HAI Book 2025 - Flipbook - Page 304
Kuchenbecker, Lindsey
60
Data-driven biomarker discovery nominates novel diagnostic plasma
biomarkers and provides insights into Alzheimer9s disease pathophysiology
in an African American cohort
Lindsey Kuchenbecker1, Kevin Thompson2, Cheyenne Hurst1, Bianca Opdenbosch1, Michael
Heckman3, Joseph Reddy3, Thuy Nguyen1, heidi Casellas1, Katie Sotelo1, Delila Redy1, John
Lucas4, Gregory Day5, Floyd Willis6, Neill Graff-Radford5, Niluer Ertekin-Taner1,5, Krishna
Kalari2, Minerva Carrasquillo1
1
Department of Neuroscience, Mayo Clinic, Jacksonville, FL, USA, Jacksonville, FL, US
Department of Quantitative Health Sciences, Mayo Clinic, Rochester, MN, USA, Rochester, MN, US
3
Department of Quantitative Health Sciences, Mayo Clinic, Jacksonville, FL, USA, Jacksonville, FL, US
4
Department of Psychiatry and Psychology, Mayo Clinic, Jacksonville, FL, USA, Jacksonville, FL, US
5
Department of Neurology, Mayo Clinic, Jacksonville, FL, USA, Jacksonville, FL, US
6
Department of Family Medicine, Mayo Clinic, Jacksonville, FL, USA, Jacksonville, FL, US
2
Background: Blood-based biomarker development is of increasing interest due to improved accessibility
compared to established diagnostic modalities for Alzheimer9s disease (AD). However, current plasma biomarker
candidates have not been tested comprehensively in diverse populations, and represent only a fraction of
pathways dysregulated in AD. This study seeks to address these knowledge gaps through untargeted plasma
biomarker discovery in African Americans (AA), an underrepresented group in AD research despite having a higher
risk of developing AD compared to non-Hispanic Whites.
Methods Untargeted proteomics were obtained using the SomaScan 7k assay on plasma samples from AA study
participants diagnosed as AD dementia (n=183) or cognitively unimpaired (CU, n=145). Machine learning
approaches were implemented to identify an optimal protein panel to discriminate between AD dementia and CU.
Predictive models were built using elastic net regression optimized with nested cross-validation on 7,298 proteins
and evaluated using the area under the receiver operating curve (AUC). The generalizability of plasma protein
predictors was evaluated in non-overlapping plasma and brain proteome datasets.
Results: Thirty-three proteins were identified as predictors. Despite the untargeted selection of predictors, many
of these proteins have been linked to core AD pathology and disease-associated pathways, including
neuroinflammation, synaptic integrity, and the matrisome. These 33 plasma proteins were able to classify AD
dementia versus CU with an AUC of 0.91, a substantial increase compared to a model including age and sex
(AUC=0.67). Sixteen of the 33 protein predictors were present in the ANMerge plasma proteomics dataset, and
14/33 predictors were present in the AMP-AD Diverse Cohorts brain proteomics dataset, each achieving AUCs of
0.83 and 0.94 when combined with age, sex, and APOE, respectively.
Conclusion: Data-driven biomarker discovery leveraging untargeted proteomics and machine learning identified
plasma proteins that could serve as a novel and highly accurate diagnostic plasma biomarker panel for AD in AA.
Keywords: blood-based biomarkers, diagnosis, health disparities, proteomics, machine learning
HAI2025 - 304
