Exploring machine learning approaches for transcriptome-based diagnosis and subgrouping of multiple system atrophy✰

Abstract

Multiple system atrophy (MSA) is a rare adult-onset neurodegenerative disease, where the cause of the disease is not well understood yet. This disease affects the nervous system and then gradually damages the nerve cells in the brain. The consequences of the disease are very serious such as affecting the body balance, body movements and involuntary nervous system, which are closely connected to many crucial body functions including breathing, digestion and bladder control. Even though this disease starts between the age of 50 and 60, it can begin any time after 30 years. This disease can be presented as two main subtypes, one is predominating parkinsonian called as MSA-parkinsonian (MSA-P) and the other one is predominating cerebellar ataxia called as MSA-cerebellar (MSA-C). For such a complicated disease, on-time diagnosis and subgrouping is very crucial. However, there are no specific tests or diagnosis methods so far for the on time identification. Hence, this study checks the possibility of using transcriptome data of MSA patients in the diagnosis and subgrouping of MSA. Mutual information is used in the feature selection and the selected features are used with three different machine learning algorithms to select the model with the highest accuracy. Results show that random forest classifier performs comparably better in both of the classifications with the accuracy value of 0.8 ± 0.20 for subgrouping and 0.86 ± 0.16 for diagnosis.

Keywords

• Multiple system atrophy (MSA)
• MSA-parkinsonian (MSA-P)
• MSA-cerebellar (MSA-C)
• Transcriptome data
• Classification models
• Machine learning prediction