CSF complement 3 and factor H are staging biomarkers in Alzheimer’s disease

Table 2 XMITTN output for ADNI and Emory cohorts assessing whether classification using two sets of variables is better than chance

Machine learning algorithm	p-value, Experiment 1, ADNI cohort	p-value, Experiment 2, ADNI cohort	p-value, Experiment 1, Emory cohort	p-value, Experiment 2, Emory cohort
Logistic	0.510	0.140
Perceptron	0.792	0.912
Decision Tree	0.197	0.161
Random Forests	0.128	0.043	0.560	0.595
Naïve Bayes	0.403	0.367
K-Nearest Neighbor	0.106	0.069
Boosted Decision Tree	0.399	0.245
Gradient Boosting	0.185	0.104
Support Vector Machine	0.125	0.033	0.266	0.014

Experiment 1 includes 6 features: age, gender, presence of at least one APOE ε4 allele, CSF Aβ42, CSF t-Tau, and CSF p-Tau₁₈₁, and no ML algorithm performed better than chance in distinguishing between the two AD stages. Experiment 2 has all previous features plus C3 and FH and levels, and achieved improved classification in two algorithms in the ADNI cohort and support vector machine in the Emory cohort (p < 0.05 shown in bold)

ISSN: 2051-5960