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

Abstract

Cognitive function is the term for the higher-order mental processes in the brain that gather and process information, and it mirrors brain activity. Cognitive function in adults exhibits variability as a result of genetic and environmental components such as gender, age[1], and lifestyle factors to name a few. Interindividual variability in cognitive trajectories has been observed in community-dwelling older adults across different cognitive domains. Inter-individual variations in cognitive response to identical physical exercise are also evident. This study aimed to explore the association between serum protein expression profiles and one measure of cognitive variability, as measured by the Wisconsin Card Sorting Test (WCST), in a healthy Thai population using a machine learning approach. This study included 199 healthy Thai subjects, ranging in age from 20 to 70 years. Cognitive performance was measured by the WCST, and the WCST % Errors was used to define the lower and higher cognitive ability groups. Serum protein expression profiles were studied by the label-free proteomics method. The Linear Model for Microarray Data (LIMMA) approach in R was utilized to assess differentially expressed proteins (DEPs) between groups; subsequently bioinformatic analysis was performed for the functional enrichment and interaction network analysis of DEPs. A random forest model was built to classify subjects from the lower and higher cognitive ability groups. Cross-validation was used for model performance evaluation. The results showed that, there were 213 DEPs identified between the poor and higher cognition groups, with 155 DEPs being upregulated in the poor cognition group. Those DEPs were significantly enriched in the IL-17 signaling pathway. Furthermore, the analysis of protein-protein interaction (PPI) network revealed that most of the selected DEPs were linked to neuroinflammation-related cognitive impairment. The random forest model achieved a test classification accuracy of 81.5%. The models sensitivity (true positive rate) was estimated to be 65%, and the specificity (true negative rate) was 85.9%. The AUC (0.79) indicates good binary classification performance. The results suggested that a measure of poor WCST performance in healthy Thai subjects might be attributed to higher levels of neuroinflammation.