NEW YORK — An artificial intelligence-based analysis of epigenetic patterns in blood samples might be able to identify people with Alzheimer's disease, a new study has found.
Alzheimer's disease affects nearly 47 million people around the world but can be difficult to diagnose, particularly in its early stages when therapeutic interventions might have the greatest effect.
"Drugs used in the late stage of the disease do not seem make much difference, so there is a tremendous amount of interest in diagnosis in the early stages of the disease," Khaled Imam, director of geriatric medicine at Beaumont Health and a co-author of the new study, said in a statement. Imam added that "blood is thought to be a desirable way of approaching this. And it would be relatively cheap and minimally invasive as compared to an MRI or spinal tap."
Imam, lead author Ray Bahado-Singh, and their colleagues studied blood samples from two dozen Alzheimer's disease patients and the same number of cognitively health controls for epigenetic differences in their leukocytes. As they reported in PLOS One on Wednesday, artificial intelligence approaches like deep learning were able to accurately predict which samples came from patients with Alzheimer's disease.
Previous studies had suggested that the brain inflammation that occurs in Alzheimer's disease leads to leukocyte production, and that those cells have methylation changes. The researchers performed genome-wide DNA methylation analysis of the blood samples using Illumina's Infinium MethylationEPIC BeadChip array.
Overall, they identified 152 differentially methylated intragenic CpG sites in the two groups.
They also used six artificial intelligences approaches to analyze their dataset, including support vector machine, random forest, and deep learning. Deep learning, the researchers noted, is a branch of machine learning that aims to mimic the neural networks of animal brains.
Each of the AI approaches could predict Alzheimer's disease with high accuracy, yielding areas under the curve of at least 0.93. Deep learning further improved upon that with an area under the curve of 0.99 and a sensitivity and specificity of 97 percent using intragenic markers. Similar results could be reached with intergenic markers, as well. The researchers noted that the addition of conventional clinical predictors or mental state analyses did not further improve performance.
The analysis highlighted a number of genes and pathways known to be disrupted in Alzheimer's disease. Epigenetically altered genes included, for instance, CR1L and CTSV, which are involved in the morphology of the cerebral cortex, as well as S1PR1 and LTB4R, which are involved in inflammatory response. Affected molecular pathways, meanwhile, were involved in brain and neuronal development as well as in brain and cardiovascular function, a finding that underscored the relationship between cardiovascular function and Alzheimer's disease.
The analysis suggested that artificial intelligence approaches like deep learning, applied to epigenetic data from blood, could identify people who may have Alzheimer's disease in the earlier stages of the disease and could inform treatment efforts.
"We found that the genetic analysis accurately predicted the absence or presence of Alzheimer's, allowing us to read what is going on in the brain through the blood," Bahado-Singh said in a statement. "The results also gave us a readout of the abnormalities that are causing Alzheimer's disease. This has future promise for developing targeted treatment to interrupt the disease process."
The researchers noted, though, that additional studies in a larger number of participants are required to validate the results.