Identification of Blood-Based Multi-Omics Biomarkers for Alzheimer’s Disease Using Firth’s Logistic Regression

Abstract

Alzheimer’s disease (AD) is a progressive and relentless debilitating neurodegenerative disease. A post-mortem microscopic neuropathological examination of the brain revealed the existence of extracellular β-amyloid plaques and intracellular neurofibrillary tangles. An accurate early diagnosis of AD is difficult because various disorders share the initial symptoms of the disease. Based on system biology, the multi-omics approach captures and integrates information from genomics, transcriptomics, proteomics, cytokinomics, and metabolomics. This study developed an AD prediction model based on the integrated blood-based multi-omics dataset involving 32 AD patients and 15 non-AD subjects. The integrated multi-omics dataset consists of 16 transcript genes, 14 metabolites, and nine cytokines. Due to the complete separation and multicollinearity issues, Firth’s logistic regression model was then developed to predict AD using the principal components. The model revealed 18 potential biomarkers of AD, consisting of seven metabolites, two transcriptomes, and nine cytokines. These potential biomarkers show an upregulated risk in the AD group compared to the non-AD subjects. The possibility of using these biomarkers as early predictors of AD is discussed.

Predicting Alzheimer’s Disease with Multi-Omic Data: A Systematic Review

“…However, there were common themes between some studies which caused the reviewers to rate 6 of the 22 papers with a medium or high risk of bias. The only study with a high risk of bias was Abdullah et al [35], which reported a prediction accuracy of 100% with a dataset of only 47 individuals and no reported means of data validation. The other 5 medium risk of bias studies were the papers of Yu et al, Darst et al, Binder et al, Khullar and Wang, and Francois et al[36].…”

“…The only study with a high risk of bias was Abdullah et al [35], which reported a prediction accuracy of 100% with a dataset of only 47 individuals and no reported means of data validation. The other 5 medium risk of bias studies were the papers of…”

“…The only study with a high risk of bias was Abdullah et al [35], which reported a prediction accuracy of 100% with a dataset of only…”

Classification of Breast Cancer Subtypes using Microarray RNA Expression Data

“…A recent finding of using hybrid anova and lasso methods for feature selection of microarray data and testing on spark environment denoted that RF perform best with 100% in two dataset and 96% in one dataset in a less time consumed [10]. Other applications of machine learning approach particularly on predictive models could be found in various field of studies including intervention and prevention of diabetic retinopathy [11], diagnosis of Alzheimer's disease [12], prediction of dengue outbreaks [13] and heart failure [14], prediction and classification on future PM10 concentrations [15], forecasting reservoir water level [16], forecasting daily sales data [17] and rainfall prediction in flood prone areas [18] and brain MRI image classification for Alzheimer's disease [19].…”