Prediction of RNA-protein sequence and structure binding preferences using deep convolutional and recurrent neural networks

Table 2 The AUC performance comparison between iDeepS and other methods on 31 experiments

Protein	iDeepS	DeepBind	DeeperBind	Oli	GraphProt
1 Ago/EIF	0.773	0.713	0.740	0.610	0.691
2 Ago2-MNase	0.591	0.595	0.606	0.512	0.595
3 Ago2-1	0.865	0.849	0.857	0.803	0.817
4 Ago2-2	0.868	0.830	0.868	0.800	0.823
5 Ago2	0.634	0.628	0.630	0.534	0.633
6 eIF4AIII-1	0.950	0.938	0.950	0.919	0.918
7 eIF4AIII-2	0.953	0.950	0.954	0.929	0.931
8 ELAVL1-1	0.932	0.924	0.930	0.889	0.915
9 ELAVL1-MNase	0.600	0.613	0.614	0.491	0.591
10 ELAVL1A	0.893	0.886	0.893	0.843	0.867
11 ELAVL1-2	0.919	0.914	0.919	0.875	0.895
12 ESWR1	0.917	0.912	0.915	0.808	0.840
13 FUS	0.934	0.942	0.939	0.846	0.860
14 Mut FUS	0.958	0.953	0.957	0.822	0.853
15 IGFBP1-3	0.717	0.702	0.713	0.569	0.697
16 hnRNPC-1	0.960	0.957	0.959	0.885	0.930
17 hnRNPC-2	0.975	0.973	0.976	0.941	0.953
18 hnRNPL-1	0.756	0.771	0.746	0.392	0.698
19 hnRNPL-2	0.747	0.769	0.746	0.474	0.708
20 hnRNPL-like	0.708	0.711	0.679	0.562	0.650
21 MOV10	0.813	0.804	0.812	0.783	0.803
22 Nsun2	0.835	0.803	0.801	0.754	0.779
23 PUM2	0.962	0.950	0.955	0.939	0.914
24 QKI	0.966	0.962	0.961	0.924	0.932
25 SRSF1	0.887	0.874	0.875	0.839	0.838
26 TAF15	0.964	0.956	0.963	0.804	0.850
27 TDP-43	0.930	0.926	0.930	0.883	0.907
28 TIA1	0.930	0.924	0.926	0.842	0.896
29 TIAL1	0.893	0.888	0.895	0.831	0.858
30 U2AF2	0.953	0.941	0.945	0.861	0.873
31 U2AF2(KD)	0.931	0.923	0.930	0.840	0.883

DeepBind, DeeperBind, Oli and GraphProt perform on the same datasets with iDeepS. The boldface indicates this performance is the best among the compared methods

ISSN: 1471-2164