How short an Illumina read needs to be discard?

[lan_ub]

I'm working on ancient DNA mitochondrial genome mapping. We extracted genomic DNA from "fossil" bones. Then we did paired-end Illumina Hiseq shotgun sequencing using single stranded library. (Based on protocol from this paper: http://www.nature.com/nprot/journal/....2013.038.html)
The problem for ancient DNA is that, most of the DNA fragments we extracted are exogenous DNA (from microbes or human or other organisms). What we want to do next is to try to map reads to a reference mitochondrial genome.

At first, I trimmed the adapters. Then I discarded the reads less than 20bp. After using BWA, only a few hundred reads can be mapped to the reference. If I keep all the reads, no matter how short they are, more than 10,000 reads can be mapped. So, my question is, what is the minimum lenth for an Illumina read that can be used? If there's any paper discussed this question that would be fantastic. Thanks in advance!

[GenoMax]

Have you checked to see what other groups have done for difficult samples (e.g. neandrathals, denisovans etc) such as yours?

If only something smaller than 20 bp is hybridizing then where did the rest of the DNA in that read originate from (artifact)?

[jimmybee]

I'm working on ancient DNA mitochondrial genome mapping. We extracted genomic DNA from "fossil" bones. Then we did paired-end Illumina Hiseq shotgun sequencing using single stranded library. (Based on protocol from this paper: http://www.nature.com/nprot/journal/....2013.038.html)
The problem for ancient DNA is that, most of the DNA fragments we extracted are exogenous DNA (from microbes or human or other organisms). What we want to do next is to try to map reads to a reference mitochondrial genome.

At first, I trimmed the adapters. Then I discarded the reads less than 20bp. After using BWA, only a few hundred reads can be mapped to the reference. If I keep all the reads, no matter how short they are, more than 10,000 reads can be mapped. So, my question is, what is the minimum lenth for an Illumina read that can be used? If there's any paper discussed this question that would be fantastic. Thanks in advance!

Generally its 25bp. A 20mer will map to a variety of different genomes, if given the chance, whereas 25mers are more stringent are more likely to be "real" when mapped to your closest reference.