الفهرس 
الغلافOnly 14 pages are availabe for public view
 |  | from | 113 |  |  |
| | | from | 113 | | |
Abstract
ABSTRACT
The Huntington disease (HD) is a severe disease with complex pathological mechanisms as it belongs to the neurodegenerative diseases. The insufficient understanding concerning the mechanics of HD is a significant barrier to therapeutic development against it. Recent studies focus on defining differentially expressed genes that are regarded as therapeutic targets at the early stages of disease progression. This objective is approached by studying physiopathology mechanisms under abnormal behavior using biomedical and omics data.
Due to the disease complexity, the process of predicting differentially expressed genes related to it with classical approaches for feature selection (FS) is not very effective. Recent studies have shown that wrapper methods perform well in the feature selection for high-dimensional data; however, they are time-consuming. Therefore, this study proposes an automated system based on two hybrid feature selection methods to reveal the most significant genes related to complex diseases in an enhanced manner. The first one engaged the empirical Bays t statistic test (E-Bays) with the genetic algorithm (GA) using two different fitness functions: K-Nearest Neighbor (KNN) and K-Means. While the second method engaged the E-Bays with the BORUTA algorithm. The two proposed methods are applied to Affymetrix microarray data related to Huntington’s disease, that composes samples from 4 different human brain regions. A some of 235 disease-associated genes are revealed as biomarkers.
Two processes were adopted to evaluate the proposed FS methods: namely classification accuracy measurement and gene ontology (GO) enrichment analysis. Classification accuracy measurement is applied to assess the importance of the revealed biomarkers using three different classifiers. In comparison to the other FS techniques employed, GA and BORUTA present the best outcomes, according to the classification findings. The maximum classification accuracy achieved reached 100% using GA with KNN fitness function, while BORUTA maximum classification accuracy was 96.875%. The classification accuracy of both recommended FS techniques fluctuates just little across trials, indicating that the methods are both practical and stable. Even though BORUTA’s experiments had large number of common biomarkers of the outcomes among trails, which demonstrate the stability of BORUTA over GA. GO enrichment analysis was performed too on the nominated biomarkers, which revealed a direct link to HD symptoms in the early stages of the disease.
Key Words:
Bioinformatics, Huntington’s disease, Feature selection, computational biology, Genetic algorithm, and BORUTA.
.