AUGUSTUS gene level sensitivity and specificity

Question

[Mehmet]

Dear All,

I am making curated genes using Artemis. These genes will be used in AUGUSTUS training and then used to find the total number of genes in a worm species. My problem is that I am not able to increase the gene level sensitivity and specificity during AUGUSTUS training. I am using 100 genes for test and 400 genes for training. I also tried to optimisation step, but I got 0.6 sensitivity. I am only considering exon-intron boundaries during manual gene curation instead of checking on Blastp.

Thank you

Answer 1

[Juke34]

It's really complicated to get a high sensitivity and specificity at gene level. We have figured out that this difficulty come by the fact that only one mismatch in your gene definition is sufficient to not take it into account for this calcul.

Only few nucleotide in less or in more ( like a different splice site position or an exon in less or in more <= it often happens with isoforms ) among a gene of several thousand nucleotide long will preclude the gene for the sensitivity calcul. In the same time 95% of exon of this gene can be exactly identical to the template.

Having 0.6 sensitivity is a relatively good result.

[Mehmet]

After I get good gene level (~0.6 or more) sensitivity, I will use the species parameters to make a new gene prediction using AUGUSTUS. I tried before but I found around 20,000 genes in genome of a worm. Then, I checked two relatively close species (my species and other species) in terms of the total gene number. For example, My species had 20.000 genes while other species had 16.000 genes. Is this acceptable?

[Juke34]

The result is not so bad. Find more than expected is quite normal with ab-initio approaches. It often overpredicts.