Gene prediction with both RNA-seq mapping and genomic sequence based inputs?

[jstjohn]

Hey everyone,
I am working on assembling a new reptile genome. I will have some pretty high coverage mRNA-seq data as well. Are there any standard gene prediction techniques that utilize both mRNA-seq data and genome sequence level data to predict genes? I found a program called Conrad that looks like it could do this kind of thing utilizing a conditional random field, but it doesn't look like it has been widely used, or maintained since 2009.

Would the best option be to use separate programs to call genes using genome sequence information, and then again using the mRNA-seq information (cufflinks or something like that maybe?), and then I could go back and somehow merge the output form the two techniques? Are there any standard methods of performing this kind of merging?

Gene annotation is another thing I will want to do with the output. I am going to hand annotate a few genes, but it would be useful if there was some kind of program that does a blast similarity based annotation with the remainder of genes.

Thank you for your suggestions!
-John

[steven]

AFAIK, not a lot of fully integrative methods our there. Looks like gene finding 15 years ago or so, when "ab initio" approaches (genomic sequence only) were distinct from "similarity" methods (transcripts alignments only).
I would use Cufflinks, or try Gmorse

[darked89]

AUGUSTUS is able to use RNA-Seq data and did a good job at gene prediction (at least in one plant genome). Problem is, one has to map reads using blat, which is not weapon of choice for fast and accurate spliced RNA-Seq mapping.
Problem: unspliced reads running over short introns

[HESmith]

The approach you propose was used recently for assembly of a nematode genome. See http://www.ncbi.nlm.nih.gov/pubmed/20980554.