I've continued to look at how to get Augustus to faithfully
predict from TAR evidence, and have an improvement to look at,
using these TAR regions where other predictors failed to find genes.

>   Here is a table of about 160 regions on Daphnia scaffold_4 that
>   have TARs outside of gene predictions.
>   http://insects.eugenes.org/species/data/work/dpx-scaffold_4-qt.nopredict.html

This new prediction set "Augustus+TarAll.14 (cds)" uses 
a version of Augustus that I modified somewhat to better use
this TAR data.  It seems to produce useful predictions in the
areas that Gnomon & JGI missed.  It is finding exons that fairly
well match the TAR regions, though I think its gene models 
are not as good as they could be.  One can at least use these
predicted proteins to estimate with blastp and interproscan what
some of the functional contents of these unpredicted genes are.

While several of these new predictions that I've checked by NCBI
blast look like weak matches to various things, some are like
this rather certain finding of extra tandem duplicates. JGI 
calls three genes in this region, but the TAR data
indicates missed exons, and Augustus predictions from TAR
show six near identical tandems with high identity:

http://insects.eugenes.org/cgi-bin/gbrowsenew/gbrowse/dpxaugust/?name=scaffold_4:2427015..2444264;
label=JGI-Gnomon-Tar999-Tar99-Tar95-AugT14-AugY

( in the middle find 1 JGI skipover gene call, 2 Gnomon skipovers 
  covering 4 tandem duplicate genes with TAR exon evidence).

-- Don Gilbert, Jan. 2008