About 50 million people worldwide suffer from Alzheimer's disease or other forms of dementia, and age is the biggest risk factor for the disease. The researchers believe that Alzheimer's disease is mostly caused by complex interactions among many factors like genes and other, but they are not aware of these factors and their key role in the disease.

Scientists from the University of Southern California have used machine learning techniques to identify potential blood markers indicative of Alzheimer's disease, which may help researchers diagnose the disease early and develop noninvasive methods to track the progress of a patient's disease. Machine learning is a subset of artificial intelligence (AI) technology that gives computers the ability to learn without proper programming.

Researcher Paul Thompson said that this type of analysis is a new way to discover specific data patterns to identify markers of Alzheimer's disease. In a very large health measurement database, this technology can help discover the predictive characteristics of Alzheimer's disease. The results of the study were published in Frontiers in Aging Neuroscience.

How to identify special biomarkers

To date, most of the research in the field of Alzheimer's disease has focused on some of the hypotheses put forward by researchers, such as the accumulation of amyloid plaques and tau in the brain. It turns out that measuring these two indicators in the blood is very difficult. Therefore, diagnostic tests are largely based on patient memory, but unfortunately, when a person begins to show signs of amnesia, he may have been suffering from the disease for decades. Finding a disease before a patient's symptoms show is the most critical step in changing the quality of life of a patient through medication and lifestyle changes.

In this study, the researchers wanted to know if there were some hidden indicators of Alzheimer's disease, which were not found by routine blood tests, but how do we find them when we don't know what to look for? The researchers then turned their attention to machine learning technology. In 2013, researcher Greg Ver Steeg developed an advanced machine learning technique called Correlation Explanation (CorEx), which can effectively sort out regional patterns, including neuroscience, psychology, and finance, all being submerged by large amounts of data. A computational biologist Shirley Pepke used the same method to calculate his cancers in the same year.

In this study, the researchers' goal was to use the same algorithm to reveal hidden or associated factor clusters associated with Alzheimer's disease. Ver Steeg said that there is currently no single predictor to help predict whether a person has a cognitive decline, but perhaps a collection of indicators may be the best signal. Our research question is whether this algorithm can be used. A set of features is found, and this feature is better able to predict Alzheimer's disease than any separately measured factor.

Relationship cluster

The researchers studied 829 elderly people from the Alzheimer's Neuroimaging Program database and analyzed the medical records of mobile phones to identify indicators of cognitive decline and brain atrophy in these individuals over the past year. Participants were divided into three different diagnostic groups: cognitive normal group, mild cognitive impairment group, and Alzheimer's disease group; these data included data of brain imaging, genetics, plasma, and demographics from more than 400 biomarkers.

When researchers use algorithms to run data, distinct clusters of relationships emerge, and amyloid and tau are important, but the algorithm also reveals there is a strong correlation with cardiovascular health, hormone levels, metabolism, and the immune system. For example, lower levels of vitamin B12 are risk factors for cardiovascular disease, and researchers can combine it with an enzyme called matrix metalloproteinases and proteins secreted by T cells.

Researcher Thompson said that because some of the measured indicators have been previously identified and these indicators are directly related to the risk of Alzheimer's disease, this study shows synergies between different features more than a single feature and that it is more effective to predict Alzheimer's disease. Perhaps just solving one of these problems does not make a big difference, but solving a series of problems may help reduce the risk of the disease.

Finally, the researchers say that scientists are now discovering more and more biomarkers for early diagnosis of diseases and better predicting disease, while also providing new targets for blood testing. It is hoped that researchers in the future can do more large-scale studies to confirm the results of this study, and that they may use this new algorithm to find hidden factors indicating multiple other diseases, such as schizophrenia and depression.

Author's Bio: 

Starting and specialized as a chemical supplier of inhibitors, fluorine compounds, APIs, superparamagnetic iron oxide particles, metabolites and impurities since 2005, BOC Sciences is a now a major supplier of pharmaceutical services. Besides, its scientists have compiled passages on various pathways such as Erk Signaling Pathway, GPCR Signaling Pathway, Hedgehog Signaling Pathway, Hippo Signaling Pathway, JAK/STAT Signaling Pathway in a hope to brief the public on how our cells are connected to function.