Mean and median error

Probalign

SSEARCH

BLAST

ClustalW


Complete benchmark

35.3  30.7

38.7  33.2

41.0  34.0

47.6  50.3

Datasets with pairwise sequence identity at most 30%

66.5*  71.2*

73.0  83.4

75.9  85.3

82.9  85.0

 BLAST does not return an alignment in 425 datasets and hence they are omitted from the calculations. HMMER is not shown since queries with unalignable flanks cannot be used to produce a reliable model. There are 14 families that contain datasets with at most 30% sequence identity. Probalign has overall lowest mean and median error. Bold indicates the best performance; the difference is larger on datasets with low sequence identity and significant with Pvalue < 0.05 (indicated by *).