T4Seeker: a hybrid model for type IV secretion effectors identification

Table 1 Performance of single features of MLP on fivefold cross-validation test

Features	SP	Precision	Recall	ACC	F1-score	MCC	AUC
Amino acid composition features
kmer	0.816	0.812	0.827	0.821	0.819	0.641	0.873
DR	0.853	0.841	0.833	0.842	0.836	0.683	0.903
AAC	0.835	0.826	0.820	0.827	0.821	0.655	0.884
DPC	0.816	0.812	0.827	0.821	0.819	0.641	0.873
TPC	0.813	0.814	0.855	0.833	0.833	0.667	0.884
PC-PseAAC	0.815	0.81	0.825	0.818	0.815	0.638	0.879
PC-PseAAC-General	0.815	0.81	0.825	0.818	0.815	0.638	0.879
SC-PseDNC	0.812	0.812	0.850	0.831	0.831	0.661	0.886
SC-PseDNC-General	0.812	0.813	0. 85	0.831	0.831	0.661	0.887
GAAC	0.702	0.708	0.749	0.727	0.727	0.452	0.77
GDPC	0.788	0.773	0.766	0.776	0.769	0.552	0.827
C/T/D features
CTDC	0.780	0.786	0.847	0.813	0.814	0.627	0.859
CTDD	0.58	0.632	0.687	0.636	0.624	0.3	0.702
CTDT	0.784	0.789	0.849	0.816	0.817	0.632	0.866
Evolutionary scale modeling features
ESM-flatten	0.828	0.827	0.866	0.846	0.845	0.692	0.899
ESM-flatten-1024	0.809	0.815	0.895	0.847	0.85	0.723	0.908
ESM-average	0.864	0.864	0.903	0.882	0.882	0.766	0.922
ESM-max	0.826	0.83	0.895	0.84	0.841	0.720	0.898
Long short-term memory features
LSTM	0.842	0.833	0.804	0.825	0.818	0.724	0.906

On the validation set, DR, ESM-flatten-1024, LSTM AUC is over 90%, which has further mining potential, so they are bold

ISSN: 1741-7007