Table 1.

Performance comparison of random forest (RF)-based ensemble method with RF-based other classifiers on benchmarking dataset.

Features	Matthews’ correlation coefficient (MCC)	Accuracy	Sensitivity	Specificity	AUC	P-value
Amino acid composition (AAC)	0.394 ± 0.022	0.697 ± 0.011	0.703 ± 0.013	0.691 ± 0.016	0.769 ± 0.009	0.288
Dipeptide composition (DPC)	0.420 ± 0.023	0.709 ± 0.012	0.746 ± 0.017	0.673 ± 0.014	0.780 ± 0.011	0.651
Composition–transition–distribution (CTD)	0.330 ± 0.032	0.665 ± 0.016	0.662 ± 0.033	0.668 ± 0.034	0.729 ± 0.019	0.001
Physicochemical properties (PCP)	0.320 ± 0.017	0.659 ± 0.008	0.696 ± 0.020	0.622 ± 0.017	0.725 ± 0.013	0.0006
Amino acid index (AAI)	0.397 ± 0.018	0.698 ± 0.009	0.698 ± 0.026	0.699 ± 0.019	0.772 ± 0.010	0.370
Hybrid	0.381 ± 0.022	0.690 ± 0.011	0.721 ± 0.036	0.658 ± 0.033	0.762 ± 0.014	0.149
PIP-EL	0.435 ± 0.019	0.717 ± 0.010	0.707 ± 0.010	0.727 ± 0.015	0.788 ± 0.011	–

The first column corresponds to the performance of individual feature group, hybrid feature, and ensemble learning. The column 2–6 respectively represent the MCC, accuracy, sensitivity, specificity, and AUC value, where each value shown as the average ± SD of 10 alternative balanced datasets. The last column represents a pairwise comparison of AUC between PIP-EL and the other methods using a two-tailed t-test. P ≤ 0.05 indicates a statistically meaningful difference between PIP-EL and the selected composition (shown in bold).