MiRTif: a support vector machine-based microRNA target interaction filter

BMC Bioinformatics

Table 2 The top 25 most informative features.

Feature	μ ⁺	σ ⁺	M ^-	σ ^-	F
3-gram, non-seed, mismatch/AU/AU	0.0197	0.0370	0.0000	0.0000	0.5313
2-gram, non-seed, mismatch/AU	0.0526	0.0606	0.0107	0.0353	0.4374
2-gram, entire, mismatch/AU	0.0441	0.0439	0.0160	0.0265	0.3991
3-gram, entire, GC/gap/gap	0.0068	0.0176	0.0228	0.0236	0.3904
3-gram, entire, mismatch/mismatch/gap	0.0060	0.0164	0.0000	0.0000	0.3636
3-gram, non-seed, gap/GU/AU	0.0095	0.0262	0.0000	0.0000	0.3631
3-gram, entire, gap/GU/AU	0.0062	0.0172	0.0000	0.0000	0.3629
3-gram, entire, mismatch/AU/AU	0.0198	0.0312	0.0044	0.0132	0.3457
3-gram, non-seed, mismatch/mismatch/AU	0.0212	0.0422	0.0022	0.0135	0.3417
2-gram, seed, GU/GC	0.0117	0.0352	0.0000	0.0000	0.3337
3-gram, entire, AU/mismatch/GU	0.0054	0.0167	0.0000	0.0000	0.3253
2-gram, entire, gap/gap	0.0838	0.1059	0.1512	0.1030	0.3226
1-gram, entire, GC	0.2399	0.0957	0.2893	0.0678	0.3021
1-gram, non-seed, gap	0.1880	0.1581	0.2841	0.1601	0.3020
2-gram, non-seed, gap/gap	0.1022	0.1406	0.1886	0.1505	0.2969
3-gram, non-seed, GC/mismatch/AU	0.0067	0.0224	0.0000	0.0000	0.2969
1-gram, entire, gap	0.1595	0.1225	0.2273	0.1066	0.2958
3-gram, non-seed, mismatch/mismatch/gap	0.0067	0.0227	0.0000	0.0000	0.2943
3-gram, entire, mismatch/mismatch/AU	0.0135	0.0259	0.0026	0.0111	0.2937
2-gram, entire, GC/gap	0.0199	0.0298	0.0357	0.0243	0.2932
2-gram, entire, GC/GC	0.0630	0.0549	0.0928	0.0471	0.2930
3-gram, entire, GU/GC/gap	0.0002	0.0028	0.0043	0.0115	0.2895
1-gram, non-seed, GC	0.1742	0.1261	0.2377	0.0952	0.2870
3-gram, entire, gap/GU/GC	0.0005	0.0047	0.0056	0.0136	0.2810
3-gram, non-seed, GU/mismatch/AU	0.0064	0.0233	0.0000	0.0000	0.2756

Features are in the format of k- gram type, region, and k- gram code. For example, "3-gram, non-seed, mismatch/AU/AU" represent a mismatch followed by an AU pair followed by an AU pair in the non-seed region (see Materials and Method – Data representation for the detailed definitions of k- gram, region and k-gram code). For each feature, its means and standard deviations in both positive and negative sets are listed. The F score is defined as |(μ⁺ - μ^-)/(σ⁺ + σ^-)|, which measures the discriminating ability of the feature.

ISSN: 1471-2105