Contaminants as a necessary evil?

[lorendarith]

If you do any sort of adapter trimming (complete sequence, specific kmers, barcodes...) and completely make your data devoid of these sequences, wouldn't one also eliminate any actual biological sequence that bares resemblance to these contaminants?

Wouldn't that cause a gap in the assemblies?

Does any of this make sense? D:

[mastal]

Contaminants as a necessary evil?

Yes, it makes sense.

I guess when we do adapter trimming we are assuming that the adapters have been designed so that those sequences don't map anywhere in the genome.

And there is obviously also a trade-off between removing all the contaminating adapter sequences and risking that in trimming too aggressively we will be removing some actual sequence that is not adapter.

[lorendarith]

I guess when we do adapter trimming we are assuming that the adapters have been designed so that those sequences don't map anywhere in the genome.

Agreed, but I don't think there is a combination (of if there is prolly, not many possibilities?) of bases so the sequence doesn't turn up in one or the other genome.

I tried blasting the first 25bp of the Illumina Truseq adapter against nt and you get quite some hits (genomes). Now, I don't know if the hits are actual biological sequences or leftover contaminants.

One could test it against a Sanger sequenced genome...

[krobison]

Most of the trimming you describe is applied only to the ends of reads, so the information will not be lost if it appears in the middle of reads.

But, it is definitely true that you need to think carefully about what you are screening out entirely. You may lose something valuable!

[rskr]

Yes and no. In theory you should be able to build a predictive model that accounts for a large percentage of your contaminants. IE if your adapter sequences matches at 18 bases and mismatches at three, then you could calculate the probability that it is an adapter vs. the alternative not an adapter, assuming an error rate of 95 percent in the sequencing.

For many analysts analysis it doesn't matter, remapping for example, there is just a bunch of junk left over. Other times it only becomes a problem when it breaks things, for example when an assembly or mapping has millions of mappings in a single place and the algorithms developers didn't build the software to handle that situation. And the most interesting of course are when you get crazy small p-values in completely irrelevant genes.

I think it is interesting to take common adapter sequences 454 for example and blast them against ncbi just to see what comes up.