Will a person develop dementia in the future? New AI algorithm predicts 15 years in advance!

Dementia refers to a chronically acquired progressive intellectual disability syndrome, which is a general term for a large class of diseases, such as Alzheimer's disease, frontotemporal dementia, Creutzfeldt-Jakob disease, Lewy body dementia, vascular dementia or traumatic brain injury dementia, etc.

It may take many years for dementia to develop from asymptomatic to full-blown. Clinically, it is mainly characterized by a slow onset of intellectual decline, accompanied by varying degrees of personality or character changes, such as reduced interest and apathy; impaired social function and childish behavior; emotional instability, anxiety, depression, or irritability, etc. Due to the complexity of the causes of dementia, and the fact that the cause of dementia caused by neuropathy has not yet been discovered, a specific drug for dementia has not yet been developed. Currently, clinical treatment can only focus on improving cognition and symptomatic treatment, while supplemented by life care and rehabilitation methods.

Because there is no drug for the disease, how to correctly judge whether a person will suffer from dementia in the future has become a very important issue. Thanks to the development of sequencing technology, blood-based biomarkers may be used as a better tool for early screening of dementia.

Recently, the research team of Professor Yu Jintai from the Department of Neurology, Huashan Hospital Affiliated to Fudan University, and the team of Professor Feng Jianfeng/Researcher Cheng Wei from the Institute of Brain-Inspired Intelligence Science and Technology of Fudan University published a research paper titled "Plasma proteomic profiles predict future dementia in healthy adults" in the journal Nature Aging.

Using large-scale proteomics analysis and artificial intelligence algorithms, the research team found that GFAP, NEFL, GDF15 and LTBP2 are the four plasma proteins that are most relevant to the disease and have the highest risk in different types of dementia. Based on these four proteins, it may be possible to predict the occurrence of dementia 15 years in advance.

The research team obtained blood sample data from 52,645 adults from the UK Biobank, of which 1,417 adults were diagnosed with dementia 14 years after sampling, including all-cause dementia (ACD), Alzheimer's disease (AD) and vascular dementia (Vascular dementia, VaD).

In these data samples, the research team compared a total of 1,463 plasma proteins and initially found (Figure 1, model 1) that there were 186 plasma proteins highly correlated with ACD, 16 plasma proteins highly correlated with AD, and 139 plasma proteins highly correlated with VaD. Taking into account the interactions and synergistic effects between proteins, the research team further analyzed and found (Figure 1, model 2) that GFAP, NEFL, LTBP2 and GDF15 were significantly correlated in all three types of dementia.

Figure 1 Correlation analysis between plasma proteins and dementia

The discovery of these risk proteins provides the research team with a tool to predict the probability of developing dementia . For example, the research team found that adults with higher levels of GFAP in their plasma were 2.91 times more likely to develop Alzheimer's disease than adults with GFAP levels at baseline (Figure 2).

GFAP protein is mainly expressed in astrocytes in the brain and is a type of protein that provides structural support for them. In previous studies, it has often been used as a marker for astrocytes, and more and more studies have shown that it can be used as a biomarker for Alzheimer's disease.

Figure 2 GFAP protein predicts the risk of clinical diagnosis of Alzheimer's disease

More importantly, the research team found that these risk proteins had already shown higher than normal values ​​15 years before the diagnosis of dementia. In particular, in the prediction of Alzheimer's disease, GFAP, GDF15 and NEFL showed significant differences from normal values ​​(Figure 3). This also means that through these proteins, the onset of dementia can be predicted more than ten years in advance, thereby enabling early screening and intervention for dementia, helping to alleviate or even prevent patients' symptoms.

Figure 3 The trajectory of plasma GFAP protein content changing over time before the diagnosis of Alzheimer's disease. Red line: AD diagnosis group; blue line: normal control group

In general, the research team demonstrated the potential of GFAP, NEFL, and GDF15 to predict the risk of dementia more than ten years in advance, and emphasized the importance of GFAP. The advantage of GFAP is that it has a strong specificity for the diagnosis of early dementia. Multiple studies have reported its increase in plasma, and for other neurodegenerative diseases other than dementia, the level of GFAP in plasma has been maintained around normal values. The disadvantage is that GFAP cannot distinguish between Alzheimer's disease and non-Alzheimer's dementia.

The research team innovatively used data-driven proteomics strategy analysis, combined with the largest research cohort to date and 14 years of follow-up data, to discover important plasma protein biomarkers for predicting dementia. The research team not only strongly supported the importance of GFAP, NEFL and GDF15 in predicting dementia, but also discovered a new biomarker - LTBP2. This is of great significance for screening high-risk groups for dementia and early intervention.

It is understood that some medical institutions for physical examinations have contacted the research team to explore the possibility of adding relevant tests to physical examination items. If everything goes well, it is expected that they can be applied to clinical tests in half a year to screen people at high risk of dementia.

In the next step, the team will conduct data collection and cross-validation on the risk cohort of people with dementias such as Alzheimer's disease in my country, correct the relevant data based on the baseline level of the Chinese population cohort, and develop the most suitable data model for predicting Alzheimer's disease and other dementias for the Chinese population cohort.

References:

[1] Guo Y, You J, Zhang Y, Liu WS, Huang YY, Zhang YR, Zhang W, Dong Q, Feng JF, Cheng W, Yu JT. Plasma proteomic profiles predict future dementia in healthy adults. Nat Aging. 2024 Feb;4(2):247-260. doi: 10.1038/s43587-023-00565-0. Epub 2024 Feb 12. PMID: 38347190.