Table 3.

Performance comparison in terms of mean F1-score for base pairs and non-local [$(i-j)\ge 6$] non-base pairs with DCA predictors (PLMC, mfDCA, plmDCA and GREMLIN) and secondary structure predictors (RNAfold and LinearPartition) on full test sets TS1 (63 RNAs), TS2 (30 RNAs) and TS3 (54 RNAs)

	TS1		TS2		TS3
	Base pairs	Non-base pairs	Base pairs	Non-base pairs	Base pairs	Non-base pairs
Full test sets
GREMLIN	0.414	0.170	0.365	0.176	0.305	0.229
plmDCA	0.445	0.200	0.401	0.196	0.341	0.240
mfDCA	0.454	0.203	0.417	0.202	0.348	0.240
PLMC	0.427	0.205	0.388	0.205	0.331	0.241
RNAfold	0.603	0.165	0.656	0.188	0.776	0.243
LinearPartition	0.623	0.166	0.668	0.179	0.764	0.258
SPOT-RNA-2D-Single	0.566	0.531	0.568	0.555	0.680	0.717
SPOT-RNA-2D	0.690	0.642	0.643	0.590	0.711	0.730
Reduced test sets
SPOT-RNA	0.660	0.199	0.670	0.192	0.778	0.407
SPOT-RNA2	0.676	0.263	0.709	0.251	0.781	0.350
SPOT-RNA-2D-Single	0.556	0.523	0.560	0.545	0.680	0.717
SPOT-RNA-2D	0.668	0.615	0.658	0.588	0.711	0.730
Reduced test sets
RNAContact	0.408	0.474	0.446	0.508	0.358	0.457
SPOT-RNA-2D-Single	0.532	0.491	0.620	0.546	0.674	0.711
SPOT-RNA-2D	0.701	0.635	0.635	0.588	0.705	0.725

Note: Comparison with SPOT-RNA and SPOT-RNA2 on reduced test sets TS1 (35 RNAs), TS2 (15 RNAs) and TS3 (54 RNAs) after removing the sequences overlapping with SPOT-RNA’s training data. Comparison with RNAContact on reduced test sets TS1 (21 RNAs), TS2 (9 RNAs) and TS3 (52 RNAs) after removing the sequences overlapping with RNAContact training data. Bold indicates the predictor with the best performance.