Predicting sumoylation sites using support vector machines based on various sequence features, conformational flexibility and disorder

Table 2 Top 25 features selected using RELIEFF algorithm.

Rank	Feature	Merit Score	P-value	Adj. P-value	Significance
1	w+E_2	0.355474	0.00E+00	0.00E+00	*
2	Consensus	0.261813	0.00E+00	0.00E+00	*
3	wDE	0.164459	1.66E-41	3.23E-40	*
4	w+2_Hydro	0.160149	3.90E-83	1.33E-81	*
5	w-I_3	0.105916	1.18E-107	5.33E-106	*
6	w-3_Hydro	0.104835	8.12E-58	1.84E-56	*
7	wK	0.078651	1.33E-02	4.03E-02	*
8	w-V_3	0.075073	1.92E-58	5.22E-57	*
9	w-2_Hydro	0.057669	1.48E-02	4.37E-02	*
10	w+3_Hydro	0.056496	1.49E-01	2.93E-01
11	w+1_Hydro	0.05232	7.13E-02	1.62E-01
12	w-1_Hydro	0.051279	4.22E-02	9.89E-02
13	w-L_3	0.051001	1.96E-03	7.00E-03	*
14	w+K_2	0.050248	1.70E-08	1.78E-07	*
15	w+P_2	0.045911	3.39E-02	8.23E-02
16	w+P_3	0.043573	2.11E-25	3.58E-24	*
17	w-K_3	0.043208	4.33E-04	1.96E-03	*
18	Flexibility	0.042334	7.52E-07	7.31E-06	*
19	w+D_2	0.041784	7.14E-01	7.77E-01
20	w-S_2	0.041097	2.95E-01	4.57E-01
21	DisorderBinary	0.040666	7.27E-14	9.88E-13	*
22	w-E_3	0.039548	6.14E-05	3.79E-04	*
23	w-A_3	0.03804	3.04E-02	7.95E-02
24	w-P_1	0.037893	4.46E-06	3.80E-05	*
25	w-E_2	0.035935	4.16E-01	5.74E-01

93 out of 137 features has been selected based on 10-fold classification performance. Features are ranked using RELIEFF [26] algorithm, implemented in Weka [39]. Details of statistical testing can be found in Methods section. For assessing significance, adjusted p-value cutoff of 0.05 is used. Feature name explanations can be found in Methods section and Figure 2a.

ISSN: 1471-2164