Title: A Rapid Screening Strategy for High-Dimensional Outcomes and Predictors

Modeling interactions within multimodal, high-dimensional datasets presents significant challenges, primarily due to the ultra-high dimensionality, intricate dependence structures, and substantial noise levels involved. While screening techniques are widely used to mitigate dimensionality, conventional methods typically focus solely on shrinking the predictor space, leaving all outcome variables intact. In cross-modal contexts, where distinct outcomes often identify different subsets of predictors, the resulting union of selected features remains extensive, and the dimensionality of the response variables is unaltered. Consequently, the practical advantages of such screening are limited, leading to excessive computational demands and reduced interpretability.

To overcome these constraints, we introduce Graph Independence Dual Screening (GIDS), a novel framework designed to simultaneously reduce the dimensionality of both response variables and predictors. We have developed computationally efficient algorithms that streamline downstream selection processes, thereby enhancing both accuracy and scalability, while also providing robust theoretical support. Comprehensive simulation studies confirm that GIDS surpasses existing methodologies that limit their screening to predictors alone.

The practical utility of GIDS was demonstrated through an analysis of the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. This study examined interactions between 865,353 genome-wide DNA methylation sites and 49,386 transcriptomic variables. GIDS successfully condensed the feature space to approximately 9,000 CpG sites and 2,000 transcripts. This reduction revealed blockwise interaction structures, identifying clusters of CpG sites and gene transcripts with strong associations. These results not only enhance computational efficiency but also provide interpretable biological insights, shedding light on the coordinated regulatory mechanisms associated with Alzheimer's disease.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC