Affiliations: [a] Institute for Translational Research, Department of Pharmacology & Neuroscience, University of North Texas Health Science Center, Fort Worth, TX, USA
| [b] Department of Family Medicine, University of North Texas Health Science Center, Fort Worth, TX, USA
Correspondence:
[*]
Correspondence to: Sid O’Bryant, PhD, University of North Texas Health Science Center, Fort Worth, TX, USA. Tel.: +1 817 735 2962; E-mail: sid.obryant@unthsc.edu.
Abstract: Background:There is a need for more reliable diagnostic tools for the early detection of Alzheimer’s disease (AD). This can be a challenge due to a number of factors and logistics making machine learning a viable option. Objective:In this paper, we present on a Support Vector Machine Leave-One-Out Recursive Feature Elimination and Cross Validation (SVM-RFE-LOO) algorithm for use in the early detection of AD and show how the SVM-RFE-LOO method can be used for both classification and prediction of AD. Methods:Data were analyzed on n = 300 participants (n = 150 AD; n = 150 cognitively normal controls). Serum samples were assayed via a multi-plex biomarker assay platform using electrochemiluminescence (ECL). Results:The SVM-RFE-LOO method reduced the number of features in the model from 21 to 16 biomarkers and achieved an area under the curve (AUC) of 0.980 with a sensitivity of 94.0% and a specificity of 93.3%. When the classification and prediction performance of SVM-RFE-LOO was compared to that of SVM and SVM-RFE, we found similar performance across the models; however, the SVM-RFE-LOO method utilized fewer markers. Conclusion:We found that 1) the SVM-RFE-LOO is suitable for analyzing noisy high-throughput proteomic data, 2) it outperforms SVM-RFE in the robustness to noise and in the ability to recover informative features, and 3) it can improve the prediction performance. Our recursive feature elimination model can serve as a general model for biomarker discovery in other diseases.
