CELLO is a subCELlular LOcalization predictive system for Gram-negative bacteria. 
CELLO utilizes a learning machine, Support Vector Machine, which is based on 
the multiple n-peptide composition (Yu et al, Proteins: Struct. Funct. Genet. 
2003:50:531-536). 

For multi-class SVM classification, we use the one-against-one (OAO) method.  
For the 5 classes of subcellular locations, we construct 5*(5-1)/2=10 SVM 
classifiers and each classifier is trained with proteins from two different 
subcellular locations.  For each penalty parameter and kernel parameter, 
cross validation combining with the OAO method is used for estimating the 
performance of the model. Therefore, for each model, 10 decision functions 
share the same parameter. Each protein in the test set will always get a vote 
from each binary classifier. We used the jury voting to determine the final 
assignment of locations to each sequence in the test set. After some 
preliminary experiments, we settle on the C+D+SP4+F3X5, denoting a combined 
SVM classifier trained with C, D, F3X5 and the combined training vectors SP4
, respectively. In the end we combine votes from  C, D, SP4 and F3X5 for 
final judgment. In the case of identical votes, we will give more weight 
to the votes from SP4.

C : amino acids composition
D : di-peptide amino acids composition
SP4 : partitioned amino acids composition
F3X5 : slice sequence averagely into five parts, and reduced amino acid 
       composition in which 20 amino acids are classified into four groups
       (charged, polar, aromatic and nonpolar)