Metabonomics data classification method based on variational auto-encoder

• Main Authors: YAN MING, ZHAO GUANGYANG, XUE LINGYUN, WEN LUHONG, XU PING, CHEN ANQI, LIU YIAN, HU SHUNDI
• Format: Patent
• Language: Chinese; English
• Published: 30.05.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTEDFOR SPECIFIC APPLICATION FIELDS; PHYSICS

The invention discloses a metabonomics data classification method based on a variational auto-encoder. According to the method, a classification model based on a variational auto-encoder is constructed, and model parameters of a trained variational auto-encoder network are shared with a classification network. According to the method, complex data processing of traditional machine learning is avoided, the tedious step of manual parameter selection in the data preprocessing process is simplified, and the classification precision is well improved through the proposed nonlinear mapping method of deep learning. The KL divergence loss is added, so that the learned hidden layer space feature is close to the standard normal state, the decoding process is more reasonable, and the classification result is better. 本发明公开一种基于变分自编码器的代谢组学数据分类方法。本发明构建基于变分自编码器的分类模型，将训练好的变分自编码器网络的模型参数与分类网络共享。本发明避免了传统机器学习的复杂数据处理，简化了数据预处理过程中人工参数选择的繁琐步骤，提出的深度学习的非线性映射方法，很好的提高了分类精度。本发明通过添加一个KL散度损失，使得学习的隐层空间特征向标准正态靠近，使得解码过程更加合理化，分类结果更好。