Table 1.
Summary of six studies for predicting AD using blood gene expression data.
| Study | Data source/# of AD and CN (Training and test datasets) | Feature selection methods (Data used for feature selection) | Classifying method | Number of selected features | Performance |
|---|---|---|---|---|---|
| Booij et al.15 |
Not publicly available (Norway)/126 AD and 126 CN (Randomly dividing all data into training and test datasets by 3:1 ratio) |
Jack-knife (training data) | PLSR | 1239 genes | ACC: 0.87 AUC: 0.94 |
| Lunnon et al.16 | ANM/104 AD and 104 CN (Randomly dividing AD and CN data into training and test datasets by 3:1 ratio) |
t-test RF with Meng score and backward elimination (training data) |
RF | 50 probes | ACC: 0.75 |
| Sood et al.17 | ANM1 and ANM2/49 AD and 64 CN, 40 AD and 71 CN (LOOCV) | Bayesian statistic (ULSAM Ageing data GEO:GSE60862) | kNN | 150 probes | AUC: 0.73 (ANM1) AUC: 0.66 (ANM2) |
| Voyle et al.18 | ANM1 and ANM+DCR/100 AD and 107 CN, 118 AD and 118 CN (ANM1 for training, ANM2 + DCR for test) | REF and pickSizeTolerance (Training data) | RF | 13 probes (12 genes) |
ACC: 0.657 AUC: 0.724 |
| Li et al.19 | ANM1 and ANM2/145 AD and 104 CN, 140 AD and 135 CN (ANM1 for training, ANM2 for test and vice versa) | Ref-REO (Training data) | Not described | 1,145 gene pairs (ANM1: training data) 1,249 gene pairs (ANM2: training data) | AUC: 0.733 (ANM2: test set) AUC: 0.775 (ANM1: test set) |
| Li et al.20 | ANM1 and ANM2/143 AD and 104 CN, 102 AD and 78 CN (ANM1 for training, ANM2 for test and vice versa) | LASSO regression (ANM1 and ANM2) | Majority voting of SVM, RR and RF | 6 genes (Full6set) | AUC: 0.866 (ANM2: test set) AUC: 0.864 (ANM1: test set) |
AD: Alzheimer’s Disease; CN: healthy control; PLSR: partial least square regression; ACC: accuracy; AUC: area under the curve; ANM: AddNueroMed; RF: Random Forest; kNN: k-nearest neighbors; RFE: recursive feature elimination; pickSizeTolerance: a function in caret package29; ULSAM: the Uppsala Longitudinal Study of Adult Men; LOOCV: leave-one-out cross-validation; LASSO: least absolute shrinkage and selection operator; SVM: support vector machine; RF: random forest; RR: logistic ridge regression.