Skip to main content
. 2020 Oct 27;11:5420. doi: 10.1038/s41467-020-19176-z

Table 1.

Summary of prediction models created using support vector machine and random forest.

Machine learning methodology Samples Model Number of variables % Accuracy score Model validation Model evaluation
TPR TNR TP FP TN FN TPR TNR TP FP TN FN AUC
randomForest (RF) All samples (1st cohort) Four metabolomes 54 73 0.87 0.55 87 45 55 13 0.58 0.64 58 36 64 42 0.72
randomForest (RF) All samples (1st cohort) DHA metabolome 23 81 0.9 0.68 90 32 68 10 0.39 0.55 39 45 55 61 0.44
randomForest (RF) All samples (1st cohort) n-3 DPA metabolome 10 69 0.83 0.5 83 50 50 17 0.78 0.32 78 68 32 22 0.58
randomForest (RF) All samples (1st cohort) EPA metabolome 3 65 0.8 0.45 80 55 45 20 0.53 0.32 53 68 32 47 0.6
randomForest (RF) All samples (1st cohort) AA metabolome 18 65 0.77 0.5 77 50 50 23 0.83 0.77 83 23 77 17 0.89
randomForest (RF) All samples (1st cohort) Clin. Score 11 50 0.6 0.36 60 64 36 40 0.03 0.88 3 12 88 97 0.53
randomForest (RF) All samples (2nd cohort) RvD4, 10S, 17S-diHDPA, 15R-LXA4, MaR1n-3 DPA 4 83 0.92 0.68 92 32 68 8 0.83 0.59 83 41 59 17 0.8
randomForest (RF) All samples (2nd cohort) RvD4, 10S, 17S-diHDPA, 15R-LXA4, 5S,12S-diHETE, 4S,14S-diHDHA, MaR1n-3 DPA 6 86 0.94 0.73 94 27 73 6 0.87 0.64 87 36 64 13 0.79
randomForest (RF) Fibroid samples (1st & 2nd cohort) RvD4, 10S, 17S-diHDPA, 15R-LXA4, MaR1n-3 DPA 4 88 0.89 0.87 89 13 87 11 N/A N/A N/A N/A N/A N/A N/A
randomForest (RF) Lymphoid samples (1st & 2nd cohort) RvD4, 10S, 17S-diHDPA, 15R-LXA4, MaR1n-3 DPA 4 83 0.89 0.7 89 30 70 11 N/A N/A N/A N/A N/A N/A N/A
randomForest (RF) Myeloid samples (1st & 2nd cohort) RvD4, 10S, 17S-diHDPA, 15R-LXA4, MaR1n-3 DPA 4 70 0.86 0.47 86 53 47 14 N/A N/A N/A N/A N/A N/A N/A
Classyfire (SVM) All samples (1st cohort) Four metabolomes 54 61 0.63 0.54 63 46 54 37 0.94 0.05 94 95 5 6 0.53
Classyfire (SVM) All samples (1st cohort) DHA metabolome 23 62 0.65 0.56 65 44 56 35 0.78 0.18 78 82 18 22 0.54
Classyfire (SVM) All samples (1st cohort) n-3 DPA metabolome 10 61 0.63 0.54 63 46 54 37 0.94 0.23 94 77 23 6 0.66
Classyfire (SVM) All samples (1st cohort) EPA metabolome 3 60 0.62 0.52 62 48 52 38 0.92 0.05 92 95 5 8 0.58
Classyfire (SVM) All samples (1st cohort) AA metabolome 18 58 0.6 0.48 60 52 48 40 0.92 0.09 92 91 9 8 0.66

TPR sensitivity, TNR specificity, TP true positives, FP false positives, TN true negatives, FN false negatives, AUC area under the curve.