Blood-based test may help detect early Alzheimer’s disease

Researchers at Beaumont Health have developed a simple blood test that could help to predict Alzheimer’s disease before symptoms develop and the brain has suffered irreversible damage. The team used artificial intelligence (AI) and deep learning processes to analyze extracted genomic DNA from samples of whole blood, revealing 152 significant genetic differences in 24 Alzheimer’s patients vs. 24 patients who did not have the disease. Brain inflammation caused by Alzheimer’s is thought to trigger the production of white blood cells, or leukocytes, which become genetically altered as they fight the disease. The researchers looked for resulting genetic markers, or methylation marks, a chemical modification of genes that changes the way they function, signaling Alzheimer’s onset. A total of six AI and deep learning platforms analyzed about 800,000 changes in the genome of the leukocytes. “What the results said to us is there are significant changes in accessible blood cells that we can use possibly to detect Alzheimer’s,” said Ray Bahado-Singh, chairman of the Beaumont department of obstetrics and gynecology and one of the study’s authors. “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.” He added that the results also highlighted abnormalities that cause Alzheimer’s. The team aims to replicate its findings in a much larger study in the next year or two. Their work was published March 31, 2021, in PLOS One.

AI-based tool helps predict course of COVID-19

Researchers at the University of Pennsylvania and Brown University have demonstrated that artificial intelligence (AI) based on chest X-rays and clinical data can predict the progression to severe illness in patients with COVID-19. In a retrospective study, patients who presented to the emergency room of a hospital in the Penn health system with a COVID-19 diagnosis confirmed by RT-PCR and a chest X-ray from their initial presentation were identified and randomized into training, validation and test sets. Inputting the chest X-rays to a deep neural network and clinical data, models were trained to predict critical or noncritical disease. The team used deep-learning insights culled from the model and clinical data to build disease progression models, which were externally tested on patients presenting to Brown University-affiliated hospitals. When chest X-rays were added to clinical data for severity prediction, the area under the receiver operating characteristic curve (ROC-AUC) rose from 0.821 to 0.846 on internal testing and from 0.731 to 0.792 on external testing. When deep-learning features were combined with clinical data for progression prediction, the concordance index rose from 0.769 to 0.805 on internal testing and 0.707 to 0.752 on external testing. Previous studies have sought to predict severity of COVID-19, but they have focused on diagnostics or binary outcomes rather than progression to a critical event, the researchers said. “Our study is unique and clinically relevant because it shows COVID-19 severity and specific time-to-critical-event windows can be predicted using clinical variables, and that using deep-learning features extracted from chest X-rays can incrementally increase the strength of those predictions and perform the prediction by radiologist-derives severity scores,” the authors wrote. Their findings appeared online March 24, 2021, in The Lancet Digital Health.

Myeloid cells differ in primary and recurrent glioblastoma

Researchers from the Free University of Brussels have mapped the glioblastoma immune landscape in both animal models and patients with newly diagnosed or recurrent glioma at single-cell resolution. Five-year survival rates in glioblastoma patients remain in the single digits, in part because of a suppressed immune response. In their work, the team focused on myeloid cells, which play a role in antitumor immunity, but also interact with tumor cells to affect their aggressiveness. There are several different macrophage populations that stem from different progenitor cells, and the team showed that “tumor-associated macrophages (TAMs) consisted of microglia- or monocyte-derived populations, with both exhibiting additional heterogeneity, including subsets with conserved lipid and hypoxic signatures… Microglia-derived TAMs were predominant in newly diagnosed tumors, but were outnumbered by monocyte-derived TAMs following recurrence, especially in hypoxic tumor environments.” The findings, which suggest that targeting specific TAM subsets could be a fruitful therapeutic strategy, were published in the March 29, 2021, issue of Nature Neuroscience.