Available via license: CC BY
Content may be subject to copyright.
TYPE Original Research
PUBLISHED 19 December 2024
DOI 10.3389/fnagi.2024.1513930
OPEN ACCESS
EDITED BY
Wencai Liu,
Shanghai Jiao Tong University, China
REVIEWED BY
Kai Xu,
Shandong First Medical University, China
Lizhihan Yu,
Texas A and M University, United States
*CORRESPONDENCE
Dantao Peng
pengdantao2000@163.com
RECEIVED 19 October 2024
ACCEPTED 05 December 2024
PUBLISHED 19 December 2024
CITATION
Zhen W, Wang Y, Zhen H, Zhang W, Shao W,
Sun Y, Qiao Y, Jia S, Zhou Z, Wang Y, Chen L,
Zhang J and Peng D (2024) Association
between Alzheimer’s disease pathologic
products and age and a pathologic
product-based diagnostic model for
Alzheimer’s disease.
Front. Aging Neurosci. 16:1513930.
doi: 10.3389/fnagi.2024.1513930
COPYRIGHT
©2024 Zhen, Wang, Zhen, Zhang, Shao, Sun,
Qiao, Jia, Zhou, Wang, Chen, Zhang and
Peng. This is an open-access article
distributed under the terms of the Creative
Commons Attribution License (CC BY). The
use, distribution or reproduction in other
forums is permitted, provided the original
author(s) and the copyright owner(s) are
credited and that the original publication in
this journal is cited, in accordance with
accepted academic practice. No use,
distribution or reproduction is permitted
which does not comply with these terms.
Association between Alzheimer’s
disease pathologic products and
age and a pathologic
product-based diagnostic model
for Alzheimer’s disease
Weizhe Zhen1, Yu Wang2, Hongjun Zhen3, Weihe Zhang2,
Wen Shao2, Yu Sun2, Yanan Qiao2, Shuhong Jia2, Zhi Zhou2,
Yuye Wang2, Leian Chen2, Jiali Zhang1and Dantao Peng1,2*
1Graduate School, Beijing University of Chinese Medicine, Beijing, China, 2Department of Neurology,
China-Japan Friendship Hospital, Beijing, China, 3Department of Orthopedics, Handan Chinese
Medicine Hospital, Handan, Hebei, China
Background: Alzheimer’s disease (AD) has a major negative impact on people’s
quality of life, life, and health. More research is needed to determine the
relationship between age and the pathologic products associated with AD.
Meanwhile, the construction of an early diagnostic model of AD, which is mainly
characterized by pathological products, is very important for the diagnosis and
treatment of AD.
Method: We collected clinical study data from September 2005 to August
2024 from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database.
Using correlation analysis method like cor function, we analyzed the pathology
products (t-Tau, p-Tau, and Aβproteins), age, gender, and Minimum Mental
State Examination (MMSE) scores in the ADNI data. Next, we investigated the
relationship between pathologic products and age in the AD and non-AD groups
using linear regression. Ultimately, we used these features to build a diagnostic
model for AD.
Results: A total of 1,255 individuals were included in the study (mean [SD]
age, 73.27 [7.26] years; 691male [55.1%]; 564 female [44.9%]). The results of the
correlation analysis showed that the correlations between pathologic products
and age were, in descending order, Tau (Corr=0.75), p-Tau (Corr=0.71), and Aβ
(Corr=0.54). In the AD group, t-Tau protein showed a tendency to decrease
with age, but it was not statistically significant. p-Tau protein levels similarly
decreased with age and its decrease was statistically significant. In contrast to
Tau protein, in the AD group, Aβlevels increased progressively with age. In the
non-AD group, the trend of pathologic product levels with age was consistently
opposite to that of the AD group. We finally screened the optimal AD diagnostic
model (AUC=0.959) based on the results of correlation analysis and by using the
Xgboost algorithm and SVM algorithm.
Conclusion: In a novel finding, we observed that Tau protein and Aβhad opposite
trends with age in both the AD and non-AD groups. The linear regression
curves of the AD and non-AD groups had completely opposite trends. Through
a machine learning approach, we constructed an AD diagnostic model with
excellent performance based on the selected features.
KEYWORDS
Alzheimer’s disease, pathologic product, machine learning, age, diagnostic model, linear
regression
Frontiers in Aging Neuroscience 01 frontiersin.org
Zhen et al. 10.3389/fnagi.2024.1513930
1 Introduction
Alzheimer’s disease (AD) is a neurodegenerative disease that
seriously jeopardizes human health and affects patients’ quality
of life (2023,2024). It is the number one cause of dementia
and mainly affects the middle-aged and elderly population (2022;
Scheltens et al., 2021). Before developing Alzheimer’s disease,
patients will go through the stages of subjective memory complaints
(SMC), mild cognitive decline and so on. How to diagnose
Alzheimer’s disease more accurately and distinguish it from
the preclinical stage of Alzheimer’s disease as well as normal
people has been a hot topic of research. Previous researchers
have attempted to differentiate AD using biomarkers, imaging,
and some behavioral-based kinesiology tests, among others, with
biomarker research undoubtedly receiving the most attention
(Bai et al., 2021;Winchester et al., 2023;Küçükali et al., 2023;
Yang et al., 2020). A large number of biomarkers have been
detected in blood, cerebrospinal fluid tests, etc., which have a
good ability to differentiate between patients with AD (Izzo et al.,
2021;Kumari et al., 2022). The researchers even spent a great
deal of time studying longitudinal changes in these biomarkers,
monitoring changes in their levels throughout the course of the
onset of Alzheimer’s disease (Jia et al., 2024;Yakoub et al., 2023).
And with the continuous advancement of histologic research
techniques, more and more biomarkers are being discovered in a
higher throughput manner. Our team has previously uncovered
a very large number of AD biomarkers through both blood and
urine testing methods, using histology-related techniques, and
has built an AD diagnostic model based on them (Wang et al.,
2023a,b).
β-amyloid (Aβ) and Tau proteins are the focus of biomarker
research as recognized markers of AD pathology. Many of both
protein families have been found to be closely associated with the
onset and progression of AD (Ferrari-Souza et al., 2022;Ashton
et al., 2021;Horie et al., 2023).Previous studies have focused on
the particular significance of these two pathologic products in the
molecular mechanisms underlying the developmental process of
Alzheimer’s disease (Zhang H. et al., 2021;Busche and Hyman,
2020). In contrast, the association between these two pathologic
products and age in the preclinical and onset stages of AD
has lacked elucidation in large-sample studies (Stern et al.,
2023).
In terms of research on diagnostic models for AD, there are
many studies that use biomarkers as features, and not a few
of them incorporate Aβand Tau protein (Ferreiro et al., 2023).
However, in previous studies, Aβand Tau protein were hardly
used as core features, and the inclusion of other biomarkers
mixed the significance of the two in modeling. At the same
time, previous studies also suffered from the shortcomings of
using algorithms mainly focusing on regression algorithms, a
single type of algorithm and a lack of sample size (Hammond
et al., 2020;Gao et al., 2023). We used classification algorithms
with excellent performance in this study to construct and train
an AD diagnostic model using Aβand Tau protein as the
core features.
2 Methods
2.1 Design
The Alzheimer’s Disease Neuroimaging Initiative (ADNI)
database1provided the data used in this investigation, which were
gathered between September 2005 and August 2024. Established
in 2003, the ADNI program is a research endeavor with the goal
of examining the course of Alzheimer’s disease and its preclinical
phases through the use of MRI, PET, biomarkers, and clinical and
neuropsychological examinations. All participating institutions’
Institutional Review Boards have given their approval for the ADNI
trial. All participants, or their authorized representatives, have
given written informed permission to ADNI in compliance with
the Declaration of Helsinki.
2.2 Participants
After undergoing a battery of cognitive functioning tests, each
individual was assigned to one of four groups: AD, mild cognitive
impairment (MCI), SMC, or control (CN). The Mini-Mental State
Examination (MMSE) scores for AD were 20–26, while for CN,
SMC, and MCI, they were 24–30. For CN, MCI, and AD, the
Clinical Dementia Rating (CDR) was 0.5, 0.5, and ≥0.5. For the
various ADNI cohorts, the enrollment processes and inclusion
criteria were generally the same. Previous descriptions have been
made of the specific enrollment processes and inclusion criteria
for the various diagnostic categories of the ADNI cohort (Petersen
et al., 2010). You could find the ADNI database protocol2with
detailed inclusion and exclusion requirements. In order to evaluate
cognitive function, we used the MMSE.
2.3 Biomarker collection and analysis
CSF Tau protein and Aβprotein data from the ADNI database
were used in our study. Methods of CSF collection and biomarker
measurement have been previously reported (Hampel et al., 2010;
Shaw et al., 2009). The ADNI database did not report outliers.
However, biomarker assays have detection intervals. The upper
limit of detection for Aβis 1,700 pg/mL and the lower limit is 200
pg/mL. The upper limit of detection for t-Tau is 1,300 pg/mL and
the lower limit is 80 pg/mL. The upper limit of detection for p-Tau
is 120 pg/mL and the lower limit is 8 pg/mL. The detections that
exceeded the detection interval the most accounted for <15% of
the total data, and only a very small number of the other detections
exceeded the detection interval. For data exceeding the detection
range of Tau or Aβproteins, we took half of the lower detection
limit value to replace data below the lower detection limit, and used
the upper detection limit value to replace data above the upper
detection limit.
1http://adni.loni.usc.edu
2https://adni.loni.usc.edu/help-faqs/adni- documentation/
Frontiers in Aging Neuroscience 02 frontiersin.org
Zhen et al. 10.3389/fnagi.2024.1513930
FIGURE 1
Variable correlation analysis of all features. Aβ,β-amyloid; t-Tau, total-Tau; p-Tau, phosphorylated tau; MMSE, Minimum Mental State Examination.
2.4 Model constructing and training
We employed two machine learning algorithms, support vector
machine (SVM) and extreme gradient boosting (XGBoost), to
build AD diagnostic models. Cross-validation is used to evaluate
the performance of machine learning models as well as for
hyperparameter tuning. Ten-fold cross-validation was applied to
the training and validation sets in order to reduce overfitting and
enhance the model’s functionality. We tuned the hyperparameters
based on the results of the ten-fold cross-validation to get the best
performing model. The best model for this study was determined by
looking at the Receiver Operating Characteristic (ROC) curve and
selecting the model with the highest area under the curve (AUC).
2.5 Statistical analysis
R software (version 4.3.1) and IBM SPSS Statistics for Windows
version 27.0 were used to conduct all statistical tests. We employed
nonparametric tests to compare the Non-AD (include CN, SMC,
EMCI, and LMCI) and AD groups for variables like total-Tau (t-
Tau), phosphorylated tau (p-Tau), Aβ, age, and MMSE that did
not match the requirements of analysis of variance (ANOVA).
The chi-square test was used for statistical analysis of counts like
gender. The threshold for a difference to be deemed statistically
significant was p<0.05. Since the data are derived from public
databases, the occurrence of missing values is often unavoidable.
In this study, the proportion of missing values to the total data has
been well over 50%. In order to minimize the error in the study,
we used direct culling of missing values in the data instead of using
interpolation. In contrast, this is the optimal way to ensure data
integrity and accuracy.
One popular technique for determining how closely variables
correlate with one another is correlation analysis. To determine the
correlation coefficient and compute the correlation of variables, we
utilize the cor function found in R-4.3.1. The correlation between
the variables is higher the closer the correlation coefficient’s
absolute value is to 1.
In several contexts, the relationship between diseased products
and age was estimated using linear regression analysis. Initially, we
examined the relationship between pathogenic products (include
Aβ, t-Tau protein, and p-Tau protein) and age in the AD and
non-AD groups. We also investigated the relationship between p-
Tau/t-Tau and Aβ/t-Tau and age using linear regression in an effort
to better understand the relationship between these pathogenic
products and age.
3 Results
3.1 Demographic and clinical
characteristics of patients
A total of 1,255 individuals were included in the study (mean
[SD] age, 73.27 [7.26] years; 691 male [55.1%]; 564 female [44.9%]).
Frontiers in Aging Neuroscience 03 frontiersin.org
Zhen et al. 10.3389/fnagi.2024.1513930
FIGURE 2
Association between t-Tau and age in AD and non-AD groups. The cyan dots and lines represent the samples and fitted lines for the non-AD group,
respectively. Red dots and lines represent samples and fitted lines for the AD group, respectively. Scatterplot of the association between t-Tau and
age (A) and Fitted curves for all participants (B). Association between t-Tau and age in the AD and non-AD groups (C) and The p-value and R-value
for each of the two groups (D).
Six factors in all were examined: pathogenic products (Aβ, t-Tau,
and p-Tau), age, sex, and MMSE scores. The average score for the
MMSE test was 26.96 [3.18], the average score for Aβwas 966.19
[458.57], the average score for t-Tau was 290.57 [136.04], and the
average score for p-Tau was 27.96 [14.91] for every subject. There
were statistically significant differences in every attribute between
the groups (Table 1).
3.2 Correlation analysis between variables
Figure 1 displays the findings of the six factors’ correlation
study with disease type. where the Corr values are between −1.0
and 1.0; the higher the correlation between the variables, the closer
the Corr value’s absolute value is to 1. Conversely, a correlation is
less the closer it is to 0. The pathogenic products and age had the
following associations, in decreasing order of absolute Corr values:
t-Tau (0.75), p-Tau (0.71), and Aβ(0.54).
3.3 Association of Tau proteins with age
As Figure 2 shows, when we studied all subjects, we
found that overall the level of Tau protein increased with
age (Figures 2A,B). Interestingly, in the AD group, Tau
protein appeared to slowly decrease with age, although this
decrease was not statistically significant (p>0.05). In the
Non-AD group, the longitudinal rise in Tau protein was
similarly associated with a lateral increase in age (p<0.05)
(Figures 2C,D).
3.4 Association of p-Tau protein with age
As Figure 3 shows, when we studied all subjects, we
found that, overall, the levels of p-Tau protein increased
with age (Figures 3A,B). Like t-Tau protein, in the AD
group, p-Tau protein appeared to slowly decrease with
Frontiers in Aging Neuroscience 04 frontiersin.org
Zhen et al. 10.3389/fnagi.2024.1513930
FIGURE 3
Association between p-Tau and age in AD and non-AD groups. The cyan dots and lines represent the samples and fitted lines for the non-AD group,
respectively. Red dots and lines represent samples and fitted lines for the AD group, respectively. Scatterplot of the association between p-Tau and
age (A) and Fitted curves for all participants (B). Association between p-Tau and age in the AD and non-AD groups (C) and the p-value and R-value
for each of the two groups (D).
age. However, the difference was that the decrease in p-
Tau protein was statistically significant (p <0.05). In the
non-AD group, the longitudinal rise in p-Tau protein was
similarly associated with a lateral increase in age (p <0.05)
(Figures 3C,D).
3.5 Association of p-Tau/t-Tau with age
Overall, as seen in Figures 4A,B, there was a slight but steady
tendency for p-Tau/t-Tau levels to rise with aging. p-Tau/t-Tau
significantly decreased in the AD group as age increased, and this
decline was statistically distinct (p<0.05). On the other hand,
p-Tau/t-Tau increased with age (p<0.05) in the non-AD group
(Figures 4C,D).
3.6 Association of Aβwith age
Overall, as Figure 5 illustrates, Aβlevels steadily declined with
age, which was different from t-Tau and p-Tau (Figures 5A,B). Aβ
levels significantly increased with age in the AD group, which was
likewise in contrast to t-Tau and p-Tau (p<0.05). Conversely,
Aβlevels in the non-AD group dropped with age (p<0.05)
(Figures 5C,D).
3.7 Association of Aβ/t-Tau with age
As shown in Figure 6, as a whole, Aβ/t-Tau levels gradually
decreased with age, which is consistent with the change of Aβwith
age (Figures 6A,B). In the AD group, Aβ/t-Tau levels increased
significantly with age (p<0.05). While in the non-AD group,
Aβ/t-Tau instead decreased with age (p <0.05) (Figures 6C,D).
Frontiers in Aging Neuroscience 05 frontiersin.org
Zhen et al. 10.3389/fnagi.2024.1513930
FIGURE 4
Association between p-Tau/t-Tau and age in AD and non-AD groups. The cyan dots and lines represent the samples and fitted lines for the non-AD
group, respectively. Red dots and lines represent samples and fitted lines for the AD group, respectively. Scatterplot of the association between
p-Tau/t-Tau and age (A) and Fitted curves for all participants (B). Association between p-Tau/t-Tau and age in the AD and non-AD groups (C) and the
p-value and R-value for each of the two groups (D).
3.8 Construction and optimization of AD
diagnostic models
We built two machine learning models for diagnosing AD using
the XGBoost classifier and the SVM classifier, respectively, based
on the six previously mentioned features. To avoid overfitting,
we further enhanced the model performance via ten-fold cross-
validation. Among them, the classifier model based on the XGBoost
algorithm has superior performance (AUC of 0.959), accuracy of
0.69, specificity of 0.86, and sensitivity of 0.95. The classifier model
based on the support vector machine (SVM) algorithm has an AUC
of 0.924, accuracy of 0.90, sensitivity of 0.96, and Specificity of 0.66
(Figure 7).
4 Discussion
Our study systematically analyzed the relationship between
changes in intracranial t-Tau protein, p-Tau protein, and Aβ
protein levels and age in AD patients and Non-AD population
through data mining and analysis of the ADNI database. The
results of the study were very interesting. We found that for
both t-Tau protein and Aβprotein, the trends with age were
diametrically opposed in AD patients and Non-AD populations.
For both t-Tau protein and p-Tau protein, the levels of these
pathogens progressively decreased the older the AD patient
was. The difference, however, was that t-Tau protein showed a
decreasing trend but was not statistically different compared to
Frontiers in Aging Neuroscience 06 frontiersin.org
Zhen et al. 10.3389/fnagi.2024.1513930
FIGURE 5
Association between Aβand age in AD and non-AD groups. The cyan dots and lines represent the samples and fitted lines for the non-AD group,
respectively. Red dots and lines represent samples and fitted lines for the AD group, respectively. Scatterplot of the association between Aβand age
(A) and Fitted curves for all participants (B). Association between Aβand age in the AD and non-AD groups (C) and the p-value and R-value for each
of the two groups (D).
the statistically significant decrease in p-Tau protein. As for Aβ
protein, the older the age of AD patients, the Aβlevel was
continuously increasing. In contrast, Aβlevels in the Non-AD
population gradually declined with age. Previous studies on the
correlation between these pathologic products and patient age
are lacking and not clearly recognized or elaborated. However,
longitudinal changes in these pathologic products over time at
the individual level have been examined in previous studies,
which is different from our observation of the association between
pathologic products and age at the population level (Barthélemy
et al., 2020). Additionally, it is not clear that some studies have
focused on changes in the levels of pathologic products long before
the onset of Alzheimer’s disease, and have not examined changes
after the onset of the disease (Jia et al., 2024). There are also
studies that do not distinguish between studies of AD patients and
Non-AD groups (Cogswell et al., 2024).
One of the main reasons we chose Aβand Tau proteins for
our study is that it has a very important impact in the course and
mechanisms of AD (Pang et al., 2022;Sadleir and Vassar, 2023).
Abnormal aggregation of Aβand Tau proteins is an important
pathogenesis and pathological hallmark of AD. Questions about
how the two are produced and how their levels change during
disease progression have been an important issue affecting our
understanding of AD, as well as a focus and difficulty in research.
Previous findings suggest that δ-secretase cleaved Tau proteins
may stimulate Aβproduction by upregulating STAT1-BACE1
signaling in AD patients (Zhang Z. et al., 2021). This is a rather
important finding. It not only reveals the molecular regulatory
process between the two pathologic products. More importantly,
it suggests to us that the molecular regulatory process that exists
between the two may allow the levels of these pathology products
to be dynamically regulated during the course of the disease. This
Frontiers in Aging Neuroscience 07 frontiersin.org
Zhen et al. 10.3389/fnagi.2024.1513930
FIGURE 6
Association between Aβ/t-Tau and age in AD and non-AD groups. The cyan dots and lines represent the samples and fitted lines for the non-AD
group, respectively. Red dots and lines represent samples and fitted lines for the AD group, respectively. Scatterplot of the association between
Aβ/t-Tau and age (A) and Fitted curves for all participants (B). Association between Aβ/t-Tau and age in the AD and non-AD groups (C) and the
p-value and R-value for each of the two groups (D).
is likely to be an important mechanism for the pathogenesis and
disease progression of AD. Not only that, Aβpathology may induce
changes in soluble tau release and phosphorylation (Mattsson-
Carlgren et al., 2020). These findings show that the relationship
between Aβand t-Tau and p-Tau is bi-directionally regulated
and mutually restrained. Fluctuations in the levels of the two
pathogens are influenced by each other. This also gives the ratio
of the two pathologic products more value for clinical studies.
In turn, abnormal aggregation of the two could ultimately drive
the disease progression by leading to loss of synapses, affecting
synaptic function and thus disrupting memory formation (Li et al.,
2018). While the roles of Aβand Tau in the pathogenesis of
AD continue to be elucidated, researchers are monitoring changes
in their levels, further exploring their potential as biomarkers
of the disease, and even screening for other AD biomarkers
with good predictive ability based on them (Chiu et al., 2021;
Boza-Serrano et al., 2022). As the technology associated with
biomedical engineering continues to advance, more assays have
been developed for the detection of Aβand Tau. In addition to
furthering our understanding of the pathologic processes of AD,
we are discovering more pathologic processes associated with the
aggregation of these pathogens. These are also one of the hotspots
and directions for future research (Pichet Binette et al., 2021).
We constructed and trained a machine learning model using
pathology products such as Aβand Tau proteins as core features.
In terms of the model’s performance, its ability to distinguish
between AD and non-AD is excellent. Previous studies have
rarely focused only on the contribution of Tau and Aβproteins
to constructing AD diagnostic models, often incorporating some
other biomarkers (Gaetani et al., 2021;Ficiarà et al., 2021;Franciotti
et al., 2023;Khan et al., 2024). These studies have their own
innovations and strengths, but inevitably have some shortcomings.
The shortcomings mainly lie in the lack of performance of the
diagnostic model, the small sample size, the excessive number of
Frontiers in Aging Neuroscience 08 frontiersin.org
Zhen et al. 10.3389/fnagi.2024.1513930
FIGURE 7
Diagnostic model of AD with pathology products as core features.
TABLE 1 Baseline demographics and clinical characteristics.
Characteristics Total
(1,255)
Non-AD
(1,022)
AD
(233)
P
value
Age (years) 73.27 72.94 74.71 <0.01
Gender
Male (%) 55.1 54.2 58.8 0.204
Female (%) 44.9 45.8 41.2 0.204
MMSE 26.96 27.88 22.94 <0.01
Aβ(pg/mL) 966.19 1,034.97 664.48 <0.01
t-Tau (pg/mL) 290.57 272.79 368.51 <0.01
p-Tau (pg/mL) 27.96 26.01 36.51 <0.01
Aβ,β-amyloid; t-Tau, total-Tau; p-Tau, phosphorylated tau; MMSE, Minimum Mental
State Examination.
incorporated features, and so on. The performance of the AD
diagnostic model constructed in our study is relatively superior.
The relative singularity of the incorporated features can highlight
more the importance of the pathology products in the model
construction. Our choice of algorithms for machine learning
that performs well in dealing with classification problems is an
important guarantee of the sophistication of our study (Li J. et al.,
2022;Yi et al., 2023;Binder et al., 2022). The Xgboost algorithm
has excellent performance in dealing with classification problems.
It is characterized by its ability to handle large volumes of data
while maintaining accuracy and predictive performance over other
classification algorithms (Yue et al., 2022;Li Q. et al., 2022).
Compared to XGBoost, the SVM algorithm is slightly less accurate
and less predictive. However, in some special problems, SVM
has advantages that other algorithms do not have, such as when
dealing with nonlinearly differentiable data and when dealing with
high-dimensional data (Huang et al., 2018;Ding et al., 2022).
Undeniably, biomarker-based machine learning diagnostic
models for AD are still the most dominant research (Shah et al.,
2023;Kononikhin et al., 2022). The biomarkers involved in
these studies include not only the pathology products that we
generally recognize, but also some lipids, proteins and so on that
are closely related to the pathogenesis of AD as screened by
new research methods (Wang et al., 2022). The great progress
in molecular biology research has also led to the expansion of
the scope of clinical biomarkers (Krokidis et al., 2023). Their
potential for clinical application will be enhanced if the cost of the
assay can be reduced. In addition to the traditional work related
to the construction of biomarker-based AD diagnostic models,
more AD-related ancillary test results have been included in the
study. Common neurological examinations in the clinic, such
as electroencephalography, are used to construct AD diagnostic
models, which also have good diagnostic performance (Parreño
Torres et al., 2023). In addition to common examination means,
more and more medical devices provide more dimensional
examination results. As a new type of model, AD diagnostic model
based on eye movement and language has a good prospect for
clinical application due to its noninvasive and easy-to-operate
characteristics (Jang et al., 2021). Overall, research efforts in AD
machine learning diagnostic modeling have produced a large
number of models with very good performance and potential for
clinical applications. However, unfortunately, more comprehensive
AD diagnostic models covering multidimensional markers have not
yet been developed.
Inevitably, there are shortcomings in our study. In the section
examining the question of the association between pathologic
products and age, we had planned to compare all preclinical stages
of AD separately. However, upon attempting this, we found that
this would make the presentation and interpretation of the results
extraordinarily difficult, and the selection of a control group of AD
patients presented considerable difficulties. In terms of AD model
construction and training, we were hampered by the fact that public
databases do not categorize the Aβand Tau protein families in great
detail. Other than that, we mainly chose the Xgboost algorithm
and SVM algorithm, which have excellent performance in solving
classification problems. We have also tried other algorithms and
will try to use more algorithms to try to build AD diagnostic
models in the future (Zhang et al., 2023). In addition to this, our
model construction has not been externally validated due to lack
of data from other database sources. However, because the data
provided by the ADNI database is collected from multiple centers
and has a large sample size, its accuracy and reliability for real-
world application will be guaranteed. In the meantime, we are
currently in the process of clinically collecting additional data to
complete external validation.
Data availability statement
The original contributions presented in the study are included
in the article/supplementary material, further inquiries can be
directed to the corresponding author.
Frontiers in Aging Neuroscience 09 frontiersin.org
Zhen et al. 10.3389/fnagi.2024.1513930
Ethics statement
All participating institutions’ Institutional Review Boards have
given their approval for the ADNI trial. All participants, or
their authorized representatives, have given written informed
permission to ADNI in compliance with the Declaration of
Helsinki. The studies were conducted in accordance with the
local legislation and institutional requirements. Written informed
consent for participation was not required from the participants or
the participants’ legal guardians/next of kin in accordance with the
national legislation and institutional requirements.
Author contributions
WZ: Conceptualization, Data curation, Formal analysis,
Funding acquisition, Investigation, Methodology, Project
administration, Resources, Software, Supervision, Validation,
Visualization, Writing – original draft, Writing – review &
editing. YuW: Data curation, Resources, Writing – review &
editing. HZ: Methodology, Resources, Software, Writing – review
& editing. WZha: Data curation, Resources, Writing – review
& editing. WS: Data curation, Resources, Writing – review &
editing. YS: Data curation, Resources, Writing – review & editing.
YQ: Data curation, Resources, Writing – review & editing. SJ:
Data curation, Resources, Writing – review & editing. ZZ: Data
curation, Resources, Writing – review & editing. YuyW: Data
curation, Resources, Writing – review & editing. LC: Data curation,
Writing – review & editing. JZ: Data curation, Writing – review
& editing. DP: Conceptualization, Formal analysis, Funding
acquisition, Methodology, Project administration, Resources,
Software, Supervision, Validation, Visualization, Writing – review
& editing.
Funding
The author(s) declare financial support was received for the
research, authorship, and/or publication of this article. This work
was supported by National Key R&D Program of China (Grant
No. 2022YFC2010103), Central health research project (Grant
No. 2020ZD10), and 2023 Independent Research Projects for
Graduate Students of Beijing University of Chinese Medicine
(Grant No. ZJKT2023116).
Conflict of interest
The authors declare that the research was conducted
in the absence of any commercial or financial relationships
that could be construed as a potential conflict
of interest.
Generative AI statement
The author(s) declare that no Gen AI was used in the creation
of this manuscript.
Publisher’s note
All claims expressed in this article are solely those of the
authors and do not necessarily represent those of their affiliated
organizations, or those of the publisher, the editors and the
reviewers. Any product that may be evaluated in this article, or
claim that may be made by its manufacturer, is not guaranteed or
endorsed by the publisher.
References
Alzheimer’s Association Report (2022). 2022 Alzheimer’s disease facts and figures.
Alzheimers. Dement. 18, 700–789. doi: 10.1002/alz.12638
Alzheimer’s Association Report (2023). 2023 Alzheimer’s disease facts and figures.
Alzheimers. Dement. 19, 1598–1695. doi: 10.1002/alz.13016
Alzheimer’s Association Report (2024). 2024 Alzheimer’s disease facts and figures.
Alzheimers. Dement. 20, 3708–3821. doi: 10.1002/alz.13809
Ashton, N. J., Pascoal, T. A., Karikari, T. K., Benedet, A. L., Lantero-
Rodriguez, J., Brinkmalm, G., et al. (2021). Plasma p-tau231: a new biomarker
for incipient Alzheimer’s disease pathology. Acta Neuropathol. 141, 709–724.
doi: 10.1007/s00401-021-02275-6
Bai, B., Vanderwall, D., Li, Y., Wang, X., Poudel, S., Wang, H., et al. (2021).
Proteomic landscape of Alzheimer’s disease: novel insights into pathogenesis and
biomarker discovery. Mol. Neurodegener. 16:55. doi: 10.1186/s13024-021-00474-z
Barthélemy, N. R., Li, Y., Joseph-Mathurin, N., Gordon, B. A., Hassenstab, J.,
Benzinger, T. L. S., et al. (2020). A soluble phosphorylated tau signature links tau,
amyloid and the evolution of stages of dominantly inherited Alzheimer’s disease. Nat.
Med. 26, 398–407. doi: 10.1038/s41591-020-0781-z
Binder, J., Ursu, O., Bologa, C., Jiang, S., Maphis, N., Dadras, S., et al.
(2022). Machine learning prediction and tau-based screening identifies
potential Alzheimer’s disease genes relevant to immunity. Commun. Biol. 5:125.
doi: 10.1038/s42003-022-03068-7
Boza-Serrano, A., Vrillon, A., Minta, K., Paulus, A., Camprubí-Ferrer, L., Garcia,
M., et al. (2022). Galectin-3 is elevated in CSF and is associated with Aβdeposits and
tau aggregates in brain tissue in Alzheimer’s disease. Acta. Neuropathol. 144, 843–859.
doi: 10.1007/s00401-022-02469-6
Busche, M. A., and Hyman, B. T. (2020). Synergy between amyloid-βand tau in
Alzheimer’s disease. Nat. Neurosci. 23, 1183–1193. doi: 10.1038/s41593-020-0687-6
Chiu, M. J., Yang, S. Y., Chen, T. F., Lin, C. H., Yang, F. C., Chen, W. P., et al. (2021).
Synergistic association between plasma Aβ(1-42) and p-tau in Alzheimer’s disease
but not in Parkinson’s disease or frontotemporal dementia. ACS Chem. Neurosci. 12,
1376–1383. doi: 10.1021/acschemneuro.1c00010
Cogswell, P. M., Lundt, E. S., Therneau, T. M., Wiste, H. J., Graff-Radford, J.,
Algeciras-Schimnich, A., et al. (2024). Modeling the temporal evolution of plasma
p-tau in relation to amyloid beta and tau PET. Alzheimers. Dement. 20, 1225–1238.
doi: 10.1002/alz.13539
Ding, C., Bao, T. Y., and Huang, H. L. (2022). Quantum-inspired support
vector machine. IEEE Trans. Neural Netw. Learn. Syst. 33, 7210–7222.
doi: 10.1109/TNNLS.2021.3084467
Ferrari-Souza, J. P., Ferreira, P. C. L., Bellaver, B., Tissot, C., Wang, Y. T., Leffa, D.
T., et al. (2022). Astrocyte biomarker signatures of amyloid-βand tau pathologies in
Alzheimer’s disease. Mol. Psychiatry 27, 4781–4789. doi: 10.1038/s41380-022-01716-2
Ferreiro, A. L., Choi, J., Ryou, J., Newcomer, E. P., Thompson, R., Bollinger, R.
M., et al. (2023). Gut microbiome composition may be an indicator of preclinical
Alzheimer’s disease. Sci. Transl. Med. 15:eabo2984. doi: 10.1126/scitranslmed.abo2984
Ficiarà, E., Boschi, S., Ansari, S., D’agata, F., Abollino, O., Caroppo, P., et al.
(2021). Machine learning profiling of Alzheimer’s disease patients based on current
cerebrospinal fluid markers and iron content in biofluids. Front. Aging Neurosci.
13:607858. doi: 10.3389/fnagi.2021.607858
Franciotti, R., Nardini, D., Russo, M., Onofrj, M., and Sensi, S. L. (2023).
Comparison of machine learning-based approaches to predict the conversion to
Frontiers in Aging Neuroscience 10 frontiersin.org
Zhen et al. 10.3389/fnagi.2024.1513930
Alzheimer’s disease from mild cognitive impairment. Neuroscience 514, 143–152.
doi: 10.1016/j.neuroscience.2023.01.029
Gaetani, L., Bellomo, G., Parnetti, L., Blennow, K., Zetterberg, H., and Di Filippo,
M. (2021). Neuroinflammation and Alzheimer’s disease: a machine learning approach
to CSF proteomics. Cells 10:1930. doi: 10.3390/cells10081930
Gao, F., Lv, X., Dai, L., Wang, Q., Wang, P., Cheng, Z., et al. (2023). A combination
model of AD biomarkers revealed by machine learning precisely predicts Alzheimer’s
dementia: China Aging and Neurodegenerative Initiative (CANDI) study. Alzheimers.
Dement. 19, 749–760. doi: 10.1002/alz.12700
Hammond, T. C., Xing, X., Wang, C., Ma, D., Nho, K., Crane, P. K., et al. (2020). β-
amyloid and tau drive early Alzheimer’s disease decline while glucose hypometabolism
drives late decline. Commun Biol 3:352. doi: 10.1038/s42003-020-1079-x
Hampel, H., Blennow, K., Shaw, L. M., Hoessler, Y. C., Zetterberg, H., and
Trojanowski, J. Q. (2010). Total and phosphorylated tau protein as biological markers
of Alzheimer’s disease. Exp. Gerontol. 45, 30–40. doi: 10.1016/j.exger.2009.10.010
Horie, K., Salvad,ó, G., Barthélemy, N. R., Janelidze, S., Li, Y., He, Y., et al. (2023).
CSF MTBR-tau243 is a specific biomarker of tau tangle pathology in Alzheimer’s
disease. Nat. Med. 29, 1954–1963. doi: 10.1038/s41591-023-02443-z
Huang, S., Cai, N., Pacheco, P. P., Narrandes, S., Wang, Y., and Xu, W. (2018).
Applications of support vector machine (SVM) learning in cancer genomics. Cancer
Genom. Proteom. 15, 41–51. doi: 10.21873/cgp.20063
Izzo, N. J., Yuede, C. M., Labarbera, K. M., Limegrover, C. S., Rehak, C., Yurko,
R., et al. (2021). Preclinical and clinical biomarker studies of CT1812: a novel
approach to Alzheimer’s disease modification. Alzheimers. Dement. 17, 1365–1382.
doi: 10.1002/alz.12302
Jang, H., Soroski, T., Rizzo, M., Barral, O., Harisinghani, A., Newton-Mason,
S., et al. (2021). Classification of Alzheimer’s disease leveraging multi-task machine
learning analysis of speech and eye-movement data. Front. Hum. Neurosci. 15:716670.
doi: 10.3389/fnhum.2021.716670
Jia, J., Ning, Y., Chen, M., Wang, S., Yang, H., Li, F., et al. (2024). Biomarker
Changes during 20 Years Preceding Alzheimer’s Disease. N. Engl. J. Med. 390, 712–722.
doi: 10.1056/NEJMoa2310168
Khan, A., Zubair, S., Shuaib, M., Sheneamer, A., Alam, S., and Assiri, B.
(2024). Development of a robust parallel and multi-composite machine learning
model for improved diagnosis of Alzheimer’s disease: correlation with dementia-
associated drug usage and AT(N) protein biomarkers. Front. Neurosci. 18:1391465.
doi: 10.3389/fnins.2024.1391465
Kononikhin, A. S., Zakharova, N. V., Semenov, S. D., Bugrova, A. E., Brzhozovskiy,
A. G., Indeykina, M. I., et al. (2022). Prognosis of Alzheimer’sdisease using quantitative
mass spectrometry of human blood plasma proteins and machine learning. Int. J. Mol.
Sci. 23:7907. doi: 10.3390/ijms23147907
Krokidis, M. G., Vrahatis, A. G., Lazaros, K., Skolariki, K., Exarchos, T. P., and
Vlamos, P. (2023). Machine learning analysis of Alzheimer’s disease single-cell RNA-
sequencing data across cortex and hippocampus regions. Curr. Issues Mol. Biol. 45,
8652–8669. doi: 10.3390/cimb45110544
Küçükali, F., Neumann, A., Van Dongen, J., De Pooter, T., Joris, G., De Rijk, P.,
et al. (2023). Whole-exome rare-variant analysis of Alzheimer’s disease and related
biomarker traits. Alzheimers. Dement. 19, 2317–2331. doi: 10.1002/alz.12842
Kumari, S., Singh, A., Singh, A. K., Yadav, Y., Bajpai, S., Kumar, P., et al. (2022).
Circulatory GSK-3β: blood-based biomarker and therapeutic target for Alzheimer’s
disease. J. Alzheimers. Dis. 85, 249–260. doi: 10.3233/JAD-215347
Li, J., Zhang, Y., Lu, T., Liang, R., Wu, Z., Liu, M., et al. (2022). Identification of
diagnostic genes for both Alzheimer’s disease and Metabolic syndrome by the machine
learning algorithm. Front. Immunol. 13:1037318. doi: 10.3389/fimmu.2022.1037318
Li, K., Wei, Q., Liu, F. F., Hu, F., Xie, A. J., Zhu, L. Q., et al. (2018). Synaptic
dysfunction in Alzheimer’s disease: Aβ, tau, and epigenetic alterations. Mol. Neurobiol.
55, 3021–3032. doi: 10.1007/s12035-017-0533-3
Li, Q., Yang, H., Wang, P., Liu, X., Lv, K., and Ye, M. (2022). XGBoost-based and
tumor-immune characterized gene signature for the prediction of metastatic status in
breast cancer. J. Transl. Med. 20:177. doi: 10.1186/s12967-022-03369-9
Mattsson-Carlgren, N., Andersson, E., Janelidze, S., Ossenkoppele, R., Insel, P.,
Strandberg, O., et al. (2020). Aβdeposition is associated with increases in soluble and
phosphorylated tau that precede a positive Tau PET in Alzheimer’s disease. Sci. Adv.
6:eaaz2387. doi: 10.1126/sciadv.aaz2387
Pang, K., Jiang, R., Zhang, W., Yang, Z., Li, L. L., Shimozawa, M., et al.
(2022). An App knock-in rat model for Alzheimer’s disease exhibiting Aβand
tau pathologies, neuronal death and cognitive impairments. Cell Res. 32, 157–175.
doi: 10.1038/s41422-021-00582-x
Parreño Torres, A., Roncero-Parra, C., Borja, A. L., and Mateo-Sotos, J.
(2023). Inter-hospital advanced and mild Alzheimer’s disease classification based on
electroencephalogram measurements via classical machine learning algorithms. J.
Alzheimers. Dis. 95, 1667–1683. doi: 10.3233/JAD-230525
Petersen, R. C., Aisen, P. S., Beckett, L. A., Donohue, M. C., Gamst, A. C., Harvey,
D. J., et al. (2010). Alzheimer’s Disease Neuroimaging Initiative (ADNI): clinical
characterization. Neurology 74, 201–209. doi: 10.1212/WNL.0b013e3181cb3e25
Pichet Binette, A., Theaud, G., Rheault, F., Roy, M., Collins, D. L., Levin,
J., et al. (2021). Bundle-specific associations between white matter microstructure
and Aβand tau pathology in preclinical Alzheimer’s disease. Elife 10:e62929.
doi: 10.7554/eLife.62929
Sadleir, K. R., and Vassar, R. (2023). Connections between ApoE, sleep,
and Aβand tau pathologies in Alzheimer’s disease. J. Clin. Invest. 133:171838.
doi: 10.1172/JCI171838
Scheltens, P., De Strooper, B., Kivipelto, M., Holstege, H., Chételat, G.,
Teunissen, C. E., et al. (2021). Alzheimer’s disease. Lancet 397, 1577–1590.
doi: 10.1016/S0140-6736(20)32205-4
Shah, J., Rahman Siddiquee, M. M., Krell-Roesch, J., Syrjanen, J. A., Kremers,
W. K., Vassilaki, M., et al. (2023). Neuropsychiatric symptoms and commonly used
biomarkers of Alzheimer’s disease: a literature review from a machine learning
perspective. J. Alzheimers. Dis. 92, 1131–1146. doi: 10.3233/JAD-221261
Shaw, L. M., Vanderstichele, H., Knapik-Czajka, M., Clark, C. M., Aisen, P.
S., Petersen, R. C., et al. (2009). Cerebrospinal fluid biomarker signature in
Alzheimer’s disease neuroimaging initiative subjects. Ann. Neurol. 65, 403–413.
doi: 10.1002/ana.21610
Stern, A. M., Van Pelt, K. L., Liu, L., Anderson, A. K., Ostaszewski, B., Mapstone,
M., et al. (2023). Plasma NT1-tau and Aβ(42) correlate with age and cognitive
function in two large Down syndrome cohorts. Alzheimers. Dement. 19, 5755–5764.
doi: 10.1002/alz.13382
Wang, Y., Sun, Y., Wang, Y., Jia, S., Qiao, Y., Zhou, Z., et al. (2023a).
Urine metabolomics phenotyping and urinary biomarker exploratory in mild
cognitive impairment and Alzheimer’s disease. Front. Aging Neurosci. 15:1273807.
doi: 10.3389/fnagi.2023.1273807
Wang, Y., Sun, Y., Wang, Y., Jia, S., Qiao, Y., Zhou, Z., et al. (2023b). Identification
of novel diagnostic panel for mild cognitive impairment and Alzheimer’s disease:
findings based on urine proteomics and machine learning. Alzheimers. Res. Ther.
15:191. doi: 10.1186/s13195-023-01324-4
Wang, Z., Ye, J., Zhang, K., Ding, L., Granzier-Nakajima, T., Ranasinghe, J. C., et al.
(2022). Rapid biomarker screening of Alzheimer’s disease by interpretable machine
learning and graphene-assisted raman spectroscopy. ACS Nano 16, 6426–6436.
doi: 10.1021/acsnano.2c00538
Winchester, L. M., Harshfield, E. L., Shi, L., Badhwar, A., Khleifat, A. A., Clarke, N.,
et al. (2023). Artificial intelligence for biomarker discovery in Alzheimer’s disease and
dementia. Alzheimers. Dement. 19, 5860–5871. doi: 10.1002/alz.13390
Yakoub, Y., Ashton, N. J., Strikwerda-Brown, C., Montoliu-Gaya, L., Karikari, T.
K., Kac, P. R., et al. (2023). Longitudinal blood biomarker trajectories in preclinical
Alzheimer’s disease. Alzheimers. Dement. 19, 5620–5631. doi: 10.1002/alz.13318
Yang, J., Feng, X., Laine, A. F., and Angelini, E. D. (2020). Characterizing
Alzheimer’s disease with image and genetic biomarkers using supervised topic models.
IEEE J. Biomed. Health Inform. 24, 1180–1187. doi: 10.1109/JBHI.2019.2928831
Yi, F., Yang, H., Chen, D., Qin, Y., Han, H., Cui, J., et al. (2023). XGBoost-SHAP-
based interpretable diagnostic framework for alzheimer’s disease. BMC Med. Inform.
Decis. Mak. 23, 137. doi: 10.1186/s12911-023-02238-9
Yue, S., Li, S., Huang, X., Liu, J., Hou, X., Zhao, Y., et al. (2022). Machine learning
for the prediction of acute kidney injury in patients with sepsis. J. Transl. Med. 20, 215.
doi: 10.1186/s12967-022-03364-0
Zhang, H., Wei, W., Zhao, M., Ma, L., Jiang, X., Pei, H., et al.
(2021). Interaction between Aβand tau in the pathogenesis of
Alzheimer’s disease. Int. J. Biol. Sci. 17, 2181–2192. doi: 10.7150/ijbs.
57078
Zhang, Y., Miao, Y., Tan, J., Chen, F., Lei, P., and Zhang, Q. (2023). Identification
of mitochondrial related signature associated with immune microenvironment
in Alzheimer’s disease. J. Transl. Med. 21, 458. doi: 10.1186/s12967-023-0
4254-9
Zhang, Z., Li, X. G., Wang, Z. H., Song, M., Yu, S. P., Kang, S. S.,
et al. (2021). δ-Secretase-cleaved tau stimulates Aβproduction via upregulating
STAT1-BACE1 signaling in Alzheimer’s disease. Mol. Psychiatry 26, 586–603.
doi: 10.1038/s41380-018-0286-z
Frontiers in Aging Neuroscience 11 frontiersin.org
