Condel Predict outcome of nonsynonymous SNVs

[Lien]

Dear all,

For the prediction of the effect nonsynonymous SNVs have on protein function or structure, I've been using SIFT, PolyPhen2 and MutationAssessor.
However, I found the paper about Condel where they use a weighted average score to obtain a general 'ranking' for a list of mutated genes. I thought it would be very useful (and less time-consuming) to use Condel, instead of the 3 softwares separately.
However, when I use the webserver of Condel, I used genomic coordinates as an input. When I enter only 1 coordinate, I sometimes get as many as 12 different predictions. Some of them with a different location in the protein, others with the same alleles, same location and the same amino acid changes. (I added an example in attachment.) Does anybody know how this can be explained and how it can be 'prevented'?

Thanks a lot,
Lien

[swbarnes2]

Your gene has multiple splice variants. You can't fix that, it's a fact of biology. I'd say it's up to the submitter to work out which variants he or she thinks are relevant to his or her experiment.

[Lien]

I understand that when you have splice variants, the mutation will appear in different locations of the protein (say position 25 and 65 in my example).
But what should I think then of those predictions with the same location in the protein, say Q65H in my example? For this Q65H mutation, both predictions of deleterious and neutral are present, with a score ranging from 0.398 tot 0.968.

Thanks a lot

[abeldavidgp]

splice variants as well

The problem is still splice variants. If you check the 'protein id' column you will see that each row of the results corresponds to a different proteint (i.e., the translation of a different Ensembl transcript object). The Ensembl Variant Effect Predictor (and consequently condel server) resports the consequence type (a separate row in your results) for each predicted gene product that overlaps the genomic location of your input SNV.

The reason why the scores are different even if the position of the SNV in the protein is the same is that the sequences of proteins resulting from different transcripts differ, and hence each predictive method treats them differently. Different multiple sequence alignments are built for them, domain predictions may differ slightly, and so on and so forth. Now, to figure out which row(s) is meaninful for your search you need to find out which specific transcript is expressed in the tissue of your interest, in other words, which of them has a proven biological function. One first thing you can do is add all columns to the results table of the condel server to find out which proteins in your result list have a Uniprot/Swissprot ID, since only those have been actualy found through experimental procedures. That may help you reduce the list a bit.

[swbarnes2]

I understand that when you have splice variants, the mutation will appear in different locations of the protein (say position 25 and 65 in my example).
But what should I think then of those predictions with the same location in the protein, say Q65H in my example? For this Q65H mutation, both predictions of deleterious and neutral are present, with a score ranging from 0.398 tot 0.968.

Thanks a lot

Take it up with the authors of Condel. Maybe if a certain exon is missing, the protein domain is so messed up that a single amino acid change can't break it any more than it's already broken.