Article Preview
Top1. Introduction
Facioscapulohumeral muscular dystrophy (FSHD) is an inherited, autosomal dominant neuromuscular disorder (Tawil, 2008; Lemmers et al., 2012). The term facioscapulohumeral is comprised of the name of muscles of the body i.e. facio (face), scapula (shoulder) and humeral (upper arm). This disease generally affects the upper body and causes the weakness in muscles. It is caused due to the contraction of polymorphic macrosattelite repeat D4Z4 on chromosome 4q35 (Lemmers et al., 2012). According to the FSHD Global Research Foundation, a new estimate says that FSHD affects 1 out of 7500 people and there is no treatment for this disease. The symptoms of the disease are more prone in men than women. For the diagnosis of FSHD, first a neurologist sees the pattern of muscles, which is done using electromyography (EMG). EMG depicts that the person is suffering from muscular dystrophy but not able to differentiate the kind of muscular dystrophy. A physician always prefers the genetic testing for the diagnosis of neuromuscular disorders. This is done by monitoring the expression levels of genes using the microarray technology.
For the genetic diagnosis of FSHD, the blood samples from persons are taken into consideration. In the laboratories, the changes in the chromosomes, DNAs or proteins in those samples are checked. The microarray technology is used to detect these smaller genetic changes. In general, the microarray data of FSHD contains tens of thousands of genes but a very few number of samples. The gene expression data in gene expression matrix is defined using . In a gene expression matrix, each row represents a gene and each column represents a sample. The value represents the expression value of a specific gene c in sample d. It is a very complex task to monitor and understand the activity of such a high number of genes to correctly diagnose a disease on time. The informative genes which are related to the disease are very less. As only a few number of genes have different level of activity under condition of interest while most of the genes exhibit a similar expression profile so they are not relevant to the classification task. Hence, to correctly diagnose the disease there is a need to reduce the number of genes. By using the data reduction techniques, we can increase the performance of classifiers by decreasing its computational burden. In addition to that we can reduce the training time, execution time, cost of classification and the risk of overfitting which will help in the easy diagnosis of the disease.
To achieve higher accuracy for the diagnosis of a disease, feature (gene) selection plays a key role in the classification. Feature selection methods are concerned with selecting the small subset of features from the original set of features. These techniques can be applied to both supervised and unsupervised methods. Former, the supervised methods are divided into three main categories: filter methods, wrapper methods and embedded methods. A filter method generates the subset of features while ignoring the communication with the classifiers and works very fast. A wrapper method generates the subsets of features using a specific classification model by taking the feature dependencies into consideration. An embedded method also includes the interaction with the classification model and they are very specific to the learning algorithm. Latter, unsupervised methods include various types of clustering methods like k-means and hierarchical clustering method.
For the proper diagnosis of a disease, a system should be made which would give the higher classification accuracy after creating a small subset of genes. Earlier, so many expert systems are made by integrating knowledge and computing techniques for the detection and diagnosis of diseases. A review of these expert systems based on the symptoms is done in (Pandey & Mishra, 2009). A comprehensive review of the use of knowledge and intelligent computing techniques in bioinformatics tasks is done in (Anand et al., 2016). These bioinformatics tasks include dimension reduction, classification, gene prediction, protein structure prediction and protein function prediction.