Using Machine Learning to Develop Blood Test For Key Alzheimer’s Biomarker

Share this post:

Alzheimer’s disease, a terminal neurodegenerative disease, has historically been diagnosed based on observing significant memory loss.  There is currently no cure or disease-modifying therapy for this illness, despite hundreds of clinical trials being conducted since 2002. It is thought that the high failure rate of these trials may be because the people enrolled are in the latest stages of the disease, and have likely already suffered a level of brain tissue loss that cannot easily be repaired. The question is how to detect this disease earlier, while there is still a chance to slow its progression.

Recent research has shown that a biological marker associated with the disease, a peptide called amyloid-beta, changes long before any memory-related issues are apparent. Examining the concentration of the peptide in an individual’s spinal fluid provides an indication of risk decades before any memory related issues occur ¹. Unfortunately, accessing spinal fluid is highly invasive, requires an anaesthetist and is expensive to conduct on large segments of the population. Hence, there is a strong effort in the research community to develop a less invasive test, such as a blood test, that can yield information about Alzheimer’s disease risk.

A recent paper by my team at IBM Research – Australia, published today in Scientific Reports, used machine learning to identify a set of proteins in blood that can predict the concentration of amyloid-beta in spinal fluid. The models we built could one day help clinicians to predict this risk with an accuracy of up to 77 percent². While the test is still in the early phases of research, it could potentially help improve the selection of individuals for drug trials: individuals with mild cognitive impairment who were predicted to have an abnormal concentration of amyloid in their spinal fluid were found to be 2.5 times more likely to develop Alzheimer’s disease².

While a wide range of other proposed blood tests for Alzheimer’s disease are being developed, this is the first study to use a machine learning approach to identify sets of proteins in blood that are predictive of a biomarker in spinal fluid. This approach is easily extended to model other spinal fluid-based biomarkers – in fact, my team is presenting new work on a blood test for another key Alzheimer’s biomarker, tau, at the 14th​ International Conference on Alzheimer’s and Parkinson’s Diseases in Lisbon at the end of  March.

The publication of this research comes just in time for Brain Awareness Week, happening this week. As our population lives longer, neurodegenerative diseases such as Parkinson’s, Alzheimer’s and Huntington’s are affecting millions of people around the world. While these mysterious and crippling diseases do not yet have a cure, the answer to slowing their growth may lie in prevention. At IBM Research, our mission is to use AI and technology to understand how to help clinicians better detect and ultimately prevent these diseases in their early stages. Whether that’s through retinal imaging, blood biomarkers or minor changes in speech, we envision a future in which health professionals have a wide array of easily accessible data available to more clearly identify and track the onset and acceleration of these conditions.

1. Palmqvist, S. et al. Earliest accumulation of β-amyloid occurs within the default-mode network and concurrently affects brain connectivity.  Nat. Comm. 8, 1214 (2017).
2. Goudey, B. et al. A blood-based signature of cerebrospinal fluid Aβ_1–42 Scientific Reports, www.nature.com/articles/s41598-018-37149-7 (2019)