Is there any reason to discard singletons after de-hosting metagenomic reads?

Question

[robert.murphy]

I an double de hosting metagenomic reads and am wondering if there is any reason to discard singletons after both alignment steps? The workflow is as:

1. Map to host 1
2. Convert to fastq and repair.sh to resort
3. Map to host 2
4. Convert to fastq and repair.sh to resort
5. Assemble

During steps 2 and 4 I believe the default behaviour of repair.sh is to discard singletons but I have the option to output them separately. Is there any reason not to recombine the singletons with the origin fastq file before assembly (step 5)?