Table 2 

The following tables display results of several structure predictions using a variety of algorithms upon datasets containing either S. cerevisiae tRNAPHE, E. coli RNase P, E. coli SSU rRNA or E. coli LSU rRNA sequences. Reading columns from left to right we show: prediction method, number of basepairs in the reference structure, number of basepairs in the predicted structure, the number of true positive basepairs in the prediction (% sensitivity as described earlier in parentheses), the number of false positive basepairs in the prediction (% selectivity as described earlier in parentheses), correlation values are the "Matthews correlation coefficient" (with approximate correlation in parentheses). Each of these MFEbased attempts to predict the famous S. cerevisiae tRNAPHE structure converges on an alternative lengthyhelix type structure. Adding prior knowledge, such as forcing modified bases in the RNA sequence to be unpaired can produce dramatic improvements. 

S. cerevisiae tRNAPHE: Single Sequence Methods 



Algorithm 
number of bps in reference 
number of bps in prediction 
True Positives (% sensitivity) 
False Positives (% selectivity) 
Correlation (%) 


RNAfold 
18 
23 
4 (22.2) 
16 (20.0) 
0.178 (21.1) 
Mfold (1) 
18 
21 
4 (22.2) 
14 (22.2) 
0.191 (22.2) 
Mfold (2) 
18 
22 
8 (44.4) 
11 (42.1) 
0.409 (43.3) 
Mfold (3) 
18 
23 
4 (22.2) 
16 (20.0) 
0.178 (21.1) 
Sfold (1) 
18 
23 
4 (22.2) 
16 (20.0) 
0.178 (21.1) 
Sfold (2) 
18 
23 
4 (22.2) 
16 (20.0) 
0.178 (21.1) 
Sfold (3) 
18 
21 
4 (22.2) 
14 (22.2) 
0.191 (22.2) 


Gardner and Giegerich BMC Bioinformatics 2004 5:140 doi:10.1186/147121055140 