Table 3.
The machine learning approaches followed by each study. Acc: Accuracy, SVM: Support Vector Machines, LR: Logistic Regression, RF: Random Forest, DNN: Deep Neural Network, SVC: Support Vector Classifier (SVC), CNN: Convolutional Neural Network, AUC: Area under the receiver operator characteristic curve, CSM: Combinatorial Searching Method, Chr: Chromosome.
| Study | Machine Learning Models | Model Assessment | Performance |
|---|---|---|---|
| Vitsios et al. [103] | Stochastic Semi-supervised Learning: Stacking, DNN, Gradient Boosting, RF, SVC, XGBoost, ExtraTrees Classifiers | 10-fold cross validation | Stacking: Avg AUC: 0.767, DNN: Avg AUC: 0.774, Gradient Boosting: Avg AUC: 0.79, RF: Avg AUC: 0.798, SVC: Avg AUC: 0.801, XGBoost: Avg AUC: 0.805, ExtraTrees: Avg AUC: 0.814 |
| Yousefian et al. [104] | Convolutional Neural Network | Autoencoder pre-training, Chr 1-10: training set, Chr 11–14: testing set and Chr 15–22: validation set | CNN: AUC: 0.96 F1-score: 0.83 |
| Bean et al. [105] | Knowledge graph edge prediction model [116] | 5-fold cross validation | Fold-Change Enrichment and random guess baseline: ALSoD: 23.33 (23.53), ClinVar: 30.05 (15.64), DisGeNet: 55.90 (81.66), Manual: 84.54 (13.27), Union: 8.92 (4.28) |
| Yin et al. [90] | Deep Neural Network (ALS-Net), Logistic Regression, SVM, Random Forest and Adaboost | 9-fold cross validation | ALS-Net: Acc: 0.769 F1-score: 0.797 LR: Acc: 0.739 F1-score:0.728 SVM: Acc: 0.725 F1-score:0.694 RM: Acc: 0.596 F1-score:0.381 Adaboost: Acc: 0.661 F1-score:0.625 (+PromoterCNN and all 4 chromosomes combined) |
| Kim et al. [91] | Multifactor dimensionality reduction; using a naïve Bayes classifier |
1000 permutation tests | Critical Acc: 0.629 and 0.640 (replication dataset) |
| Greene et al. [107] | Multifactor dimensionality reduction | 1000 permutation tests | Best SNP pairwise model: Acc: 0.6551 and 0.5821 (replication dataset); with p < 0.048 and p < 0.021 (replication dataset) |
| Sha et al. [108] | Two-locus probabilistic models, Multifactor dimensionality reduction, Combinatorial Searching Method | 1000 permutation tests | Two-locus models: rs4363506-rs3733242: p = 0.032 rs4363506-rs16984239: p = 0.042 MDR model: rs4363506-rs12680546: p = 0.156 CSM model: rs4363506-rs12680546: p = 0.2 |