Identification of Hot Spots in Protein Structures Using Gaussian Network Model and Gaussian Naive Bayes

Table 3

Performance comparison of the proposed models with the work by Ozbek et al. [12] and the simulated methods proposed by Haliloglu et al. [13] and Demirel et al. [14], where hm1– means that a total of high frequency modes are used together.

Reference	GNM modes	Cutoff	sw	sen	spe	pre	acc
Ozbek et al. [12]	hm1			0.14	0.89	0.05	0.86	0.0737
	hm2			0.16	0.80	0.05	0.85	0.0762
	hm3	6.5 Å	1	0.24	0.88	0.07	0.85	0.1084
	hm1–3			0.25	0.86	0.07	0.83	0.1094
	hm1–5			0.29	0.84	0.07	0.81	0.1128
Haliloglu et al. [13] (simulated)	hm1			0.1988	0.9019	0.0780	0.8738	0.1120
	hm1-2			0.2690	0.8819	0.0868	0.8574	0.1312
	hm1–3	7.0 Å	1	0.3041	0.8580	0.0820	0.8358	0.1292
	hm1–4			0.3275	0.8429	0.0800	0.8222	0.1286
	hm1–5			0.3450	0.8339	0.0797	0.8143	0.1295
Demirel et al. [14] (simulated)	hm1			0.0468	0.9773	0.0792	0.9400	0.0588
	hm1-2			0.0526	0.9697	0.0677	0.9330	0.0592
	hm1–3	7.0 Å	1	0.0409	0.9615	0.0424	0.9246	0.0417
	hm1–4			0.0819	0.9573	0.0741	0.9222	0.0778
	hm1–5			0.0936	0.9532	0.0769	0.9187	0.0844
This work	hm8	7.3 Å	3	0.1930	0.9436	0.1250	0.9136	0.1517
	hm8	7.1 Å	9	0.2515	0.9095	0.1039	0.8831	0.1470
	hm1–10	7.4 Å	1	0.2924	0.8992	0.1080	0.8749	0.1577
	hm1–11	7.4 Å	1	0.3041	0.8873	0.1012	0.8639	0.1518
	hm1–13	7.4 Å	1	0.3275	0.8736	0.0976	0.8518	0.1503
	hm1–20	7.2 Å	1	0.4269	0.8207	0.0903	0.8049	0.1491