| Autor: |
Cisterna‐García, Alejandro, Chen, Hsiang‐Han, Norton, Joanne, Gentsch, Jen, Bergmann, Kristy, Wang, Fengxian, Budde, John P., Cruchaga, Carlos, Botía, Juan A., Ibanez, Laura |
| Zdroj: |
Alzheimer's & Dementia: The Journal of the Alzheimer's Association; Jun2023 Supplement 1, Vol. 19, p1-5, 5p |
| Abstrakt: |
Background: Cerebrospinal fluid (CSF) Alzheimer's disease (AD) biomarkers are used to clinically diagnose AD with an accuracy of 70‐80%. However, they are invasive and expensive and need to be further studied on preclinical‐AD. Additionally, a blood‐based biomarker can be of great use. In order to obtain a blood‐based biomarker, we have sequenced plasma cell‐free RNA (cfRNA) from preclinical‐AD individuals to develop prediction models for AD using machine learning (Figure 1). Method: Plasma samples were obtained from the Knight‐Alzheimer‐Disease‐Research‐Center [preclinical‐AD (n = 67), clinical‐AD (n = 92), and controls (n = 48)] (Table 1) and sequenced in two batches. Sequences were processed following standard pipelines and normalized using DESeq2. Preclinical‐AD (n = 47) and controls (n = 26) from one sequencing batch were used to train the predictive model, then preclinical‐AD (n = 20) and controls (n = 22) from the other batch were used as testing population. We applied a Z‐score scaling to the normalized gene counts and used Kullback–Leibler (KL) divergence to select the genes that had similar expression distribution in both sequencing experiments (Figure 2). We used different KL thresholds to select genes as predictors and then Ridge regression to find the best predictive models. After we tested the models on preclinical‐AD, they were further tested on the clinical‐AD individuals. Finally, Parkinson's disease (n = 96), dementia with Lewy bodies (n = 17), and frontotemporal dementia (n = 16) individuals were used to test for specificity analyses of the AD prediction models (Table 2). Result: Best predictors sizes include 40, 90, and 220 genes yielding models with 85.7%, 90.5%, and 95.2% testing accuracy respectively. Predicted AD risk consistently correlates with Aβ42 levels. The best model included genes such as MT‐ATP6, GAB2, or MAPK14, which have been associated with AD. We tested the specificity of these models for other neurodegenerative diseases, and we observed specificity for AD in comparison to PD, FTD, and DLB (Figure 3). Conclusion: Plasma cfRNA is a promising tool to screen for preclinical AD. It has the potential to be used as a minimally invasive biomarker for AD since it has an accuracy higher or comparable to CSF biomarkers and it is AD‐specific. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Supplemental Index |
| Externí odkaz: |
|
