Utilisation of Machine Learning Approaches Improves RNA-Seq Transcriptome Analyses in Alzheimer’s Disease Brain

Alzheimer’s disease (AD) is a neurodegenerative disorder that progressively deteriorates a person’s memory, as well as their ability to think and move. It has been reported to be the most common cause of dementia. Alterations in gene expression have been increasingly recognised as key contributors to the onset and progression of AD, driving interest in transcriptomic approaches to better understand the disease at a molecular level. The development of machine learning (ML) approaches in transcriptomics have been rapid in the past decade, and this advancement can be applied to the study of AD transcriptomes. An ML program that enhances the alignment data through filtering out low confidence splice junction reads, Splam, has been developed by Chao et al. (2023). However, this program has not been utilised and assessed in the transcriptomic study of a complex neurological disease such as AD. This study investigates both the transcriptome of AD brain and the potential of an ML program to enhance alignment-stage data quality and influence downstream analyses. Using the Integrative Genomics Viewer, a selection of filtered reads was visualised, uncovering the types of splice junction reads Splam discards to refine the alignment data. From the differential expression (DE) analysis, we found a higher number of DE transcripts using ML-filtered data compared to unfiltered data, potentially unmasking aspects of AD brain DE profile obscured by alignment noise. The gene loci expressing those transcripts were also determined to be more AD-relevant by comparing these findings with external studies, and contribute to more related gene ontology enrichment terms. We identified gene loci expressing transcripts of interest shared between ML-filtered and unfiltered data, as this consistency in detection suggests that these genes are robust candidates for downstream analyses and biomarkers in AD.