ABSTRACT
To uncover the genetic underpinnings of brain disorders, brain imaging genomics usually jointly analyzes genetic variations and imaging measurements. Meanwhile, other biomarkers such as proteomic expressions can also carry valuable complementary information. Therefore, it is necessary yet challenging to investigate the underlying relationships among genetic variations, proteomic expressions, and neuroimaging measurements, which stands a chance of gaining new insights into the pathogenesis of brain disorders. Given multiple types of biomarkers, using sparse multi-view canonical correlation analysis (SMCCA) and its variants to identify the multi-way associations is straightforward. However, due to the gradient domination issue caused by the naive fusion of multiple SCCA objectives, SMCCA is suboptimal. In this paper, we proposed two adaptive SMCCA (AdaSMCCA) methods, i.e. the robustness-aware AdaSMCCA and the uncertainty-aware AdaSMCCA, to analyze the complicated associations among genetic, proteomic, and neuroimaging biomarkers. We also imposed a data-driven feature grouping penalty to the genetic data with aim to uncover the joint inheritance of neighboring genetic variations. An efficient optimization algorithm, which is guaranteed to converge, was provided. Using two state-of-the-art SMCCA as benchmarks, we evaluated robustness-aware AdaSMCCA and uncertainty-aware AdaSMCCA on both synthetic data and real neuroimaging, proteomics, and genetic data. Both proposed methods obtained higher associations and cleaner canonical weight profiles than comparison methods, indicating their promising capability for association identification and feature selection. In addition, the subsequent analysis showed that the identified biomarkers were related to Alzheimer's disease, demonstrating the power of our methods in identifying multi-way bi-multivariate associations among multiple heterogeneous biomarkers.
Subject(s)
Alzheimer Disease , Proteomics , Algorithms , Alzheimer Disease/diagnostic imaging , Alzheimer Disease/genetics , Biomarkers , Brain/diagnostic imaging , Genomics , Humans , Multivariate Analysis , NeuroimagingABSTRACT
BACKGROUND: Late-Onset Alzheimer's Disease (LOAD) is a leading form of dementia. There is no effective cure for LOAD, leaving the treatment efforts to depend on preventive cognitive therapies, which stand to benefit from the timely estimation of the risk of developing the disease. Fortunately, a growing number of Machine Learning methods that are well positioned to address this challenge are becoming available. RESULTS: We conducted systematic comparisons of representative Machine Learning models for predicting LOAD from genetic variation data provided by the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort. Our experimental results demonstrate that the classification performance of the best models tested yielded â¼72% of area under the ROC curve. CONCLUSIONS: Machine learning models are promising alternatives for estimating the genetic risk of LOAD. Systematic machine learning model selection also provides the opportunity to identify new genetic markers potentially associated with the disease.
Subject(s)
Alzheimer Disease/genetics , Age of Onset , Aged , Benchmarking , Cohort Studies , Female , Genomics , Humans , Machine Learning , Male , Neuroimaging/methods , ROC CurveABSTRACT
This study aimed to provide a simple way to approach group differences by independent component analysis when researching functional connectivity changes of resting-state network in brain disorders. We used baseline resting state functional magnetic resonance imaging from the Alzheimer's disease neuroimaging initiative dataset and performed independent component analysis based on different kinds of subject selection, by including two downloaded templates and single-subject independent component analysis method. All conditions were used to calculate the functional connectivity of the default mode network, and to test group differences and evaluate correlation with cognitive measurements and hippocampal volume. The default mode network functional connectivity results most fitting clinical evaluations were from templates based on young healthy subjects and the worst results were from heterogeneous or more severe disease groups or single-subject independent component analysis method. Using independent component analysis network maps derived from normal young subjects to extract all individual functional connectivities provides significant correlations with clinical evaluations.