Table 1.

Average classification performance of the quality features in each subset group.

Quality feature	Group A	Group B	Group C
RAW_Basic_Statistics	0.500	0.500	0.500
RAW_Per_base_sequence_quality	0.681	0.738	0.747
RAW_Per_tile_sequence_quality	0.645	0.668	0.658
RAW_Per_sequence_quality_scores	0.693	0.738	0.754
RAW_Per_base_sequence_content	0.650	0.622	0.632
RAW_Per_sequence_GC_content	0.684	0.686	0.652
RAW_Per_base_N_content	0.674	0.732	0.745
RAW_Sequence_Length_Distribution	0.502	0.523	0.511
RAW_Sequence_Duplication_Levels	0.606	0.586	0.602
RAW_Overrepresented_sequences	0.771	0.763	0.740
RAW_Adapter_Content	0.551	0.526	0.538
RAW_Kmer_Content	0.559	0.574	0.549
MAP_SE_no_mapping	0.840	0.817	0.841
MAP_SE_uniquely	0.829	0.821	0.850
MAP_SE_multiple	0.860	0.805	0.859
MAP_SE_overall	0.840	0.817	0.841
MAP_MI_no_mapping	0.837	0.809	0.825
MAP_MI_uniquely	0.839	0.817	0.841
MAP_MI_multiple	0.847	0.798	0.841
MAP_MI_overall	0.846	0.810	0.827
LOC_Promoter	0.719	0.717	0.717
LOC_5_UTR	0.706	0.692	0.687
LOC_3_UTR	0.745	0.711	0.725
LOC_1st_Exon	0.699	0.688	0.702
LOC_Other_Exon	0.733	0.724	0.733
LOC_1st_Intron	0.687	0.716	0.705
LOC_Other_Intron	0.686	0.717	0.697
LOC_Downstream	0.681	0.688	0.697
LOC_Distal_Intergenic	0.727	0.719	0.710
TSS_−4500	0.690	0.692	0.676
TSS_−3500	0.696	0.707	0.710
TSS_−2500	0.682	0.694	0.699
TSS_−1500	0.702	0.685	0.708
TSS_−500	0.703	0.731	0.723
TSS_+500	0.706	0.718	0.718
TSS_+1500	0.686	0.702	0.724
TSS_+2500	0.691	0.709	0.722
TSS_+3500	0.677	0.703	0.712
TSS_+4500	0.691	0.692	0.701
MAP_PE_con_no_mapping	0.833	0.711	0.711
MAP_PE_con_uniquely	0.852	0.772	0.772
MAP_PE_con_multiple	0.830	0.708	0.708
MAP_PE_dis_uniquely	0.771	0.676	0.676
MAP_PE_cod_no_mapping	0.837	0.626	0.626
MAP_PE_cod_uniquely	0.773	0.673	0.673
MAP_PE_cod_multiple	0.849	0.655	0.655
MAP_PE_overall	0.853	0.724	0.724

Classification performance is measured as area under Receiver Operating Characteristic curve (auROC) from 0.5 for a random classification and 1.0 for a perfect classification. MAP features perform best over all three groups. The RAW features that perform well do so over all three groups. For LOC and TSS only some of the features show good performance on average; however, these still can be more important for some of the subsets in each group.