Find now at http://wfleabase.org/prerelease/dpulex_jgi060905/
Genome assembly used:  dpulex_jgi060905 (release date 2006/09/05)

 Nine eukaryote proteome annotations (tBLASTn) [1]
  including Human, Mouse, Fruitfly, Worm, Arabidopsis, Yeast, others
 Gene predictor (SNAP/homology with location, transcripts, proteins) [2]


[1] Model organism protein blast matches (HSP groups):
Source proteomes are taken from model organism databases.
  modDM = Drosophila melanogaster (flybase) proteome
  modMM = Mus musculus (MGI) proteome
  modCE = C. elegans (WormBase) proteome
  modSC = Sacc. cer. (SGD) proteome
Protein tBLASTn matches are  grouped by HSP overlap.  Names are protein ID + HSP group.
  ID Key: CG00000_G1 = primary Gene match (best hit), CG0000_G2 = secondary Gene  match on 
  same scaffold, CG0000_S[3..n] = tertiary and further matches of same protein on same scaffold as _G1,
  CG0000_o[1..n] = further matches on other scaffolds.  Matches at p <= 1e-3 are collected,
  secondary matches that mostly overlap better matches are removed.  HSPs are grouped by both
  protein fragment overlap (same part of protein) and target genome overlap.  HSP groups
  include several gene events: alternate splice exons in same gene, tandem and distant duplications,
  new genes composed of parts of several other genes, as well as computational artifacts.
  See <a href="http://insects.eugenes.org/species/news/genome-summaries/gene-GO-function-association/">
  here for further details and use in Gene Ontology groups</a> D. Gilbert, may 06

Example HSP GFF:
  source field = MOD database; score = blast bitscore; Parent= only for HSPs (no ID) with ID as above.
  tkey = protein target HSP group (location subset); tloc = protein target location; align = no. aa residues 
  aligned
##gff-version 3
scaffold_13340  modMM   HSP     13833673        13833942        79.0    -       .       Parent=MGI:88491_G1;tk
ey=MGI:88491-HSP:23-120;tloc=23-120;align=98
scaffold_13340  modDM   HSP     11004286        11004747         230    +       .       Parent=CG8236_G1;tkey=
CG8236-PA-HSP:1-153;tloc=1-153;align=154

----------------------------------------------

[2] Gene Predictions:

 2. SNAP gene predictions with protein homology guidance (DGIL_SNO is source tag)
    This SNO version is better quality than (1) SNP by sensitivity & specificity statistics.
 (not done yet: 1. SNAP gene predictions (no homology guidance; pure ab-initio; DGIL_SNP source tag) 

Recent gene prediction qualities for twelve Drosophila genomes
have been assessed and summarized here, forming the basis for these
choices.
http://insects.eugenes.org/species/news/genome-summaries/genepredictions-compared.txt

SNAP (Korf 2004) guided by protein homology evidence is one of the
best ab-initio predictors when (a) new genes are sought, (b) there are
no close relatives with an experimentally verified genome annotation.
SNAP works well on the range of eukaryote genomes (plant to animal,
small to big) minimal homology data. The draw-back is that SNAP
overpredicts, but in a way that identifies gene-like features better
than other predictors. 

SNAP with ho. evidence produces a much closer gene mapping where
there is homology, yet retains unique gene calls in non-homologous regions.