Sex‐adjusted blood biomarkers differentiate AD from LATE

Abstract Background Alzheimer’s disease (AD) is a complex neurodegenerative disorder that affects cognition, memory, and behavior, while limbic‐predominant age‐related TDP‐43 encephalopathy (LATE) is a recently defined common neurodegenerative disease that mimics the clinical symptoms of AD (Table 1). At present, there are no effective diagnostic biomarkers for LATE which can mainly be confirmed by autopsy. Method We applied machine learning algorithms to identify informative blood biomarkers differentiating AD from LATE. To identify blood‐based biomarkers for discriminating AD from LATE based on imbalanced data, we proposed an innovative integrated feature selection‐based approach, Preprocessing, Environmental factor analysis, Feature selection, and Validation (PEFV). Result We focused on the association analysis of blood markers that differentiate LATE and/or AD, on ROSMAP (Figure 1), by considering the factors including medication and diet. We pre‐identified some sensitive medication and dietary factors, and then based on these factors, we performed sex‐stratified group analyses. We identified 3 small sets of blood biomarkers that distinguish AD from LATE adjusted for each case, including All, Male, and Female. In addition, we analyzed 3 scenarios for each case: pure LATE vs. LATE+AD, pure AD vs. AD+LATE, and pure AD vs. pure LATE (Tables 2‐4). While usually, a single biomarker cannot achieve high accuracy and reliability due to the complexity of both AD and LATE, the set of biomarkers we identified for each scenario was small. For example, in Male case, 8 biomarkers were needed to distinguish LATE+AD from LATE with an accuracy rate of 61%, an improvement of about 10% over using all blood test features; 3 biomarkers were needed to distinguish LATE+AD from AD with an accuracy rate of 77%, an improvement of about 23% over using all blood test features; 4 biomarkers were needed to distinguish AD from LATE with an accuracy rate of 75%, an improvement of about 22% over using all blood test features. Conclusion In this study, based on blood tests, we explored blood‐based biomarkers that can differentiate AD from LATE with imbalanced data by our proposed approach PEFV and used it to discover informative blood biomarkers distinguishing AD from LATE.