I think that it will depend of several factors:

1) Software and contaminant coverage: There are some tools, such as SOAPdenovo that just throw away any Kmer up to some count (255 for SOAPdenovo, variable for ABySS depending of the distribution...), so they may not be used during the assembly. It is the case for example for plant genomes. Usually you don't assembly the chloroplast (you can find between 3 and 10%, that usually means 100 times deeper coverage than the nuclear genome), but you can recover it from the discarded reads.

2) Do you have a reference genome for the contaminant ? Is this different from the target ? If both answers are yes, you can just map the reads to them and use the non mapped reads. Nevertheless if the genome is similar to your genome, you have the risk to lose some reads in the most conserved regions. Some examples could be plant genomes contaminated with some fungi. Some reads can map to the most conserved coding regions.

3) If you assembly assembly everything together, can you distinguish target and the symbionts ? Do they have "blast-able" sequences ? Do they have really different GC content ? They can be post-assembly methodologies to differentiate the sequences (blast it is the most common way).

My suggestion, if you have a reference genome and they are different, you can map them, and use samtools to get the "non-contaminant" ("Filter before option"). If not, you can just assembly them with different tools playing with the kmer count filter. Once you have the assembly, you can try to identify them with a Blast ("Filter after option").

I hope this helps,

Good luck.

I think that it will depend of several factors:

1) Software and contaminant coverage: There are some tools, such as SOAPdenovo that just throw away any Kmer up to some count (255 for SOAPdenovo, variable for ABySS depending of the distribution...), so they may not be used during the assembly. It is the case for example for plant genomes. Usually you don't assembly the chloroplast (you can find between 3 and 10%, that usually means 100 times deeper coverage than the nuclear genome), but you can recover it from the discarded reads.

2) Do you have a reference genome for the contaminant ? Is this different from the target ? If both answers are yes, you can just map the reads to them and use the non mapped reads. Nevertheless if the genome is similar to your genome, you have the risk to lose some reads in the most conserved regions. Some examples could be plant genomes contaminated with some fungi. Some reads can map to the most conserved coding regions.

3) If you assembly assembly everything together, can you distinguish target and the symbionts ? Do they have "blast-able" sequences ? Do they have really different GC content ? They can be post-assembly methodologies to differentiate the sequences (blast it is the most common way).

My suggestion, if you have a reference genome and they are different, you can map them, and use samtools to get the "non-contaminant" ("Filter before option"). If not, you can just assembly them with different tools playing with the kmer count filter. Once you have the assembly, you can try to identify them with a Blast ("Filter after option").

I hope this helps,

Good luck.

Hi Aureliano,

Thank you very much for the extensive information, it really helps for sure. I completely agree with you in all the points. I only proposed the "filter after option" because we are talking about an metazoan and its bacterial symbiont, which hopefully should make it very distinguishable. Nevertheless, I have no experience in assembling a mixture of reads from very different sources and that's why I was wondering if I could make that a real option.

In fact we were aware of the presence of the symbiont genome because I used a large reads subset to map to a close related bacteria and it maps almost fully, but I am afraid that removing the unmapped reads will not enable a further improvement of the symbiont genome. Nevertheless, for the host genome I have it clear that its best to use the decontaminated reads (filtered for host mitochondria and bacterial) even if they still carry symbiont reads.

Thanks for the attention

:)

Luis

If either of your target species (i.e., the non-model organism or the symbiont) is closely related to a species whose genome has been sequenced, perhaps you could align your reads against that genome. This might help you to distinguish which reads came from which organism. For instance, say your target non-model organism is closely related to another species whose genome has been sequenced. Reads with significant alignments would be allocated to the non-model species, while reads that don't align (or don't exceed the designated alignment threshold) would be allocated to the symbiont.

The nature of your reads is important here. Naturally, the first step you should take would be to filter and trim the sequences to remove low quality cycles from either end of each read (e.g., quality score 20 for Illumina based encoding). If they are paired-end reads with a combined length that exceeds the insert size, it would probably be a good idea to merge them prior to attempting any alignment. This would increase their "alignability" from a statistical standpoint.

Also, I recommend you take a look at Titus Brown's Khmer toolkit. I think you'll find the partitioning component particularly useful.

http://ged.msu.edu/angus/metag-assembly-2011/index.html

Hi,

from my experience you will always have contigs of the contaminant in the assembly, even after mapping the reads against a closely related species. We normally get the contigs out through the GC content and/or after the gene finding step, just looking at the one exon structure vs the spliced genes... a blast also helps.

Another though is, to minimize the amount of falsely mapped reads to just considered reads as contamination, if both mate pairs are mapping with a minimal score against the reference.

Last, if you are not sure that you eliminated too many reads, you can always do the gapfiller step for the assembly, using the complete read set. You just need to be sure to set the minimal mapping score high enough, see for example in image "-smalt_minScore"

Good luck.

Thomas

Hello Luis,

Apparently you can identify the bacterial sequences - otherwise how would you know about them? No matter which software is used, I would keep them to trace the assembly of the bacterial contigs. It would not interfere if you have sufficient computing capacities. In reality, you might have computing restrictions and try various partial assemblies including the one with filtered reads on the earliest step. You may need a lot of creativity to be able to distinguish between the target and contaminating sequences. It is unlikely that finishing the bacterial genomes of the symbionts worth the effort, but unless you accomplish the finishing either for your target or the symbionts (satisfy finishing criteria either for your target or the contaminants), you never can be sure.

Luis,

"Nevertheless, I have no experience in assembling a mixture of reads from very different sources and that's why I was wondering if I could make that a real option"

That is not a problem, you can assemble them together. I had the same situation in a recent publication of mine (doi:10.1093/gbe/evt012), where we shotgun sequenced (454) the genomes of five eukaryotes that contain a bacterial endosymbiont in their cytoplasm. It was no problem separating the contigs (assembly by Newbler) later.

Hi Luis,

We are sequencing virus genomes from samples such as urine that have a mix of host genome, mitochondria, bacterial contamination and virus. We usually find viral genome assemblies work best (fewer, bigger contigs) if the host and contaminating reads are subtracted via reference-based alignment (bwa) and samtools before we begin the de novo assembly. Some contaminating reads will slip through and may assemble into small contigs, but its usually quite easy to spot them due to gene content and coverage and filter them out.