Table 2.
Simulation results for covariate selection.
| Variable selection accuracy | ||||||||
|---|---|---|---|---|---|---|---|---|
| TPR | FPR | F1 | MCC | TPR | FPR | F1 | MCC | |
| Random | Hub | |||||||
| mSSL-DPE | 0.840 | 0.000 | 0.912 | 0.898 | 0.868 | 0.000 | 0.929 | 0.915 |
| mLDM | 0.609 | 0.003 | 0.751 | 0.734 | 0.612 | 0.003 | 0.754 | 0.738 |
| SINC (Ω = I) | 0.808 | 0.003 | 0.871 | 0.889 | 0.813 | 0.001 | 0.887 | 0.894 |
| SINC (τ = 1) | 0.914 | 0.001 | 0.943 | 0.953 | 0.925 | 0.000 | 0.952 | 0.960 |
| SINC (τ learned) | 0.917 | 0.000 | 0.947 | 0.956 | 0.926 | 0.001 | 0.952 | 0.960 |
| Cluster | Band | |||||||
| mSSL-DPE | 0.837 | 0.000 | 0.910 | 0.900 | 0.853 | 0.000 | 0.920 | 0.906 |
| mLDM | 0.605 | 0.003 | 0.747 | 0.730 | 0.597 | 0.003 | 0.741 | 0.725 |
| SINC (Ω = I) | 0.791 | 0.006 | 0.854 | 0.873 | 0.808 | 0.000 | 0.887 | 0.893 |
| SINC (τ = 1) | 0.908 | 0.000 | 0.941 | 0.951 | 0.921 | 0.001 | 0.947 | 0.957 |
| SINC (τ learned) | 0.910 | 0.000 | 0.942 | 0.952 | 0.922 | 0.000 | 0.950 | 0.959 |
NOTE: mSSL-DPE refers to the method of Deshpande, Ročková, and George (2019), mLDM to the method of Yang, Chen, and Chen (2017), SINC (Ω = I) to the modified version of the proposed model with the precision matrix fixed, and SINC to the proposed model. SpiecEasi is omitted from the comparison, as it does not perform selection or adjustment for covariates. Random, Hub, Cluster, and Band refer to the underlying shape of the network, as illustrated in Figure 1. Bold values reflect top performing methods.