计算机科学
数据集成
数据挖掘
数据建模
数据科学
系统集成
理论计算机科学
大数据
信息集成
人工智能
数据分析
作者
Heng Zhang,Zhi-Ping Liu
标识
DOI:10.1109/tbdata.2026.3725910
摘要
Integrating heterogeneous omics data is crucial for comprehending complex biological systems; however, the theoretical underpinnings of its empirical benefits over single-omics methodologies remain ambiguous. To address this gap, we introduce the “More is Better” theorem, which provides a theoretical rationale for the performance enhancements observed through multi-omics data integration. The fundamental intuition is that, within the adaptive weighting framework, the model's hypothesis space can be broadened as additional omics data are integrated, potentially diminishing model risk. Nevertheless, we also highlight a significant constraint: the inclusion of noisy or marginally relevant omics may not enhance performance and could even result in deterioration. This underscores the underlying principle that expanding the hypothesis space does not inherently ensure a reduction in risk. Drawing on this principle, we propose Supervised Adaptive Sparse Canonical Correlation Analysis (SASCCA), a scalable multi-omics integration approach that selectively accentuates informative omics types through adaptive weighting while enforcing sparsity for biomarker selection. We derive theoretical assurances of SASCCA's superiority over single-omics models and present an efficient optimization algorithm with guaranteed convergence. Extensive experiments on real-world datasets corroborate our theoretical predictions and demonstrate the efficacy and robustness of SASCCA.
科研通智能强力驱动
Strongly Powered by AbleSci AI