AI and Blood Biomarkers Show Promise in Predicting Disease Risk Years in Advance

A study has found that a women's risk of developing dementia may be predicted 25 years before symptoms begin through a test of a blood-based biomarker called 'phosphorylated tau 217' -- a protein linked to the brain changes seen in Alzheimer's disease. Higher levels of phosphorylated tau 217, or p-tau217, were strongly associated with future mild cognitive impairment and dementia -- of which Alzheimer's disease is the most common form -- among older women who were cognitively healthy at the study's start.

The study, published in The Journal of the American Medical Association (JAMA) Network Open, analysed data of 2,766 participants in the Women's Health Initiative Memory Study, a US national study that enrolled women aged 65 to 79 in the late 1990s and followed them for up to 25 years. Women who developed memory or thinking problems, including dementia, were identified during follow-up.

Higher levels of p-tau217 in blood at the start of the study were related with a higher chance of developing dementia later in life, with increasing levels of the biomarker related with increasing dementia risk. Higher p-tau217 levels were also more strongly associated with poorer cognitive outcomes among women aged 70 and above, compared to those younger, and among those with the APOE e4 genetic risk factor for Alzheimer's disease.

The study also found that p-tau217 was more predictive of dementia among women who were randomly assigned estrogen plus progestin hormone therapy versus placebo. Blood-based biomarkers like p-tau217 are especially promising because they are far less invasive and potentially more accessible than brain imaging or spinal fluid tests.

In separate research on liver cancer, a team in Japan focused on a specific protein called MYCN. Scientists already knew this protein played a role in liver cancer in damaged livers, but they didn't know how. To figure it out, the researchers used a mouse model to see what happens when MYCN is turned up too high. They found that when MYCN teamed up with another gene called AKT, 72% of the mice developed tumors within just 50 days.

To understand why this happens, the team used a technique called spatial transcriptomics. They discovered a specific cluster of 167 genes that change in "tumor-free" areas when MYCN levels rise. They've labeled this environment the "MYCN niche."

Using this data, the researchers built a machine-learning model. This algorithm looks at the gene patterns in a liver and gives it a score. If the score is high, it means the liver environment is primed for cancer. When they tested this on human data, it worked with 93% accuracy. Interestingly, the score was even better at predicting future trouble when they looked at the healthy-looking tissue around a tumor rather than the tumor itself.

Meanwhile, scientists affiliated with the German Center for Diabetes Research (DZD) suggest that a simple blood test, combined with AI, could help identify individuals at high risk of developing type 2 diabetes and its complications at an early stage. A new study published in Biomarker Research analyzed blood samples from participants across several study cohorts with known prediabetes risk profiles.

Researchers focused on DNA methylation patterns, which are chemical modifications that regulate gene activity without altering the DNA sequence itself. Using machine learning techniques, the team identified 1,557 epigenetic markers that together formed a biological "fingerprint" of prediabetes risk.

Using these markers, the AI model was able to assign individuals to high-risk prediabetes clusters with an accuracy of about 90%, even when tested in an independent validation cohort. Many of the epigenetic markers were specific to particular clusters and reflected different biological signaling pathways. Several had already been linked in earlier studies to type 2 diabetes, chronic inflammation, cardiovascular and kidney disease.

Previous research conducted by the DZD divided prediabetes into at least six distinct clusters. Three clusters are associated with moderate risk, while the other three carry a high risk of developing type 2 diabetes and related complications.