IMMO：面向不完整多组学数据的整合分析框架

IMMO: An Integrated Analysis Framework for Incomplete Multi-Omic Data

摘要: 随着多组学数据整合方法的发展，微生物多组学数据在揭示复杂疾病机制方面发挥着越来越重要的作用。然而，在实际研究中，多组学数据普遍存在样本不完全匹配、模态广泛缺失以及高维稀疏性等问题，严重制约了组学信息之间的有效整合。为应对这些挑战，本文提出一种基于动态掩码机制的多组学整合模型(IMMO)，采用联合自编码器架构，引入动态掩码策略，在训练过程中自适应地处理缺失数据，实现对不完全多组学数据的表征学习与数据重构。在炎症性肠病(IBD)和糖尿病数据集上的实验结果表明，IMMO在数据重建和疾病分类任务中表现出良好的性能，且所学习到的潜在特征能够有效捕捉与疾病相关的关键微生物模式。这为不完全多组学数据的整合分析提供了一种稳定、高效且可解释的方案。

Abstract: Microbiome multi-omics data are increasingly valuable for deciphering complex diseases. However, their integration is often hindered by incomplete sample matching, widespread missing modalities, and high-dimensional sparsity. To address these challenges, we propose IMMO (Integration Model for Incomplete Multi-Omics), a joint autoencoder-based framework that incorporates a dynamic masking mechanism to adaptively handle missing data during training. Evaluated on inflammatory bowel disease (IBD) and diabetes cohorts, IMMO demonstrates strong performance in both data reconstruction and disease classification, with latent representations capturing disease-relevant microbial patterns. Our approach offers a robust, efficient, and interpretable solution for integrative analysis of incomplete multi-omics data.