Computer methods and programs in biomedicine · 2026
AGSI: Adaptive group-enhanced strategy for iterative integration of single-cell multi-omics.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Zhang Fanyu, Shang Junliang, Jiang Shoujia, Zhang Xiaohan, Yan Shijia, Sun Yan, Liu Jin-Xing
Аннотация
BACKGROUND AND OBJECTIVE: Single-cell multi-omics data integration is critical for understanding cellular heterogeneity and disease mechanisms. However, current methods face two key limitations: (1) uniform evaluation of cross-modal correspondence across all genes, neglecting the modular organization of biological systems, and (2) static integration strategies that fail to accommodate varying degrees of cell-level heterogeneity. To address these challenges, this study proposes AGSI, an adaptive framework for robust multi-omics integration through co-regulated gene modules and iterative reliability assessment.
METHODS: AGSI employs Latent Dirichlet Allocation to identify co-regulated gene modules and evaluates cross-modal correspondence at the module level. AGSI combines Wasserstein-enhanced similarity metrics with dual reliability modeling to progressively identify and integrate cells with high cross-modal concordance. Adaptive thresholding dynamically adjusts selection criteria throughout the iterative refinement process.
RESULTS: Extensive experiments on multiple datasets including PBMC, SNARE-seq mouse brain, 10x mouse brain, and large-scale human myocardial infarction data demonstrate that AGSI significantly outperforms seven state-of-the-art methods. Notably, AGSI achieves up to 25.6% F1 improvement over its ablation baseline and 11.9% over the best competing method on complex neural datasets, and maintains over 85% accuracy even under 50% data dropout.
CONCLUSIONS: AGSI provides a robust and scalable solution for multi-omics integration that preserves biological interpretability while achieving superior technical performance. AGSI is well suited to biomedical analyses requiring accurate cell type identification. The implementation code is available at https://github.com/CDMBlab/AGSI.
METHODS: AGSI employs Latent Dirichlet Allocation to identify co-regulated gene modules and evaluates cross-modal correspondence at the module level. AGSI combines Wasserstein-enhanced similarity metrics with dual reliability modeling to progressively identify and integrate cells with high cross-modal concordance. Adaptive thresholding dynamically adjusts selection criteria throughout the iterative refinement process.
RESULTS: Extensive experiments on multiple datasets including PBMC, SNARE-seq mouse brain, 10x mouse brain, and large-scale human myocardial infarction data demonstrate that AGSI significantly outperforms seven state-of-the-art methods. Notably, AGSI achieves up to 25.6% F1 improvement over its ablation baseline and 11.9% over the best competing method on complex neural datasets, and maintains over 85% accuracy even under 50% data dropout.
CONCLUSIONS: AGSI provides a robust and scalable solution for multi-omics integration that preserves biological interpretability while achieving superior technical performance. AGSI is well suited to biomedical analyses requiring accurate cell type identification. The implementation code is available at https://github.com/CDMBlab/AGSI.