Table 4. Performance evaluation of various input encoding methods using the dataset of Lee and Han [21].

API classifier	Protein encoder	Aptamer encoder	Sensitivity	Specificity	Accuracy	Youden’s Index	MCC
Ranfom Forest	CTD	iCTF	0.862	0.516	0.689	0.379	0.405
	CTD	PseKNC(k = 2)	0.842	0.537	0.69	0.379	0.399
	CTD	PseKNC(k = 3)	0.855	0.52	0.687	0.375	0.399
	CTD	TAC	0.715	0.6	0.658	0.315	0.319
	DPC	iCTF	0.933	0.454	0.693	0.387	0.442
	DPC	PseKNC(k = 2)	0.928	0.474	0.701	0.402	0.452
	DPC	PseKNC(k = 3)	0.932	0.462	0.697	0.394	0.448
	DPC	TAC	0.881	0.561	0.721	0.442	0.469
	iCTF	iCTF	0.931	0.493	0.712	0.424	0.473
	iCTF	PseKNC(k = 2)	0.949	0.499	0.724	0.448	0.502
	iCTF	PseKNC(k = 3)	0.952	0.498	0.725	0.45	0.506
	iCTF	TAC	0.887	0.567	0.727	0.454	0.481
	TPC	iCTF	0.931	0.459	0.695	0.389	0.443
	TPC	PseKNC(k = 2)	0.997	0.466	0.731	0.463	0.546
	TPC	PseKNC(k = 3)	0.986	0.466	0.726	0.452	0.53
	TPC	TAC	0.978	0.492	0.735	0.47	0.538
[21]			0.768	0.661	0.714	0.429	0.431

The bold font denote the best result in each performance metric. According to MCC, TPC+PseKNC(k = 2) was selected as our final choice of encoders. All the results in detail are available in S7 Table.