Computational Biology and Bioinformatics

SIEVE

Description:

To run SIEVE, use the online tool. This server implements the algorithm described in our paper (PLoS Pathogens, 5(4); e1000375.) to predict type III secreted effectors using protein sequence alone. Upload a fasta format protein sequence file then hit the 'Submit Job' button. The server will process your input file and make predictions of the likelihood that each input sequence is secreted by a type III system. After completion the file returned will provide a table with your protein identifiers (supplied in the fasta file), SIEVE Zscore (the SIEVE raw discriminant adjusted by the mean and standard deviation from a large prediction set), and the raw SVM discriminant score. From experimental validation studies, (Infection and Immunity 79(1): 33-43), we have established that a SIEVE score greater than 1.5 is a very confident prediction.

Type III secretion systems are used by Gram-negative bacteria to deliver proteins directly to the host cell cytoplasm. These proteins, called effectors, have a range of functions but generally interact with host cell pathways to allow the bacteria to evade host cell defenses and establish an environment (either intracellular or extracellular) that allows the bacteria to persist and replicate. Effectors are known to have a signal encoded in their N-terminal regions that directs their secretion by the type III secretion apparatus, but this signal does not have readily identifiable sequence motifs that allow its identification using traditional sequence analysis methods. Our method uses a machine-learning approach (a support vector machine, SVM) to learn patterns from known effectors. The learned patterns can then be used to provide very accurate predictions of secretion for new proteins (as in this server). It is important to note that these are only predictions and it is likely that even some predictions with relatively high probabilities will not be secreted. Experimental validation is necessary to demonstrate actual secretion.

Two other methods exist for prediction of type III secreted effectors that use very similar approaches. Using a recently validated set of proteins we have found that these methods are as effective as ours at predicting novel effectors, but that each predicts a slightly different set of effectors.

Links to the other servers are:

EffectiveT3 Server
ModLab Prediction Server

For questions and further information please see the Help tab above, or contact

Computational Biology & Bioinformatics

Research Areas & Software

Online Tools

About CBB