[Gene Prediction] AUGUSTUS program

[syintel87]

Hello, I am running AUGUSTUS which is a program for gene prediction.
If there is someone who has experience of using this program, would you help me?

- My organism: Meloidogyne Chitwoodi
- What I have: ESTs of M.chitwoodi, de novo assembled contigs of M.chitwoodi
- What I did: 1) I ran blat with ESTs and contigs of M.chitwoodi in order to generate hints in gff format. 2) I ran augustus with hints and contigs of M.chitwoodi by setting brugia as species parameter.

- My questions are:
1) How much the species parameter would have an impact on the result? With what proportions are the results affected by "hints" and "species"?
2) If I want to use more information from "hints" than "species", how could I control this? (For example, in cfg file, could I adjust some numbers to give more weights on hints than species?)
3) What is the difference between ep(exon part) and exon?

Thanks in advance.

[Wallysb01]

First, from what I understand, its not a great idea to use blat to have mapped the ESTs and contigs for this purpose. You should use exonerate or run them through PASA. I’d have to double check, but I think its mostly above refining exon/intron boundaries and small indels.

1a) Species parameter will matter a lot if you don’t have a trained set of augustus parameters for an organism relatively closely related to your species. From what I can tell the brugia might be close enough, but I don’t work with nematodes, so I can’t really say for sure if gene structure is very conserved within the classes. But if its the same class, I’d be ok with it. For instance, augustus’ human parameters work pretty well for all of mammals.
1b) The hints are very important for augustus. In a sense, they sculpt the gene models when multiple options are possible. They can also allow for Augustus predicting alternative splicing.

2) I don’t know. I haven’t actually run Augustus all by itself except to train it for a new species (which didn’t have a pre-trained species in the same class). Then I ran Augustus within MAKER. I would strongly suggest you do so as well. Maker does a very good job of addressing this point you are asking about here. Basically MAKER can use several different types of information: 1) EST/de novo RNAseq transcripts from the same species 2) the same thing from other species, 3) protein alignments from other species, 4) ab initio predictors (i.e. Augustus and SNAP). What it does, is aligns those transcripts/proteins with blast, then does a local alignment with Exonerate to improve the alignments, then uses the alignments as hints for the ab initios. Finally, maker runs a series of checks to create the best possible gene model from all the information above. Its a really great pipeline that’s very user friendly and give great annotations if you have the right pieces of evidence. You should read some about it here: http://gmod.org/wiki/MAKER_Tutorial_2013

3) That sounds like an alignment gff, but I’m not entirely sure. What I’ve seen are match/match part type annotations, were in the gff the first line would be the entire range described by “match” in the second column, then all the individual exons would in lines below them with “match part” in the second column. I haven’t seen "exon part” before though, but it sound similar.