摘 要
药物相互作用预测对于合理用药和药物研发至关重要，传统方法依赖实验测定，耗时且成本高昂。本研究旨在开发基于人工智能的药物相互作用预测系统，以提高预测效率和准确性。首先构建了包含药物化学结构、药理特性等多源异构数据的大型数据库，并采用深度学习算法中的图神经网络模型对药物分子结构进行表征，充分挖掘药物分子内部原子间复杂关系。同时引入注意力机制，使模型能够聚焦于药物相互作用的关键特征区域，此为创新点之一。在模型训练过程中，利用迁移学习策略，借助已有的大规模预训练模型参数初始化，减少所需训练样本量并加速收敛过程，这是主要贡献之一。通过与多种传统机器学习算法及现有主流预测模型对比实验表明，所开发系统在准确率、召回率等评价指标上均有显著提升，证明其有效性和优越性。该系统可为临床合理用药提供科学依据，有助于新药研发过程中早期发现潜在药物相互作用风险，降低研发成本和周期。
关键词：药物相互作用预测；人工智能；图神经网络；注意力机制；迁移学习
Abstract
Drug interaction prediction is critical for rational drug use and pharmaceutical development, with traditional methods relying on experimental determination that are time-consuming and costly. This study aims to develop an artificial intelligence-based drug interaction prediction system to enhance prediction efficiency and accuracy. A large-scale database encompassing multi-source heterogeneous data, including drug chemical structures and pharmacological properties, was constructed. The graph neural network model from deep learning algorithms was employed to characterize drug molecular structures, thoroughly exploring the complex relationships between atoms within drug molecules. Additionally, an attention mechanism was introduced, enabling the model to focus on key feature regions crucial for drug interactions, which represents one of the innovations. During the model training process, transfer learning strategies were utilized by initializing with parameters from existing large-scale pre-trained models, reducing the required number of training samples and accelerating the convergence process, serving as one of the main contributions. Comparative experiments with various traditional machine learning algorithms and current mainstream prediction models demonstrated significant improvements in evaluation metrics such as accuracy and recall, proving its effectiveness and superiority. This system can provide a scientific basis for clinical rational drug use and help identify potential drug interaction risks early in the new drug development process, thereby reducing research and development costs and duration.
Keywords：Drug Interaction Prediction；Artificial Intelligence；Graph Neural Network；Attention Mechanism；Transfer Learning
目 录
摘 要 I
Abstract II
引 言 1
第一章 药物相互作用研究现状 2
1.1 药物相互作用的重要性 2
1.2 传统预测方法的局限性 2
1.3 人工智能技术的应用前景 3
第二章 系统架构设计与实现 5
2.1 数据获取与预处理 5
2.2 模型选择与构建 5
2.3 系统性能评估指标 6
第三章 关键技术与算法优化 7
3.1 特征提取与表示学习 7
3.2 深度学习模型改进 7
3.3 多模态数据融合方法 8
第四章 实验验证与结果分析 10
4.1 实验设计与数据集构建 10
4.2 预测性能对比分析 10
4.3 系统应用案例展示 11
结 论 13
参考文献 14
致 谢 15
