Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Open Access

Peer-reviewed

Research Article

Proteomic associations with cognitive variability as measured by the Wisconsin Card Sorting Test in a healthy Thai population: A machine learning approach

Abstract

Inter-individual cognitive variability, influenced by genetic and environmental factors, is crucial for understanding typical cognition and identifying early cognitive disorders. This study investigated the association between serum protein expression profiles and cognitive variability in a healthy Thai population using machine learning algorithms. We included 199 subjects, aged 20 to 70, and measured cognitive performance with the Wisconsin Card Sorting Test. Differentially expressed proteins (DEPs) were identified using label-free proteomics and analyzed with the Linear Model for Microarray Data. We discovered 213 DEPs between lower and higher cognition groups, with 155 upregulated in the lower cognition group and enriched in the IL-17 signaling pathway. Subsequent bioinformatic analysis linked these DEPs to neuroinflammation-related cognitive impairment. A random forest model classified cognitive ability groups with an accuracy of 81.5%, sensitivity of 65%, specificity of 85.9%, and an AUC of 0.79. By targeting a specific Thai cohort, this research provides novel insights into the link between neuroinflammation and cognitive performance, advancing our understanding of cognitive variability, highlighting the role of biological markers in cognitive function, and contributing to developing more accurate machine learning models for diverse populations.

Citation: Chen C, Khanthiyong B, Thaweetee-Sukjai B, Charoenlappanit S, Roytrakul S, Surit P, et al. (2025) Proteomic associations with cognitive variability as measured by the Wisconsin Card Sorting Test in a healthy Thai population: A machine learning approach. PLoS ONE 20(2): e0313365. https://doi.org/10.1371/journal.pone.0313365

Editor: Nafisa M. Jadavji, Southern Illinois University Carbondale, UNITED STATES MINOR OUTLYING ISLANDS

Received: October 23, 2024; Accepted: January 21, 2025; Published: February 20, 2025

Copyright: © 2025 Chen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The author-generated code and the datasets generated and/or analyzed during the current study are uploaded as supplementary files.

Funding: S.N-T. received partial support from the Global and Frontier Research University Fund, Naresuan University (grant number R2567C003; https://www.nu.ac.th/) and the Reinventing University Program 2024, Ministry of Higher Education, Science, Research and Innovation (MHESI), Thailand (grant number R2567A141; https://www.mhesi.go.th/). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. There was no additional external funding received for this study.

Competing interests: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Introduction

Cognitive function denotes the higher-order mental processes in the brain that gather and process information, reflecting brain activity. In adults, inter-individual cognitive variability is pervasive [1], influenced by both genetic and environmental factors [2,3], including gender, age [4], and lifestyle choices [3]. For example, variability in cognitive trajectories has been observed in community-dwelling older adults across different cognitive domains, such as episodic memory, vocabulary, executive function, attention, and psychomotor speed [5]. Differences in cognitive responses to the same physical exercise among individuals are also apparent [6,7]. Studying normal cognitive variability is essential for establishing a baseline understanding of typical cognition, which is crucial for identifying anomalies that may signal early cognitive disorders [8]. Additionally, understanding normal cognitive differences aids in developing strategies and educational practices that ensure the society is tailored to support diverse cognitive strengths [9].

Genetic and environmental factors may exhibit complex direct and indirect interactions that influence normal cognitive variability. Genetic factors contribute approximately 50% to 70% of cognitive variability at the population level. However, genetic influences on cognitive function increase from birth to maturity, with these effects being pronounced in more advantaged socioeconomic groups [10]. Previous research indicates that differences in cognitive flexibility between healthy Thai males and females fluctuate across age, with such cognitive sex differences notable in subjects over 60 years [11]. Additionally, experiences and knowledge gained through education alter activity in cholinergic pathways, leading to an attenuation of these sex-dependent cognitive differences [12]. This complexity poses a significant challenge for researchers attempting to unravel the contributing elements and their interactions that influence normal cognitive variability. Thus, using information technology and computer-based algorithms to probe these various complex interconnections offers great potential for enhancing our understanding of cognitive variability.

Machine learning (ML) algorithms serve as robust approaches for processing large-scale datasets and detecting intricate patterns that conventional statistical methods might fail to reveal [13,14]. ML models have been employed to classify cognitive profiles [15–17] and to predict cognitive health in the global population [18–20]. When using ML algorithms, the racial and ethnic background of subjects is an essential issue to consider [21], since racial bias is a prevalent challenge facing ML in human studies [22], with consequences that can lead to racial disparities in healthcare access and outcomes [23].

Given that cognitive function reflects brain activity, alterations in molecular factors such as neuronal and neurotransmission-related proteins may offer valuable insights into cognitive variability in healthy individuals. Indeed, previous studies have demonstrated that biomarkers related to brain activity can serve as indicators of cognitive function [24–26]. Accordingly, this study aims to investigate the association between serum protein expression profiles and one measure of cognitive variability, as assessed by the Wisconsin Card Sorting Test (WCST), in a healthy Thai population using ML algorithms. By targeting a specific Thai cohort, this research contributes to the development of ML models tailored to diverse populations in cognitive studies.

Methods

Participants and data

This study included 199 healthy Thai subjects, ranging in age from 20 to 70 years, with samples collected and cognitive tests conducted from October 20, 2014, to August 25, 2018, as previously reported [27]. The researchers assessed the archived samples and data from July 6, 2022, to July 6, 2023, for the preparation of this publication; all subjects were assigned a study ID that concealed individual identities. The cognitive performance of each participant was evaluated using the WCST, a test that measures cognitive flexibility [28,29]—the capacity to adapt the behavioral response mode to changing conditions—which is commonly employed to assess frontal lobe function, especially the prefrontal cortex [30]. 3 ml of blood was collected from the cubital vein of each subject immediately following the completion of the WCST. The serum protein expression profiles of the subjects were then analyzed using the label-free proteomics method, as previously described [11].

The mean and standard deviation (SD) of the percentage of total errors (%Errors) in WCST were computed in male and female subjects separately, covarying for age [11]. A negative correlation has been observed between WCST %Errors and the Full-Scale Intelligence Quotient (FSIQ) [31,32], a metric used to assess a person’s overall level of general cognitive and intellectual functioning [33]. Male and female subjects with %Errors > 1SD above their sex-specific mean were considered to have poor cognitive performance and were assigned to the lower cognitive ability group [34]. All methods were performed in compliance with relevant guidelines and regulations (Declaration of Helsinki). The Institutional Review Board (IRB) of Naresuan University, Thailand approved this research (COA No. 0262/2022). Participation was voluntary, and written informed consent was obtained from each subject.

Bioinformatic analysis

Differentially expressed proteins (DEPs) between the two cognition groups were identified using the Linear Model for Microarray Data (LIMMA) approach in R version 4.2.3. The LIMMA approach applies linear modeling to the expression data and employs empirical Bayesian techniques to adjust the standard errors of the estimated log-fold changes [35]. The significance of the differential expression was determined using the adjusted P-value, with a threshold set at P ≤ 0.01 [36]. This statistical rigor enhances the validity of our findings, ensuring that the results are both robust and biologically meaningful.

Pathway analysis and Gene Ontology (GO) enrichment analysis were performed on the website of Annotation, Visualization, and Integrated Discovery (DAVID). DAVID utilizes Fisher’s Exact Test to assess the enrichment of GO terms and pathways among the DEPs, ensuring that these results are statistically significant and reliable [37,38]. The protein-protein interaction network of the identified DEPs was studied using Pathway Studio version 12.5 [39]. DEP expression levels were displayed by Multi-Experiment Viewer (MeV) version 4.9.0 [40]. P ≤ 0.05 was considered significant.

Preprocessing

The training and testing datasets were respectively proportioned at 0.6 (n = 119) and 0.4 (n = 80) of the total sample to optimize the balance between training and validation sets. This approach is designed to enhance the model’s generalization capability and minimize the risk of overfitting. By allocating a sufficient portion of the data to the training set while preserving a substantial validation set, we can achieve more reliable and robust model performance, consistent with best practices in ML [41].

To address the relatively small proportion of subjects defined as having poor cognitive performances, the synthetic minority over-sampling technique (SMOTE) was employed [42]. SMOTE was applied exclusively to the training dataset during cross-validation to minimize overfitting. The testing dataset, excluded from SMOTE or cross-validation procedures, was used to evaluate the final performance metrics of the model [18].

The DEPs that were significantly enriched in cognition-related pathways, as identified by GO enrichment analysis, pathway analysis, and protein-protein interaction network analysis, were selected as model variables. Age was also included in the model since it is strongly connected to cognitive impairment in the healthy population [43,44].

Machine learning model

The ML processes were conducted using the tidy models meta-package in R [45]. This approach facilitated the creation of a unitary preprocessing dataset, enabling reliable comparisons across various ML models, including K-nearest neighbors, regression, decision tree, and random forest (RF). The RF model was selected for this study due to its superior performance. The analysis code can be found in S1 File.

The principles and procedures of the RF model are well-documented [46–48]. In brief, the RF is an ensemble learning method based on decision tree algorithms and is applicable to both classification and regression analysis [49]. The significance of each variable in the final model is assessed using a ranked measure of variable importance.

Model validation

Cross-validation is a technique employed to evaluate the performance and generalizability of an ML model. It involves dividing the data into multiple subsets, training the model on certain subsets, and validating it on others. This method minimizes overfitting and yields a more accurate estimate of model performance [50]. In this study, a 10-fold cross-validation was conducted on the training dataset to evaluate the models’ performances.

In addition, model hyperparameters were tuned during the cross-validation [45]. Specifically, the following hyperparameters were tuned in this study: the number of trees was 1000, the minimum number of data points required for a node further splitting was 2, and 10 predictors were randomly sampled at each split when building the tree models.

Performance metrics

In this study, we utilized several performance metrics to evaluate the RF model comprehensively. Overall accuracy measures the proportion of correctly classified instances among all instances. Matthews correlation coefficient (MCC) is a balanced measure that accounts for true and false positives and negatives, providing a comprehensive understanding of model performance [51]. The F1-score, which is the harmonic mean of precision and recall, offers insight into the balance between the precision and the completeness of the model [52]. Finally, the Area Under the Receiver Operating Characteristic Curve (AUC) assesses the model’s ability to distinguish between classes, offering a summary of performance across all classification thresholds [53]. Fig 1 shows the overall working procedure of this study.

thumbnail
Download:
Fig 1. Overall working procedure of this study.

https://doi.org/10.1371/journal.pone.0313365.g001

Results

Demographic data of the study population

The demographic characteristics of the study population are detailed in Table 1 (see full data set in S1 Table). The mean age of the participants was 45.6 ± 19.3 years, with 55.3% being female (n = 110). The cutoff point as defined by %Errors > 1SD from the mean in males was 62.3, whereas in females, it was 67.4.

thumbnail
Download:
Table 1. Demographic data of the study population.

https://doi.org/10.1371/journal.pone.0313365.t001

Bioinformatic analysis

There were 213 differentially expressed proteins identified between the poor and higher cognition groups, with 155 DEPs being upregulated in the poor cognition group and the remaining 58 DEPs being downregulated (see Fig 2). Data from the subsequent GO enrichment analysis demonstrated that DEPs were substantially enriched in the following biological pathways: regulation of protein stability (P = 0.01), macromolecule methylation (P = 0.03), regulation of neuron projection development (P = 0.04), and retinoic acid catabolic process (P = 0.04), as shown in Table 2. Regarding the pathway analysis [54–56], these DEPs showed significant enrichment in the IL-17 signaling pathway (P = 0.05); see Table 3.

thumbnail
Download:
Fig 2. Volcano plot of DEPs between poor and normal cognition groups.

Color blue represents DEPs that are downregulated in the poor cognition group, red represent DEPs unregulated in poor cognition group.

https://doi.org/10.1371/journal.pone.0313365.g002

thumbnail
Download:
Table 2. Gene Ontology analysis of DEPs between lower and higher cognitive ability groups.

https://doi.org/10.1371/journal.pone.0313365.t002

thumbnail
Download:
Table 3. Pathway analysis of DEPs between lower and higher cognitive ability groups.

https://doi.org/10.1371/journal.pone.0313365.t003

Furthermore, protein-protein interaction (PPI) network analysis indicated that most of the 16 DEPs enriched in the aforementioned pathways were linked to neuroinflammation-related cognitive impairment (see full data set in S1 Table). The other four DEPs found to be involved in this PPI network were serotonin receptor 7 (HTR7), metabotropic glutamate receptor 4 (GRM4), choline transporter-like protein 2 (SLC44A2), and pro-adrenomedullin (ADM) (see Fig. 3). Glutamatergic [57], serotonergic [58], and cholinergic [59] systems have been shown to have a role in cognitive impairment. Additionally, ADM has been suggested as a potential biomarker for cognitive impairment [60,61]. As a result, all those 20 proteins were included as variables in the RF model.

thumbnail
Download:
Fig 3. Protein-protein interaction network of selected DEPs.

https://doi.org/10.1371/journal.pone.0313365.g003

Model performance

The RF model achieved a test classification accuracy of 81.5% (see Table 4). The model’s sensitivity (true positive rate) was estimated to be 65%, while the specificity (true negative rate) was 85.9%. The AUC (0.79) indicates good binary classification performance [62] (see Fig. 4).

thumbnail
Download:
Table 4. Performance metrics of the RF model.

https://doi.org/10.1371/journal.pone.0313365.t004

thumbnail
Download:
Fig 4. Receiver operating characteristic curve of the RF model.

https://doi.org/10.1371/journal.pone.0313365.g004

Importance of the model variables

Fig 5 depicts the 10 most influential variables. MAPK9 (P45984) had the strongest influence on the classification model, and this DEP was significantly enriched in the IL-17 signaling pathway. Age, ADM (P35318), HTR7 (P34969), SLC44A2 (Q8IWA5), and GRM4 (Q14833) are among the most important variables, as expected. The remaining four variables are NTNG1 (Q9y2I2), FTSG3 (Q8IY81), CCAR2 (Q8N163), and UPS25 (Q9UHP3), in descending order of importance. In the poor cognition group, 7 of those 9 DEPs were upregulated, whereas SLC44A2 and GRM4 were downregulated (see Fig. 6).

thumbnail
Download:
Fig 5. Significance of variables in the RF model.

https://doi.org/10.1371/journal.pone.0313365.g005

thumbnail
Download:
Fig 6. Heat map of the top 9 protein variables.

https://doi.org/10.1371/journal.pone.0313365.g006

Discussion

This study built a classification model using ML algorithms to detect healthy Thai subjects with poor cognitive performance based on proteomic data. The RF model showed good performance with an accuracy of 0.815, a specificity of 0.859, and an AUC of 0.790, despite a limited sensitivity of 0.65. Overall, the screening performance of our classifier is considered acceptable [62,63]. This performance surpasses previous studies that utilized ML algorithms to study cognitive function in a Thai population. For example, a study using cardiovascular risk factors to compute the probability of mild cognitive impairment (MCI) in the Thai population reported AUCs ranging from 0.58 to 0.61 [64]. Another study applied phonemic verbal fluency (PVF) tasks combined with ML techniques to predict MCI in Thai participants, achieving an AUC of 0.73 [65]. Additionally, research on the Thai version of the CERAD neuropsychological battery for MCI screening reported the best model performance with an AUC of 0.77 [66]. Our integration of serum proteomic data provided a deeper insight into the value of biological markers for enhancing model precision for cognitive outcomes in the Thai population. However, a recent study leveraging high-dimensional neuroimaging data to classify cognitive performance in Portuguese individuals achieved an accuracy of 86.67% [67]. Their higher accuracy compared to our RF model may be attributed to the direct relevance of neuroimaging features to cognition, while proteomic data reflects broader systemic processes that may introduce additional variability. Nevertheless, due to its limited sensitivity and moderate F1-score [68], the performance of the RF model should be interpreted with caution. The low sensitivity was anticipated owing to the dataset’s skew toward the negative cases (higher cognitive ability) [69,70] and the moderate sample size [71,72].

MAPK9 had the most impact on the RF model; it was enriched in the IL-17 signaling pathway along with another key variable, UPS25, with both DEPs being upregulated in the lower cognitive ability group. IL-17 signaling regulates inflammation by modulating inflammatory gene expression [73]. These proinflammatory factors, if unrestrained, may contribute to the pathology of a variety of autoimmune and chronic inflammatory conditions [74]. Dysfunctional and persistent inflammatory processes have been suggested as potential causative drivers of impaired cognitive functioning [75,76]. This is consistent with previous findings that proinflammatory molecules can cause the progression of brain deficits [77], and IL-17 reportedly initiates the onset of synaptic and cognitive impairments in the early stages of Alzheimer’s disease [78], a process that may be mediated by activation of IL-17 receptors and MAPK [79].

Furthermore, serotonin receptor 7 (HTR7), metabotropic glutamate receptor-4 (GRM4), and choline transporter-like protein 2 (SLC44A2) were also among the most influential variables. It has been demonstrated that inhibiting HTR7 modulates immune responses and decreases the severity of intestinal inflammation [80]. Serotonin stimulation of HTR7 activates downstream signaling modules such as MEK/MAPK [81]. This suggested that HTR7 may be positively connected with IL-17-mediated neuroinflammation and the poor cognitive functioning resulting from brain inflammation. This notion is supported in part by a prior result that HTR7 antagonism may have positive effects on schizophrenia-like cognitive deficits [82].

According to previous animal research, GRM4 controls adaptive immunity and restrains neuroinflammation [83]. Another study found that mutant mice lacking GRM4 were less capable of learning and integrating new spatial information into previously generated memory traces [84]. The involvement of spatial information learning in the WCST process has been proposed [85]. Choline transporter-like protein 2 (SLC44A2) is a high-affinity choline carrier [86]. Choline is an indispensable constituent in the biosynthesis of acetylcholine (ACh), and its transportation into the presynaptic terminals of cholinergic neurons requires a high-affinity choline transporter [87]. ACh is assumed to play an essential role in executive function, and cholinergic decline is related to poor WCST performance in healthy individuals [88]. Furthermore, SLC44A2 is a newly discovered plasma membrane and mitochondrial ethanolamine transporter [89], and ethanolamine has been linked to anti-inflammatory effects [90]. The downregulation of GRM4 and SLC44A2 in the lower cognitive ability group samples provides a potential biological mechanism for why subjects in this group performed poorly on the WCST. However, the weights of SLC44A2 and GRM4 are small, consistent with there being multiple interactions across diverse neurotransmitter systems in maintaining central nervous system homeostasis [91,92].

Netrin-G1 (NTNG1) and pro-adrenomedullin (ADM) are the second and third most important variables, and both DEPs were upregulated in the lower cognitive ability group samples. NTNG1 is a member of the Netrin family that is little studied in the brain but is implicated in inflammatory processes, including microglial function [93]. ADM has long been thought to be a biomarker for cognitive impairment [60,94]. ADM accumulation in the human brain contributes to memory loss with age [61], and blood ADM levels elevate in multiple pathological states, such as acute ischemic stroke and vascular cognitive decline with white matter alterations [95]. There is evidence indicating that inflammatory states stimulate the in vivo production of ADM [96]. These data, together with the evidence shown above, suggest that those in the lower cognitive ability group may have higher levels of neuroinflammation, resulting in poor cognitive performance on WCST. Age ranks as the seventh most important predictor in the model, with a moderate weight. Although it is widely accepted that aging is positively correlated to cognitive decline [44,97], our findings imply that this variable did not emerge as an influential factor in classifying cognitive variability among healthy Thai subjects.

This study has limitations that should be noted. First, the lower and higher cognitive ability groups were defined by their WCST % Errors > 1SD from the mean, but there is a controversy about this cutoff for defining an abnormal score on the WCST [34], and further study with different cutoff points might be needed to replicate the findings of the current study. Second, only age and proteomics data were included in the RF model. Given the potential direct and indirect interactions among multiple factors influencing cognitive function [98], additional factors such as socioeconomic status, lifestyle factors, and cardiovascular risk factors should be included in the model to improve model performance. Third, although the WCST has demonstrated validity and reliability as a stand-alone cognitive assessment tool for healthy individuals of various ages and educational backgrounds [99], and %Errors offers a general measure of performance on the WCST that closely correlates with FSIQ [31], variability in different domains of cognition requires a more comprehensive battery of cognitive testing. This would have been beyond the scope of the current study. Fourth, due to the moderate sample size and imbalanced number of subjects in the lower and higher cognitive ability groups, generalizing the current findings should be done with caution. Last, although the current statistical analysis outlines its benefits, it does not cover additional metrics such as effect size measures and confidence intervals. Including these metrics in future research will provide a more comprehensive evaluation and strengthen the robustness and interpretability of the findings.

Conclusion

This study highlights the significant association between serum protein expression profiles and cognitive variability in a healthy Thai population using ML algorithms. By identifying 213 DEPs, with 155 upregulated in the lower cognition group and enriched in the IL-17 signaling pathway, our findings indicate a strong link between neuroinflammation and cognitive performance differences. The RF model using this proteomic data demonstrates strong classification accuracy and performance, with high specificity, outperforming previous studies that employed ML algorithms to examine cognitive function in Thai populations. This research underscores the critical role of biological markers in enhancing cognitive prediction and contributes to the development of more accurate ML models for diverse populations. These findings advance our understanding of cognitive variability and pave the way for early detection of cognitive disorders, offering promising directions for future research in cognitive health.

Supporting information

S1 Table. WCST scores and protein expression levels.

https://doi.org/10.1371/journal.pone.0313365.s001

(XLSX)

S1 File. Author-generated code.

https://doi.org/10.1371/journal.pone.0313365.s002

(DOCX)

Acknowledgments

The authors would like to thank all participants in this study. We would also like to thank the Faculty of Medical Science, Naresuan University and The National Center for Genetic Engineering and Biotechnology, Pathum Thani, Thailand for the facility supports.

References

  1. 1. Judd N, Aristodemou M, Klingberg T, Kievit R. Interindividual differences in cognitive variability are ubiquitous and distinct from mean performance in a battery of eleven tasks. J Cogn. 2024;7(1):45. pmid:38799081
  2. 2. Davies G, Armstrong N, Bis JC, Bressler J, Chouraki V, Giddaluru S, et al; Generation Scotland. Genetic contributions to variation in general cognitive function: a meta-analysis of genome-wide association studies in the CHARGE consortium (N=53 949). Mol Psychiatry. 2015;20(2):183–92. pmid:25644384
  3. 3. Hsu H-C, Bai C-H. Individual and environmental factors associated with cognitive function in older people: a longitudinal multilevel analysis. BMC Geriatr. 2022;22(1):243. pmid:35321640
  4. 4. Whitley E, Deary IJ, Ritchie SJ, Batty GD, Kumari M, Benzeval M. Variations in cognitive abilities across the life course: cross-sectional evidence from understanding society: the UK household longitudinal study. Intell. 2016;59:39–50. pmid:27932853
  5. 5. Wu Z, Woods RL, Wolfe R, Storey E, Chong TTJ, Shah RC, et al; ASPREE Investigator Group. Trajectories of cognitive function in community-dwelling older adults: a longitudinal study of population heterogeneity. Alzheimers Dement (Amst). 2021;13(1):e12180. pmid:33969173
  6. 6. Schwarck S, Schmicker M, Dordevic M, Rehfeld K, Müller N, Müller P. Inter-individual differences in cognitive response to a single bout of physical exercise-a randomized controlled cross-over study. J Clin Med. 2019;8(8):1101. pmid:31349593
  7. 7. Herold F, Törpel A, Hamacher D, Budde H, Zou L, Strobach T, et al. Causes and consequences of interindividual response variability: a call to apply a more rigorous research design in acute exercise-cognition studies. Front Physiol. 2021;12:682891. pmid:34366881
  8. 8. Ku P-H, Yang Y-R, Yeh N-C, Li P-Y, Lu C-F, Wang R-Y. Prefrontal activity and heart rate variability during cognitive tasks may show different changes in young and older adults with and without mild cognitive impairment. Front Aging Neurosci. 2024;16:1392304. pmid:38863782
  9. 9. Boa Sorte Silva NC, ten Brinke LF, Bielak AAM, Handy TC, Liu-Ambrose T. Improved intraindividual variability in cognitive performance following cognitive and exercise training in older adults. J Int Neuropsychol Soc. 2024;30(4):328–38. 37860873. PMID
  10. 10. Tucker-Drob EM, Briley DA, Harden KP. Genetic and environmental influences on cognition across development and context. Curr Dir Psychol Sci. 2013;22(5):349–55. pmid:24799770
  11. 11. Chen C, Khanthiyong B, Thaweetee-Sukjai B, Charoenlappanit S, Roytrakul S, Thanoi S, et al. Proteomic association with age-dependent sex differences in Wisconsin Card Sorting Test performance in healthy Thai subjects. Sci Rep. 2023;13(1):20238. pmid:37981639
  12. 12. Chen C, Khanthiyong B, Charoenlappanit S, Roytrakul S, Reynolds GP, Thanoi S, et al. Cholinergic-estrogen interaction is associated with the effect of education on attenuating cognitive sex differences in a Thai healthy population. PLoS One. 2023;18(7):e0278080. pmid:37471329
  13. 13. Bhatt P, Sethi A, Tasgaonkar V, Shroff J, Pendharkar I, Desai A, et al. Machine learning for cognitive behavioral analysis: datasets, methods, paradigms, and research directions. Brain Inform. 2023;10(1):18. pmid:37524933
  14. 14. Thiele JA, Faskowitz J, Sporns O, Hilger K. Choosing explanation over performance: Insights from machine learning-based prediction of human intelligence from brain connectivity. PNAS Nexus. 2024;3(12):pgae519. pmid:39660075
  15. 15. Alahmadi HH, Shen Y, Fouad S, Luft CD, Bentham P, Kourtzi Z, et al. Classifying Cognitive Profiles Using Machine Learning with Privileged Information in Mild Cognitive Impairment. Front Comput Neurosci. 2016;10:117. pmid:27909405
  16. 16. Ahmed MU, Begum S, Gestlöf R, Rahman H, Sörman J, editors. Machine learning for cognitive load classification – a case study on contact-free approach. Artificial intelligence applications and innovations 2020. Cham: Springer International Publishing; 2020.
    • 17. Sun J-W, Fan R, Wang Q, Wang Q-Q, Jia X-Z, Ma H-B. Identify abnormal functional connectivity of resting state networks in Autism spectrum disorder and apply to machine learning-based classification. Brain Res. 2021;1757:147299. pmid:33516816
    • 18. Na KS. Prediction of future cognitive impairment among the community elderly: a machine-learning based approach. Sci Rep. 2019;9(1):3335. pmid:30833698
    • 19. Cremers LGM, Huizinga W, Niessen WJ, Krestin GP, Poot DHJ, Ikram MA, et al. Predicting global cognitive decline in the general population using the disease state index. Front Aging Neurosci. 2019;11:379. pmid:32038225
    • 20. Poudel GR, Barnett A, Akram M, Martino E, Knibbs LD, Anstey KJ, et al. Machine learning for prediction of cognitive health in adults using sociodemographic, neighbourhood environmental, and lifestyle factors. Int J Environ Res Public Health. 2022;19(17):10977. pmid:36078704
    • 21. Desaire H, Stepler KE, Robinson RAS. Exposing the brain proteomic signatures of Alzheimer’s disease in diverse racial groups: leveraging multiple data sets and machine learning. J Proteome Res. 2022;21(4):1095–104. pmid:35276041
    • 22. Ntoutsi E, Fafalios P, Gadiraju U, Iosifidis V, Nejdl W, Vidal M-E, et al. Bias in data-driven artificial intelligence systems—an introductory survey. WIREs Data Min Knowl Discovery. 2020;10(3):e1356.
    • 23. Obermeyer Z, Powers B, Vogeli C, Mullainathan S. Dissecting racial bias in an algorithm used to manage the health of populations. Science. 2019;366(6464):447–53. pmid:31649194
    • 24. Tian F, Wang Y, Qian ZM, Ran S, Zhang Z, Wang C, et al. Plasma metabolomic signature of healthy lifestyle, structural brain reserve and risk of dementia. Brain. 2024.
    • 25. Kivisäkk P, Carlyle BC, Sweeney T, Trombetta BA, LaCasse K, El-Mufti L, et al. Plasma biomarkers for diagnosis of Alzheimer’s disease and prediction of cognitive decline in individuals with mild cognitive impairment. Front Neurol. 2023;14:1069411. pmid:36937522
    • 26. Xiao Z, Wu X, Wu W, Yi J, Liang X, Ding S, et al. Plasma biomarker profiles and the correlation with cognitive function across the clinical spectrum of Alzheimer’s disease. Alzheimers Res Ther. 2021;13(1):123. pmid:34225797
    • 27. Khanthiyong B, Thanoi S, Reynolds GP, Nudmamud-Thanoi S. Association study of the functional Catechol-O-Methyltranferase (COMT) Val(158)Met polymorphism on executive cognitive function in a Thai sample. Int J Med Sci. 2019;16(11):1461–5. pmid:31673237
    • 28. Nagahama Y, Fukuyama H, Yamauchi H, Katsumi Y, Magata Y, Shibasaki H, et al. Age-related changes in cerebral blood flow activation during a Card Sorting Test. Exp Brain Res. 1997;114(3):571–7. pmid:9187292
    • 29. Nyhus E, Barcelo F. The Wisconsin Card Sorting Test and the cognitive assessment of prefrontal executive functions: a critical update. Brain Cogn. 2009;71(3):437–51. pmid:19375839
    • 30. Konishi S, Nakajima K, Uchida I, Kameyama M, Nakahara K, Sekihara K, et al. Transient activation of inferior prefrontal cortex during cognitive set shifting. Nat Neurosci. 1998;1(1):80–4. pmid:10195114
    • 31. Kopp B, Maldonado N, Scheffels JF, Hendel M, Lange F. a meta-analysis of relationships between measures of Wisconsin Card Sorting and intelligence. Brain Sciences. 2019;9(12):349. pmid:31795503
    • 32. Lung FW, Chen PF, Shu BC. Performance of Wisconsin Card Sorting Test in five-year-old children in Taiwan: relationship to intelligence and cognitive development. PLoS One. 2018;13(8):e0202099. pmid:30161263
    • 33. Lange RT. Full Scale IQ. In: Kreutzer JS, DeLuca J, Caplan B, editors. Encyclopedia of Clinical Neuropsychology. New York, NY: Springer New York; 2011. p. 1103–5.
      • 34. Gunner JH, Miele AS, Lynch JK, McCaffrey RJ. Performance of non-neurological older adults on the Wisconsin Card Sorting Test and the Stroop Color-Word Test: normal variability or cognitive impairment? Arch Clin Neuropsychol. 2012;27(4):398–405. pmid:22591916
      • 35. Smyth GK. limma: linear models for microarray data. In: Gentleman R, Carey VJ, Huber W, Irizarry RA, Dudoit S, editors. Bioinformatics and computational biology solutions using r and bioconductor. statistics for biology and health. New York, NY: Springer New York; 2005. p. 397-420.
        • 36. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Series B Stat Methodol. 1995;57(1):289–300.
        • 37. Sherman BT, Hao M, Qiu J, Jiao X, Baseler MW, Lane HC, et al. DAVID: a web server for functional enrichment analysis and functional annotation of gene lists (2021 update). Nucleic Acids Res. 2022;50(W1):W216–21. pmid:35325185
        • 38. Huang DW, Sherman BT, Lempicki RA. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc. 2009;4(1):44–57.
        • 39. Nikitin A, Egorov S, Daraselia N, Mazo I. Pathway studio--the analysis and navigation of molecular networks. Bioinformatics. 2003;19(16):2155–7. pmid:14594725
        • 40. Chu VT, Gottardo R, Raftery AE, Bumgarner RE, Yeung KY. MeV+R: using MeV as a graphical user interface for bioconductor applications in microarray analysis. Genome Biol. 2008;9(7):R118. pmid:18652698
        • 41. Dhabarde MS, editor. Approach towards model evaluation, model selection, and algorithm selection in machine learning; 2019.
          • 42. Chawla NV, Bowyer KW, Hall LO, Kegelmeyer WP. SMOTE: synthetic minority over-sampling technique. J Artif Intell Res. 2002;16(1):321–57.
          • 43. Jia J, Zhao T, Liu Z, Liang Y, Li F, Li Y, et al. Association between healthy lifestyle and memory decline in older adults: 10 year, population based, prospective cohort study. BMJ. 2023;380:e072691. pmid:36696990
          • 44. Deary IJ, Corley J, Gow AJ, Harris SE, Houlihan LM, Marioni RE, et al. Age-associated cognitive decline. Br Med Bull. 2009;92(1):135–52. pmid:19776035
          • 45. Kuhn M, Wickham H. Tidymodels: a collection of packages for modeling and machine learning using tidyverse principles; 2020.
            • 46. Acharjee A, Larkman J, Xu Y, Cardoso VR, Gkoutos GV. A random forest based biomarker discovery and power analysis framework for diagnostics research. BMC Med Genomics. 2020;13(1):178. pmid:33228632
            • 47. Chen X, Ishwaran H. Random forests for genomic data analysis. Genomics. 2012;99(6):323–9. pmid:22546560
            • 48. Fan Y, Murphy TB, Byrne JC, Brennan L, Fitzpatrick JM, Watson RWG. Applying random forests to identify biomarker panels in serum 2D-DIGE data for the detection and staging of prostate cancer. J Proteome Res. 2011;10(3):1361–73. pmid:21166384
            • 49. Breiman L. Random Forests. Mach Learn. 2001;45(1):5–32.
            • 50. Granholm V, Noble WS, Käll L. A cross-validation scheme for machine learning algorithms in shotgun proteomics. BMC Bioinf. 2012;13(Suppl 16):S3. S16-S3 pmid:23176259
            • 51. Matthews BW. Comparison of the predicted and observed secondary structure of T4 phage lysozyme. Biochim Biophys Acta. 1975;405(2):442–51. pmid:1180967
            • 52. Lavazza L, Morasca S. Comparing ϕ and the F-measure as performance metrics for software-related classifications. Empir Softw Eng. 2022;27(7):185.
            • 53. Bradley AP. The use of the area under the ROC curve in the evaluation of machine learning algorithms. Pattern Recognit. 1997;30(7):1145–59.
            • 54. Kanehisa M, Goto S. KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res. 2000;28(1):27–30. pmid:10592173
            • 55. Kanehisa M. Toward understanding the origin and evolution of cellular organisms. Protein Sci. 2019;28(11):1947–51. pmid:31441146
            • 56. Kanehisa M, Furumichi M, Sato Y, Kawashima M, Ishiguro-Watanabe M. KEGG for taxonomy-based analysis of pathways and genomes. Nucleic Acids Res. 2023;51(D1):D587–92. pmid:36300620
            • 57. Zhang Y, Chu J-M-T, Wong G-T-C. Cerebral glutamate regulation and receptor changes in perioperative neuroinflammation and cognitive dysfunction. Biomolecules. 2022;12(4):597. pmid:35454185
            • 58. Švob Štrac D, Pivac N, Mück-Šeler D. The serotonergic system and cognitive function. Transl Neurosci. 2016;7(1):35–49. pmid:28123820
            • 59. Bohnen NI, Grothe MJ, Ray NJ, Müller M, Teipel SJ. Recent advances in cholinergic imaging and cognitive decline-Revisiting the cholinergic hypothesis of dementia. Curr Geriatr Rep. 2018;7(1):1–11.
            • 60. Kuriyama N, Koyama T, Ozaki E, Saito S, Ihara M, Matsui D, et al. Association between cerebral microbleeds and circulating levels of mid-regional pro-adrenomedullin. J Alzheimers Dis. 2022;88(2):731–41. pmid:35694922
            • 61. Larrayoz IM, Ferrero H, Martisova E, Gil-Bea FJ, Ramírez MJ, Martínez A. Adrenomedullin contributes to age-related memory loss in mice and is elevated in aging human brains. Front Mol Neurosci. 2017;10:384. PMID: 29187812
            • 62. Šimundić AM. Measures of diagnostic accuracy: basic definitions. Ejifcc. 2009;19(4):203–11. pmid:27683318
            • 63. Power M, Fell G, Wright M. Principles for high-quality, high-value testing. Evid Based Med. 2013;18(1):5–10. 22740357. PMID
            • 64. Buawangpong N, Aramrat C, Pinyopornpanish K, Phrommintikul A, Soontornpun A, Jiraporncharoen W, et al. Risk prediction performance of the thai cardiovascular risk score for mild cognitive impairment in adults with metabolic risk factors in Thailand. Healthcare (Basel). 2022;10(10):1959.). pmid:36292406
            • 65. Metarugcheep S, Punyabukkana P, Wanvarie D, Hemrungrojn S, Chunharas C, Pratanwanich PN. Selecting the most important features for predicting mild cognitive impairment from Thai verbal fluency assessments. Sensors (Basel). 2022;22(15):5813.). pmid:35957370
            • 66. Tunvirachaisakul C, Supasitthumrong T, Tangwongchai S, Hemrunroj S, Chuchuen P, Tawankanjanachot I, et al. Characteristics of mild cognitive impairment using the Thai version of the consortium to establish a registry for Alzheimer’s disease tests: a multivariate and machine learning study. Dement Geriatr Cogn Disord. 2018;45(1-2):38–48. pmid:29617684
            • 67. Zhang S, Ge M, Cheng H, Chen S, Li Y, Wang K. Classification of cognitive ability of healthy older individuals using resting-state functional connectivity magnetic resonance imaging and an extreme learning machine. BMC Med Imaging. 2024;24(1):72. pmid:38532313
            • 68. Allwright S. What is a good F1 score and how do I interpret it?; 2022. Available from: https://stephenallwright.com/good-f1-score/.
            • 69. Banerjee P, Dehnbostel FO, Preissner R. Prediction is a balancing act: importance of sampling methods to balance sensitivity and specificity of predictive models based on imbalanced chemical data sets. Front Chem. 2018;6:362. pmid:30271769
            • 70. van den Goorbergh R, van Smeden M, Timmerman D, Van Calster B. The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression. J Am Med Inform Assoc. 2022;29(9):1525–34. 35686364. PMID
            • 71. Moghaddam DD, Rahmati O, Panahi M, Tiefenbacher J, Darabi H, Haghizadeh A, et al. The effect of sample size on different machine learning models for groundwater potential mapping in mountain bedrock aquifers. CATENA. 2020;187:104421.
            • 72. Rajput D, Wang W-J, Chen C-C. Evaluation of a decided sample size in machine learning applications. BMC Bioinf. 2023;24(1):48. pmid:36788550
            • 73. Amatya N, Garg AV, Gaffen SL. IL-17 Signaling: The Yin and the Yang. Trends Immunol. 2017;38(5):310–22. pmid:28254169
            • 74. Monin L, Gaffen SL. Interleukin 17 Family Cytokines: Signaling Mechanisms, Biological Activities, and Therapeutic Implications. Cold Spring Harb Perspect Biol. 2018;10(4):a028522. pmid:28620097
            • 75. Traub J, Frey A, Störk S. Chronic Neuroinflammation and Cognitive Decline in Patients with Cardiac Disease: Evidence, Relevance, and Therapeutic Implications. Life (Basel, Switzerland). 2023;13(2):329. pmid:36836686
            • 76. Lecca D, Jung YJ, Scerba MT, Hwang I, Kim YK, Kim S, et al. Role of chronic neuroinflammation in neuroplasticity and cognitive function: A hypothesis. Alzheimer's & dementia. 2022;18(11):2327–40. pmid:35234334
            • 77. Liu Y, Ouyang Y, You W, Liu W, Cheng Y, Mai X, et al. Physiological roles of human interleukin-17 family. Exp Dermatol. 2023;33(1):
            • 78. Brigas HC, Ribeiro M, Coelho JE, Gomes R, Gomez-Murcia V, Carvalho K, et al. IL-17 triggers the onset of cognitive and synaptic deficits in early stages of Alzheimer’s disease. Cell Reports. 2021;36(9):109574. pmid:34469732
            • 79. Di Filippo M, Mancini A, Bellingacci L, Gaetani L, Mazzocchetti P, Zelante T, et al. Interleukin-17 affects synaptic plasticity and cognition in an experimental model of multiple sclerosis. Cell Rep. 2021;37(10):110094. pmid:34879272
            • 80. Kim JJ, Bridle BW, Ghia J-E, Wang H, Syed SN, Manocha MM, et al. Targeted inhibition of serotonin type 7 (5-HT7) receptor function modulates immune responses and reduces the severity of intestinal inflammation. J Immunol. 2013;190(9):4795–804. 23554310. PMID
            • 81. Sahu A, Gopalakrishnan L, Gaur N, Chatterjee O, Mol P, Modi PK, et al. The 5-Hydroxytryptamine signaling map: an overview of serotonin-serotonin receptor mediated signaling network. J Cell Commun Signal. 2018;12(4):731–5. pmid:30043327
            • 82. Nikiforuk A. Targeting the serotonin 5-HT7 receptor in the search for treatments for CNS disorders: rationale and progress to date. CNS Drugs. 2015;29(4):265–75. pmid:25721336
            • 83. Fallarino F, Volpi C, Fazio F, Notartomaso S, Vacca C, Busceti C, et al. Metabotropic glutamate receptor-4 modulates adaptive immunity and restrains neuroinflammation. Nat Med. 2010;16(8):897–902. pmid:20657581
            • 84. Gerlai R, Roder JC, Hampson DR. Altered spatial learning and memory in mice lacking the mGluR4 subtype of metabotropic glutamate receptor. Behav Neurosci. 1998;112(3):525–32. pmid:9676970
            • 85. Martin J. Card sorting test performance: the role of visual working memory and the effect of visual feedback. University Of Tasmania; 2006.
              • 86. Traiffort E, O’Regan S, Ruat M. The choline transporter-like family SLC44: properties and roles in human diseases. Mol Aspects Med. 2013;34(2-3):646–54. pmid:23506897
              • 87. Hamid R, Sant HS, Kulkarni MN. Choline transporter regulates olfactory habituation via a neuronal triad of excitatory, inhibitory and mushroom body neurons. PLoS Genet. 2021;17(12):e1009938. pmid:34914708
              • 88. Young-Bernier M, Kamil Y, Tremblay F, Davidson PSR. Associations between a neurophysiological marker of central cholinergic activity and cognitive functions in young and older adults. Behav Brain Funct. 2012;8(1):17. pmid:22537877
              • 89. Taylor A, Grapentine S, Ichhpuniani J, Bakovic M. Choline transporter-like proteins 1 and 2 are newly identified plasma membrane and mitochondrial ethanolamine transporters. J Biol Chem. 2021;296:100604. pmid:33789160
              • 90. Andreadou I, Rekka EA, Kourounakis PN. Effect of novel anti-inflammatory ethanolamine derivatives with antioxidant properties on drug metabolising enzymes. Eur J Drug Metab Pharmacokinet. 2003;28(1):7–10. pmid:14503659
              • 91. Borodinsky LN, Belgacem YH, Swapna I, Sequerra EB. Dynamic regulation of neurotransmitter specification: relevance to nervous system homeostasis. Neuropharmacology. 2014;78:75–80. pmid:23270605
              • 92. Pendyam S, Mohan A, Kalivas PW, Nair SS. Role of perisynaptic parameters in neurotransmitter homeostasis--computational study of a general synapse. Synapse. 2012;66(7):608–21. pmid:22460547
              • 93. Ziegon L, Schlegel M. Netrin-1: a modulator of macrophage driven acute and chronic inflammation. Int J Mol Sci. 2021;23(1):275.). pmid:35008701
              • 94. Li FJ, Zheng SR, Wang DM. Adrenomedullin: an important participant in neurological diseases. Neural Regen Res. 2020;15(7):1199–207. pmid:31960799
              • 95. Ihara M, Washida K, Yoshimoto T, Saito S. Adrenomedullin: a vasoactive agent for sporadic and hereditary vascular cognitive impairment. Cereb Circ Cogn Behav. 2021;2:100007. pmid:36324729
              • 96. Hofbauer KH, Schoof E, Kurtz A, Sandner P. Inflammatory cytokines stimulate adrenomedullin expression through nitric oxide-dependent and -independent pathways. Hypertension. 2002;39(1):161–7. pmid:11799096
              • 97. Murman DL. The impact of age on cognition. Semin Hear. 2015;36(3):111–21. pmid:27516712
              • 98. Pinto JM, Fontaine AM, Neri AL. The influence of physical and mental health on life satisfaction is mediated by self-rated health: a study with Brazilian elderly. Arch Gerontol Geriatr. 2016;65:104–10. pmid:27017415
              • 99. Miranda AR, Franchetto Sierra J, Martinez Roulet A, Rivadero L, Serra SV, Soria EA. Age, education and gender effects on Wisconsin card sorting test: standardization, reliability and validity in healthy Argentinian adults. Neuropsychol Dev Cogn B Aging Neuropsychol Cogn. 2020;27(6):807–25. pmid:31744387