多组学数据和卷积自编码器的癌症分型算法

Cancer Subtyping Algorithm Using Multi-Omics Data and Convolutional Autoencoders

整合多组学数据对癌症患者进行分型,对于提高患者的诊断、治疗和预后效果是至关重要的。传统的统计学方法,例如主成分分析等,对于处理高纬度的多组学数据集的能力有限。为有效整合多组学数据,提出了一种基于卷积神经网络的自编码器框架MCAEI (Multi-Omics Convolutional Autoen-coder Integration)。所提出的卷积自编码器设置了三个卷积层和反卷积层以及一个全连接自编码器来对多组学数据进行压缩和降维,将MCAEI应用于三种癌症并进行了分型工作。此外,所提出的方法与普通、稀疏、降噪自编码器进行比较,实验结果表明MCAEI方法更优。对于得到的最佳生存亚型,还进行了差异基因表达分析和富集通路分析。

Integrating multi-omics data for staging cancer patients is essential to improve patient diagnosis, treatment, and prognosis. However, traditional statistical methods, such as principal component analysis, face limitations when dealing with high-dimensional multi-omics datasets. To effectively integrate multi-omics data, a convolutional neural network-based autoencoder framework, MCAEI (Multi-omics Convolutional Autoencoder Integration), is proposed. The proposed convolutional au-toencoder is composed of three convolutional layers, three corresponding deconvolutional layers, and a fully connected autoencoder. It is utilized to compress and reduce the dimensionality of mul-ti-omics data. The MCAEI method is then applied to three types of cancer for subtype classification. In addition, the proposed method was compared with the normal, sparse, denoising autoencoder. The results demonstrated the superiority of the MCAEI method. For the best survival subtypes ob-tained, differential gene expression analysis and enrichment pathway analysis were also per-formed.