Table 2. Summary of simulation filtering results.

Simulation Settings			Total # Features Removed			Avg # Network Features Removed
Signal	Total #	Data	Technique			Technique
Strength	Features	Type	SuMO-Fil	Low Mean	Low Variance	SuMO-Fil	Low Mean	Low Variance
Weak	45305	$\mathcal{X}$	3376	1091	904	0.34	0.76	0
Weak	45305	$\mathcal{G}$	2704	876	720	0.22	0.53	0.00
Moderate	45305	$\mathcal{X}$	3515	1315	165	0	0	15
Moderate	45305	$\mathcal{G}$	2684	1062	135	0	0.28	9.2
Strong	45305	$\mathcal{X}$	7339	1088	165	0	0	15
Strong	45305	$\mathcal{G}$	6364	878	140	0	0.11	10
Weak	25305	$\mathcal{X}$	2089	671	547	0.34	0.78	0
Weak	25305	$\mathcal{G}$	1389	448	375	0.21	0.46	0
Moderate	25305	$\mathcal{X}$	2245	820	165	0.01	0	15
Moderate	25305	$\mathcal{G}$	1507	558	130	0.01	0.34	9.1
Strong	25305	$\mathcal{X}$	5791	659	165	0	0	15
Strong	25305	$\mathcal{G}$	3729	441	140	0	0.10	10
Weak	15305	$\mathcal{X}$	841	232	196	0.40	0.81	0.01
Weak	15305	$\mathcal{G}$	835	228	192	0.15	0.45	0
Moderate	15305	$\mathcal{X}$	1278	280	162	0.02	0	14.7
Moderate	15305	$\mathcal{G}$	1263	274	124	0.01	0.33	8.8
Strong	15305	$\mathcal{X}$	2242	229	161	0	0	14.7
Strong	15305	$\mathcal{G}$	2132	229	138	0	0.11	9.9

The total number of features removed and the number network features erroneously removed are averaged across 500 simulations under each setting accounting for various number of total features and network signal strengths. The simulations contained 15 network features within data type $\mathcal{X}$ and 10 network features within data type $\mathcal{G}$. An ideal filtering technique would eliminate no network features to maintain signal for the primary analysis. Additionally, an ideal filtering technique would remove enough features to make an impact on run times for the primary analysis while maintaining sparsity assumptions. The results indicate that the variance filtering technique performs favorably for weak network signals, but performs poorly for the moderate and strong signal strengths in regards to both the number of features removed and the number of network features erroneously removed. The SuMO-Fil performs favorably compared to both the mean and variance filtering techniques under most simulation settings.