Investigating the pathogenesis of Alzheimer’s disease with machine learning

Alzheimer’s disease (AD) affects millions worldwide, but there are no effective treatments. AD is defined by the initial accumulation of amyloid plaques before the formation of neurofibrillary tangles of tau protein leads to neurodegeneration. Multiple clinical trials directed at amyloid have not shown clinical benefits. Further research into the mechanisms of AD is necessary to identify other potential drug targets. Transcriptomic analysis is a powerful approach for studying mechanisms, but the commonly used methods have limitations such as a vulnerability to outliers. Machine learning (ML) algorithms address some of these limitations yet have rarely been applied to AD RNA-sequencing data. Here we apply ML to bulk RNA-seq data from postmortem human brain tissue (Religious orders study and Memory and Aging Project, Mount Sinai Brain Bank) to investigate the mechanisms underlying AD pathology and cognitive decline. The transcripts associated with AD pathology were identified, with LTF predictive of an AD diagnosis and severity of amyloid pathology (validated experimentally). PRTN3 and ADAMTS2 (but not LTF) were the key transcripts associated with the rate of cognitive decline, while neurofibrillary tangles in the anterior cingulate cortex was the key neuropathology. Next, we used ML to find the overlapping pathways between our transcripts of interest and genes from AD genome-wide association studies. Interactions were identified including between LTF and TREM2 while other transcripts (EMP3, SWAP70, WWTR1, VGF) were associated with AD markers and multiple AD risk genes. Finally, given the effect size of APOE status on AD risk, a stratified analysis confirmed the significance of our target genes and anterior cingulate tau pathology in APOE4 carriers while also demonstrating distinct AD mechanisms in APOE4 carriers. Overall, this work has identified new potential therapeutic targets in AD and shown that APOE4 carriers and non-carriers may require targeted interventions