Finding Contaminants and Removing them

[ajkarloss]

Add option in quality check of sequences - to screen for possible contaminants
Use mash to predict the contaminants in the raw sequence
-- Prepare/Download the contaminant database from NCBI
-- Prokaryotes database - will need to be updated regularly

-- Make a summary with Håkon script - nb as such not ok for metagenomics - can be precised

PB: We need to remove phiX - maybe trimming -> ask Thomas advise on issue

[evezeyl]

as Karin said: we might some advises as for the best way of creating the database: the default database contains all sequences...

• do we have a way to clean the database: clean entry names (maybe better to modify R script name filter)?
  -- complete genomes or not - all eukaryotes sequences?
  -- the database will need to be update regularly
  • frequency updates ? can we automatize as much as possible? is it eg. possible to scheldulde a way for updating/creating database with specified parameters?

Karin do not want any modification of the files here -> maybe remove phiX and adaptors Trim - but do not output files -> send them directly in the chanel - would that be a good enough solution?
Not removing phiX and adaptors should aftect mashscreen ...

we need slight modification from Håkon's script: https://github.com/hkaspersen/misc-scripts/blob/master/scripts/mash_screen.R

• on the organism of interst (ie in Håkons' script we filter organism of interest based on name: ex: "Listeria monocytogenes" but <Listeria.monocytogenes> was not filtrered and poped up as likely contaminant because of this dot inserted in the name in the mash database -> so we might need to find an improvement of the filter.
• line 74: needs to be modified for pattern matching - according to nextflow script
• maybe add an option to transpose the output tables (question of preference - I prefer it transposed - easy to modify)
• short explanation of what the filter is/do to help selecting for options-> on bifrost/Håkon (towards 0 we get also rare reads matching and toward 1: high values filter out all of the low-abundance sequences and we only get the ones that dominate the files
• we might require some package installed for R and Bifrost/conda? (ie. had to install cairo librairy on my ubuntu system to be able to use the script - and additional svglite package in R - but maybe already in R system)

[Thomieh73]

I think this paper was really helpful when it comes to Human contamination.

Interestingly, in discussion with Jen Lu, who worked on Kraken2 I heard that when she is doing classifications of metagenomic reads, she includes a unmasked human genome, in order to catch anything that looks like human.