EvidentialGene :
  Evidence Directed Gene predictions for Eukaryotes 

  Don Gilbert, gilbertd at indiana edu  
  Summer-fall 2010
  See http://arthropods.eugenes.org/genes2/

Annotation summary

Informant evidence for gene models is transcript and protein
data.  Transcripts from long (e.g. dbEST) and short reads (e.g.
SRA-Illumina), are mapped to the genome assembly with GMAP (long
reads) or GSNAP (short reads).

EST assemblies are constructed using PASA, from all transcript
data.  Rnaseq assemblies are first constructed from aligned short
reads with Cufflinks, and combined with EST reads for a full EST
assembly.

Protein genes from related sequenced genomes are BLASTX aligned to
repeat/transposon-soft-masked genome, then refined with Exonerate to 
protein gene models.

Gene models are predicted with evidence-directed AUGUSTUS
predictor. Augustus is trained for gene parameters
using full cDNA genes derirved from the PASA EST assemblies.

Several prediction sets from different evidence sets and
parameters are combined, selecting highest evidence scoring model
per location.  A final consensus gene set has the best match to
EST and protein evidence, including full gene models from cDNA
and plant genes.

Predicted genes are UTR-extended and/or improved by PASA. Genes
are annotated with Uniprot descriptions, and classified by
evidence scores including transposable elements.

Software references:
  AUGUSTUS: http://augustus.gobics.de/
  GMAP-GSNAP: http://research-pub.gene.com/gmap/
  PASA: http://pasa.sourceforge.net/
  Cufflinks: http://cufflinks.cbcb.umd.edu/
  exonerate: http://www.ebi.ac.uk/~guy/exonerate/
  NCBI BLAST:  blast.ncbi.nlm.nih.gov/

This has been applied in 2010 to Daphnia pulex v2, Daphnia magna v1, 
  Pea aphid v2, Nasonia vit v2, Th. cacao v1,
  
See also
  Perfect(ing) Arthropod Genes with Next Gen Informatics 
  Talk & poster at Arthropod Genomics Symposium June 2010 Don Gilbert
  http://arthropods.eugenes.org/arthropods/about/
  arperfgenes1006talk.pdf


Annotation Overview
===================
0.  prepare genome
    {MY_SPECIES} assembly from NCBI genbank/genomes or elsewher
    parallelize genome and evidence data for TeraGrid cluster runs.
    
1.  collect gene evidence
    1.1 Est from NCBI dbEST mapped to genome with gmap
    1.2 RNAseq from NCBI SRA, mapped to genome with gsnap
        assembled to contigs with cufflinks, velvet (denovo), other
    1.3 Proteins from related genomes,
        mapped to genome with blastx + exonerate
    1.4 Reference/validated gene transcripts (from version 1, etc)        
    1.x Repeat finding, tRNA/etc finding 
    
2.  prepare augustus config, evidence hints
    2.0 training genes from full gene EST assemblies by PASA 
    2.1 train augustus for species/MY_SPECIES/ 
    2.2 Augustus configs and evidence hints  
        
3.  predict with several evidence sets
    3.1 : _epir = EST ep + Protein cp + all intron = classical gene evidence
    3.2 : _ep = EST assembly + introns
    3.3 : other weighted evidence partitions as needed
    3.4 : collate partitioned results, convert augustus.gff to augmap.gff.
    
4.  annotate predictions with evidence
    4.1. evidence annotate by base overlaps: protein, est, rnaseq, intron, 
    4.2. mark ncRNA or at least genes < 40aa
    4.3. homology, parology annotation from predicted proteins

5.  evaluate predictions, pick best
    5.1  Choose best model/location from weighted evidence scores
    5.2  Summarize quality scores

6.  PASA to add alternate transcripts, UTR extensions
    from EST and RNAseq assemblies 

7.  Genome map views in GBrowse
    7.2 repeats features including transposons

Software scripts and recipe details are available.