2015/02/12 by Manjari Narayan, Narayan, Manjari, Genevera I. Allen +4 · 1 citation
Biochemistry, Genetics and Molecular Biology · Neuroscience · #Applications (stat.AP) #FOS: Biological sciences #FOS: Computer and information sciences #Functional Brain Connectivity Studies #Gene Regulatory Network Analysis #Gene expression and cancer classification #Methodology (stat.ME) #Quantitative Methods (q-bio.QM)
paper · pdf · doi:10.48550/arxiv.1502.03853
openalex publication_date 2015/02/12 · openalex created_date 2022/10/02 · openalex updated_date 2026/07/28
Gaussian Graphical Models (GGM) are popularly used in neuroimaging studies\nbased on fMRI, EEG or MEG to estimate functional connectivity, or relationships\nbetween remote brain regions. In multi-subject studies, scientists seek to\nidentify the functional brain connections that are different between two groups\nof subjects, i.e. connections present in a diseased group but absent in\ncontrols or vice versa. This amounts to conducting two-sample large scale\ninference over network edges post graphical model selection, a novel problem we\ncall Population Post Selection Inference. Current approaches to this problem\ninclude estimating a network for each subject, and then assuming the subject\nnetworks are fixed, conducting two-sample inference for each edge. These\napproaches, however, fail to account for the variability associated with\nestimating each subject's graph, thus resulting in high numbers of false\npositives and low statistical power. By using resampling and random\npenalization to estimate the post selection variability together with proper\nrandom effects test statistics, we develop a new procedure we call R3 that\nsolves these problems. Through simulation studies we show that R3 offers\nmajor improvements over current approaches in terms of error control and\nstatistical power. We apply our method to identify functional connections\npresent or absent in autistic subjects using the ABIDE multi-subject fMRI\nstudy.\n