Table 1. Big-Π, NMT, PTS1: web URL, taxonomic range and prediction accuracy.

WWW-server, URL and taxonomic range	Sensitivity (%)	False-positive prediction rate (%)
Big-Π (8,10,16)
http://mendel.imp.univie.ac.at/gpi/gpi_server.html (metazoa/protozoa)	80	0.2
http://mendel.imp.univie.ac.at/gpi/plant_server.html (plants)	94	0.1
NMT (7,9)
http://mendel.imp.univie.ac.at/myristate/ (non-fungal eukarya+viruses/fungi)	>95	0.5
PTS1 (19,20)
http://mendel.imp.univie.ac.at/myristate/ (fungi, metazoa, other eukarya)	>90	0.2–0.8

All servers request a single fasta-formatted sequence and a taxon selection as input. Only for taxonomic ranges where the amount of learning data (especially the number of non-redundant substrate sequences) is sufficient, predictors have been developed as indicated in this table. Details of the prediction functions and their validation procedures are described in detail in the references indicated. Scores calculated from different predictors are not directly comparable. In a practical application, it is also not clear from the score alone how large is the risk of wrongly accepting the prediction as correct. Therefore, the score is translated into probabilities of false-positive prediction calculated for the queries analysed (9,16,17) following the BLAST E-value style (18). Two values characterize the accuracy of the corresponding predictors. Sensitivity or coverage is the probability to predict the biological property for a true target (with a threshold score=0). The false-positive prediction rate (which complements the specificity to 100%) assesses the probability that the biological property is assigned to a random, non-related protein sequence (with a threshold score=0).