MEA

MEA predicts RNA maximum expected accuracy structures from RNA base pair probabilities and optionally compares them to a reference structure. In a special mode it skips the prediction and compares a given structure to the reference. For the prediction, MEA allows to penalize long base pairs, using parameters alpha, beta, gamma, and delta. For the comparison of secondary structures, several measures are computed from the confusion matrix of the RNA base pairs.

Inputs

The tool accepts dot plot files as generated by RNAfold -p.

For (predicted) structure and reference, the tool accepts dot-bracket structures with pseudoknots (supporting bracket pairs (),{},[],<>,Aa,Bb,...)

Outputs

If predicting a structure, the tool outputs the sequence and the predicted dot bracket strucuture with computed score in parenthesis following the structure. This mimicks the output of the Vienna tools.

The result of structure comparison is reported as a line of numbers

TP FP FN TN SENS PPV F1 MCC

where

TP = # true positives
FP = # false positives
FN = # false negatives
TN = # true negatives
SENS = TP/(TP+FN) 'Sensitivity'http://www.bioinf.uni-leipzig.de/Software/mea/
PPV = TP/(TP+FP) 'Positive Predictive Value'
F1 = PPV*SENS / (PPV+SENS), if PPV+SENS!=0; 0, otherwise 'F1-score'
MCC = (TP*TN - FP*FN) / sqrt( (TP+FP)*(TP+FN)*(TN+FP)*(TN+FN) ) 'Mathews correlation coefficient'

Special rules for prediction evaluation:

Slide rule: tolerate shift of one base pair end by one base. This rule directly affects the number of true positives.
Conflict rule: predicted base pairs are false only if they conflict with the reference; two base pair conflict if and only if they share one end This rule directly affects the number of false positives.