T4Seeker: a hybrid model for type IV secretion effectors identification

Table 2 Performance of concatenated features and ablation studies on validation dataset

Features	SP	Precision	Recall	ACC	F1-score	MCC	AUC
Concatenated features
LSTM + ESM-average + DR	0.87	0.862	0.872	0.87	0.866	0.789	0.938
T4Seeker = LSTM + ESM-flatten-1024 + DR	0.875	0.869	0.888	0.881	0.88	0.821	0.947
Ablation studies
ESM-average + DR	0.863	0.852	0.841	0.851	0.846	0.702	0.908
ESM-flatten-1024 + DR	0.829	0.827	0.866	0.845	0.845	0.767	0.920
LSTM + DR	0.859	0.859	0.839	0.847	0.848	0.79	0.921
LSTM + ESM-average	0.874	0.862	0.85	0.858	0.853	0.793	0.934
LSTM + ESM-flatten-1024	0.871	0.865	0.793	0.825	0.82	0.722	0.929

AUC of T4Seeker is the highest among fusion features on validation sets, so it is bolded

ISSN: 1741-7007