce a multi-head attention mechanism to enhance the learning ef-fect of the model. Finally, through training and testing on DIP yeast protein datasets, the experi-mental results demonstrate the higher recognition accuracy of our method compared to the tradi-tional topology-based strategy.