Table 1.

Evaluation of SCR predictions

Features used inneural network	Optimization on MCC				Optimization on Q2				AUC
	MCC	Q2	SE	SP	Q2	MCC	SE	SP
Structural features
SS	0.315	0.681	0.768	0.541	0.687	0.308	0.837	0.445	0.724
RSA	0.391	0.711	0.757	0.636	0.716	0.388	0.807	0.57	0.767
CB14	0.423	0.726	0.769	0.655	0.731	0.414	0.85	0.541	0.783
SS + RSA	0.414	0.719	0.751	0.668	0.728	0.406	0.853	0.527	0.784
SS + CB14	0.436	0.719	0.703	0.745	0.739	0.432	0.852	0.556	0.802
RSA + CB14	0.417	0.727	0.795	0.618	0.729	0.41	0.84	0.55	0.777
STR (SS + RSA + CB14)	0.433	0.726	0.747	0.692	0.735	0.429	0.824	0.592	0.797
Sequence features
PBL	0.408	0.721	0.783	0.623	0.728	0.404	0.861	0.513	0.777
SSP	0.364	0.698	0.746	0.621	0.707	0.354	0.854	0.469	0.749
RSAP	0.389	0.713	0.776	0.61	0.716	0.387	0.808	0.568	0.766
PBL + SSP	0.424	0.735	0.844	0.559	0.735	0.423	0.855	0.543	0.788
PBL + RSAP	0.405	0.727	0.842	0.541	0.727	0.402	0.868	0.501	0.775
SSP + RSAP	0.418	0.731	0.826	0.578	0.732	0.413	0.862	0.521	0.782
SEQ (PBL + SSP + RSAP)	0.423	0.725	0.765	0.661	0.733	0.417	0.865	0.522	0.788
Combined features
SS + CB14 + PBL	0.465	0.752	0.836	0.617	0.753	0.464	0.861	0.58	0.812
SS + CB14 + PBL + SSP	0.476	0.75	0.782	0.698	0.755	0.468	0.867	0.575	0.817
SS + CB14 + SEQ	0.467	0.753	0.841	0.512	0.753	0.467	0.841	0.512	0.814
STR + PBL	0.465	0.751	0.83	0.624	0.752	0.461	0.864	0.572	0.814
STR + PBL + SSP	0.468	0.751	0.815	0.647	0.752	0.461	0.853	0.589	0.814
STR + SEQ	0.474	0.753	0.809	0.662	0.755	0.471	0.846	0.61	0.814

SE and SP are sensitivity and specificity, respectively. The best two predictions in each category are shown in bold and underlined numbers.