Table 1. Reference-dependent alignment accuracy (precision and recall) on our in-house benchmark.

	CNFpred		HHpred-global		HHpred-local		DRNF-Viterbi		DRNF-MaxAcc
	recall	prec	recall	prec	recall	prec	recall	prec	recall	prec
	Evaluated by DeepAlign-generated reference alignments
(0,1]	0.475	0.474	0.395	0.344	0.351	0.475	0.616	0.615	0.596	0.635
(0.45,0.55]	0.240	0.236	0.141	0.111	0.118	0.268	0.382	0.377	0.353	0.394
(0.55,0.65]	0.422	0.426	0.338	0.282	0.284	0.435	0.591	0.594	0.579	0.628
(0.65,0.8]	0.601	0.596	0.530	0.468	0.470	0.580	0.748	0.747	0.732	0.763
(0.8,1]	0.830	0.829	0.778	0.732	0.774	0.865	0.848	0.845	0.816	0.837
	Evaluated by TMalign-generated reference alignments
(0,1]	0.436	0.426	0.318	0.358	0.435	0.322	0.545	0.532	0.526	0.548
(0.45,0.55]	0.206	0.194	0.094	0.115	0.221	0.094	0.313	0.292	0.288	0.308
(0.55,0.65]	0.377	0.369	0.255	0.296	0.391	0.252	0.510	0.497	0.495	0.521
(0.65,0.8]	0.567	0.555	0.436	0.487	0.545	0.436	0.676	0.666	0.656	0.676
(0.8,1]	0.779	0.780	0.704	0.751	0.756	0.865	0.815	0.814	0.801	0.825