Filtering on quality

[Farhat]

What kind of trimming and/or quality filter do you use before mapping reads? Are there any commonly accepted filtering criteria?

I am trying to map some data from an older sequencer and my filtering (Q>15, no more than 6 bases outside quality range allowed) seems to filter out about 85% of the reads.

[Jose Blanca]

I use a combination of several tools lucy, exonerate, blast, Univec, quality trimming. If you want to take a look I wrote a document about our read cleaning process.

[strob]

fastx toolkit (incorporated in Galaxy)

[Farhat]

Thanks. I am aware of software that does this, what I would like to know is the criteria used in filtering, (e.g., Q>15, maximum six bases outside the quality range, etc.)

[Jose Blanca]

Our default pipelines depend on the kind of sequences (mainly on the sequencing platform). Each pipepile has several steps (you can look for them in the ngs_backbone page). For these steps our defaults are:

For the unknown vectors Blast against the UniVec database. We keep the hits with more than 15 bases and a similarity above 96%. We remove this regions from the reads.

For the adaptors longer than 15 bases, exonerate alignment. Again we keep the hits with more than 15 bases and a similarity above 96%.

We look for the adaptors shorted than 15 bases using regular expressions.

Quality for short sequences. We apply a quality threshold of 20 and a window of 1. We keep the regions with more than 3 good bases.

Quality for the long sequences. We use lucy with the default parameters. If we have the cloning site we use it.

For the sequences without quality we use trimpoly to look for N and remove the regions with too many Ns.

Sometimes, we remove 3 or 4 nucleotides from the edges.