Improving accuracy of protein-protein interaction prediction by considering the converse problem for sequence representation

Table 3 The performance of four vector representing schemes for protein sequences

Organism	Methods	Benchmark negatives					Random negatives
		AUC	Acc	Sn	Sp	Pre	AUC	Acc	Sn	Sp	Pre
E. coli	AG-QP360	0.994	0.982	0.996	0.982	0.894	0.899	0.811	0.821	0.802	0.804
	AG-CTF	0.996	0.968	0.987	0.940	0.889	0.886	0.797	0.794	0.799	0.798
	AG-P100	0.994	0.965	0.989	0.979	0.889	0.889	0.799	0.798	0.799	0.799
	AG-Q340	0.989	0.964	0.987	0.959	0.807	0.854	0.771	0.743	0.789	0.787
S. cerevisiae	AG-QP360	0.993	0.968	0.998	0.969	0.786	0.960	0.902	0.887	0.929	0.917
	AG-CTF	0.991	0.964	0.986	0.960	0.767	0.948	0.880	0.879	0.927	0.909
	AG-P100	0.991	0.963	0.985	0.959	0.765	0.947	0.849	0.798	0.899	0.889
	AG-Q340	0.989	0.945	0.982	0.939	0.684	0.902	0.844	0.788	0.898	0.877

Cutoff for each method was set according to the maximal F-measure statistic. Acc: accuracy; Sn: sensitivity; Sp: Specificity; Pre: precision.

ISSN: 1471-2105