Swingle, et al. describe three hidden Markov models for PvdS boxes:

A) Derived from experimental data in P. syringae pv. tomato DC3000,
manually realigned to optimize conservation of nucleotides in -35 and
-10 regions.

B) Derived from regular expression search for generalized IS boxes in
the pyoverdine locus in P. aeruginosa PAO1 (cf. Ravel, et al.).

C) Combined model obtained by aligning set B to set A.

The aligned training sets for each model are provided in Clustal
format (file name suffix .aln).  The HMM constructed for each training
set were re-calibrated for the target genome.  The complete scan
results for each replicon are supplied in GFF format (file name suffix
.gff).