National Institute of Health

Scientists develop machine learning algorithms that can predict pathological “tau” protein buildup in the brains of patients with early Alzheimer’s disease

The cognitive and functional decline in patients with Alzheimer’s disease is associated, among other factors, with the aberrant accumulation of a protein called “tau” inside the nerve cells of the brain. However, the current method to detect tau accumulation is expensive, risky, and poorly available at healthcare centers. Now, scientists from Korea have an innovative solution: machine-learning-based models that not only beat these challenges but also open doors to better screening, prognosis, and treatment.

Korean researchers successfully develop machine learning algorithms that accurately predict the accumulation of abnormal tau proteins in prodromal Alzheimer’s disease.

Photo courtesy: Shutterstock.

Alzheimer’s disease, which makes slow progress and ultimately leads to severe memory loss and cognitive decline, has considerably captured popular pathos and even featured in classic films like “Still Alice.” This is also fueled by the fact that the disease is difficult to diagnose and prognose and still lacks a treatment. Fortunately, the research effort into understanding the fundamentals of how the disease develops moves ahead at full steam.

This research effort has revealed the key role of the aberrant tangling of filaments of a protein called “tau” in our brain during Alzheimer’s. These tangles accumulate inside neurons (nerve cells) and disrupt their ability to communicate with each other, leading to decline in cognitive or motor functions. Studies have linked progressive tau accumulation with progressive cognitive decline in Alzheimer’s disease, meaning that tau accumulation could be a prognostic marker.

But the current technological standard for tau accumulation detection is positron emission tomography (PET) scanning, which is limited by expense, availability, and risk of excess radiation exposure. Scientists have, therefore, been trying to find an improved detection method, or better, a way to predict it in Alzheimer’s. “I believe predicting tau accumulation is an important goal not just for prognosis but because future treatment strategies may target the tau protein,” explains Dr. Sang Won Seo from the Sungkyunkwan University School of Medicine, Korea, who studies Alzheimer’s disease.

Dr. Seo’s team has an innovative solution to the tau observation problem: machine learning. Their new breakthrough is published in Nature’s Scientific Reports. This study was supported by a fund (2018-ER6203-02) by the Research of Korea Centers for Disease Control and Prevention.

In their study, the scientists first recruited 64 patients with very early stage Alzheimer’s, or prodromal Alzheimer’s, when there is definitive mild cognitive decline due to Alzheimer’s but no functional decline. They then used various combinations of multimodal biomarkers and clinical characteristics of Alzheimer’s, such as brain structure and volume, memory, Braak stages (indicative of Alzheimer’s pathology), Alzheimer’s scale scores, and genotype, and patient demographics, such as age, gender, and education, as input variables in six machine learning models to predict tau positivity (or the presence of aberrant tau accumulation) based on the random forest and gradient boosting machine algorithms. “Not all clinical settings can acquire data on all biomarkers, so we tested various combinations to make it more accessible, cost-effective, and thus widely applicable to different clinical settings,” Dr. Seo explains.

Both types of models showed good prediction performance. In the gradient boosting machine based models, the cortical thickness of the parietal lobe (the brain area processing sensory information such as touch, taste, and temperature) was the most important predictive feature, followed by performance on memory tests, cortical thickness of the occipital lobe (brain area responsible for processing visual information), and performance on the number cancellation test. In the random forest–based models, more or less the same features proved important, with word recognition score turning out to be more important than number cancellation test scores. Overall, clinical and neuropsychological data resulted in better performance than demographic data did.

Speaking of the applications of these algorithms, Dr. Seo says, “They can be useful for screening study populations in clinical trials to receive targeted tau-based treatments, helping to reduce screening failure. They can also help patients receive better treatment via improved prediction of disease severity and progress speed.”

While these are the immediate potential applications, with validation in bigger datasets (much larger than 64 patients) and in other medical conditions, the applications can be broadened further. “We believe our models could facilitate better stratification or intervention design in not just Alzheimer’s but other tauopathies as well,” says Prof. Seo. “Our secondary goal was to enable the maximization of the potential applicability of our models.”

Reference

About National Institute of Health in Korea

The Korea National Institute of Health (KNIH), one of the major operating components of the Ministry of Health and Welfare, leads the nation’s medical research. Over the past seven decades, the KNIH has made unwavering efforts to enhance the public’s health and innovate biomedical research. The KNIH seeks to eradicate diseases and make people healthier. The KNIH establishes a scientific basis and evidence underlying health policy as well as provides national research infrastructures. We also promote public health research. To this end, we make efforts to enrich a health research environment by granting funds to research projects and keeping our resources, data, and facilities more open and accessible to researchers.

Website: http://www.nih.go.kr/eng/

About Dr. Sang Won Seo

Dr. Sang Won Seo is a Professor at the Department of Neurology, Sungkyunkwan University School of Medicine, Samsung Medical Center. Dr. Seo’s lab studies the relationship among Alzheimer’ disease, small vessel disease, and cognition. He was a visiting scholar at Epidemiology, Johns Hopkins Bloomberg School of Public Health (2014) and at Neuropathology, Memory Aging Center University of California at San Francisco (2015). He received his MD from Yonsei Medical School in 1997 and his Ph.D. in Neurology in 2008 from Yonsei Medical School. He has published about 300 papers in international peer-reviewed journals.