how to deal with adjoining SNPs?

[cometarossa]

Dear All,
I'm dealing with some sequencing data coming from an Illumina platform and assembled on a reference genome. We employed standard filters to reduce gaps (and therefore false mismatches), but still I'm finding numerous adjoining SNPs in the dataset. I'm afraid these could be artifacts caused by reads assembly; people in bibliography present some contrasting opinions, sometimes considering these kind of SNPs as regular (and valuable) data, some other times getting rid of them.
What I'd like to ask you is if anybody supports biological reasons at the base of adjoining SNPs formations ore else if there is any filtering tool devoted to this problem. I could brutally eliminate these SNPs from the dataset, but maybe somebody came up with a more elegant way .

[pmcget]

You could have a mixture of real and artifactual MNPs - so a brute force filtering of them wouldn't be advisable.

Are you seeing a read-position bias to the adjacent SNPs? If you are then these could be a symptom of declining base quality at the end of reads. If so read trimming would remove this artifact e.g. using the bwa aln -q option.
Once the position bias is eliminated then you could have some real MNPs remaining (multi-nucleotide polymorphisms).

We have found some of these ourselves (validated by Sanger sequencing):

http://www.ncbi.nlm.nih.gov/pubmed?term=21901792

Also see Rosenfeld et al 2010

http://www.ncbi.nlm.nih.gov/pubmed/20488869

These MNPs can cause problems for mutation annotation software as usually they report the adjacent mutations separately - rather than the combined effect of both mutations on the codon.

[cometarossa]

Thank you pmcget for the kind reply. That are some really interesting clues, I'll work in that direction to untangle this knot.
Cheers!