Table 2.

Comparing the five-fold cross-validation performance of k-mer and word embedding with and without feature selection on three gold standard datasets.

Datasets	feature	Acc (%)	Sens (%)	Spec (%)	Pre (%)	MCC (%)
RPI369	k-mer	68.71	67.29	70.30	69.88	37.74
	embedding without feature selection	71.97	70.27	73.76	73.19	44.24
	embedding with feature selection	73.06	75.32	71.14	72.64	46.67
RPI488	k-mer	89.29	83.17	95.17	94.33	79.09
	embedding without feature selection	87.64	83.17	91.93	90.82	75.52
	embedding with feature selection	89.92	82.75	96.72	96.32	80.59
RPI1807	k-mer	96.88	98.44	94.96	96.04	93.72
	embedding without feature selection	96.73	97.90	95.28	96.28	93.37
	embedding with feature selection	97.10	97.89	96.14	96.91	94.13

The boldface indicates this measure performance is the best among the compared sequence feature encoding.