Identifying Longitudinal Intermediate Phenotypes Between Genotypes and Clinical Score via Exclusive Relationship-Induced Association Analysis in Alzheimer's Disease

As a widely focused topic, brain imaging genetics has achieved great successes in the diagnosis of complex brain disorders. In clinical application, the imaging phenotypes affected via genetic factors will change over time. A clinical score-relevant exclusive relationship-induced multimodality learning (CS-ERMM) framework is proposed for integrating longitudinal neuroimage, genetics, and clinical score data. Specifically, first, the exclusive lasso term is used to construct the exclusive multimodality learning method, which can convey the unique information at a specific time point. The relationship-induced term is then introduced to automatically learn the relatedness among the multiple time-points from data, which explores the association between genotypes and longitudinal imaging phenotypes to facilitate the understanding of the degenerative process. Finally, the clinical score outcomes are integrated into such association model, which discovers longitudinal phenotypic markers associated with the Alzheimer's disease risk single nucleotide polymorphism that are relevant to clinical score outcomes. We also design a proximal alternating optimization strategy to solve the constructed CS-ERMM model. Extensive experimental results on brain imaging genetic data from the Alzheimer's disease neuroimaging initiative dataset have validated that our method outperforms several competing approaches, which achieve strong associations and identify important consistent markers across longitudinal phenotypes related to genetic risk biomarkers for disease interpretation.