In an idealized world, the detection of rare bits of DNA would be constrained by sampling probabilities. Say your bacteria are present at 1 in 1 million and you have 3 million reads. Each read has a chance of 999,999/1,000,000 of not detecting your bacteria, and the overall chance of not detecting it would be (999,999/1,000,000)^3,000,000 = 5%. 

In practice, a rare DNA has other sampling probabilities involved, such as priming during PCR, and potential to be lost during ligation or other molecular biology steps. If the bacteria is similar to other genomes present and reads are distinguished by SNPs, then the sequencing error rate may dominate. For example, if you are doing 16S sequencing and your bacteria is 2 SNPs different from a related bacteria in the sample, then you'll "detect" your bacteria even if it isn't there because of sequencing errors creating that sequence every 10,000 reads of the related bacteria.

So, some empirical testing is needed.

Hi, that is really a tricky question. 

It all depends on some luck in the lib preparation (amplification step) and the depth of coverage that one sequences. The only literature that I can suggest is to read in to RNA seq papers that discuss the coverage required to obtian information on low transcripted genes/ isoforms. 

You will see that there is a certain saturation from which on you wont get much further... 

Hope that helps a bit.

In an idealized world, the detection of rare bits of DNA would be constrained by sampling probabilities. Say your bacteria are present at 1 in 1 million and you have 3 million reads. Each read has a chance of 999,999/1,000,000 of not detecting your bacteria, and the overall chance of not detecting it would be (999,999/1,000,000)^3,000,000 = 5%. 

In practice, a rare DNA has other sampling probabilities involved, such as priming during PCR, and potential to be lost during ligation or other molecular biology steps. If the bacteria is similar to other genomes present and reads are distinguished by SNPs, then the sequencing error rate may dominate. For example, if you are doing 16S sequencing and your bacteria is 2 SNPs different from a related bacteria in the sample, then you'll "detect" your bacteria even if it isn't there because of sequencing errors creating that sequence every 10,000 reads of the related bacteria.

So, some empirical testing is needed.

Thanks for your detailed information, since some bacteria will invade into the tissue, so doing microbiota analysis in the tissue by using NGS will lost many information? or not? since there are so few in number for some specific type of bacteria, what about the cell culture method? Thanks!

Perhaps my experience in detecting Helicobacter pyl. in gastric biopsy can help you. I did tests for the limit of detection of Hp done with a combined PCR-melting curve-analysis. I found that about 5-10 bacterias (that means 7-18 femtogramm DNA) can be detected in a background of 50 ng human DNA in one PCR-tube. One can calculate how many bacterias must be "catched" in the analysed bit of tissue to get a positive result. In respect to the former answers I think NGS does not reach the same sensitivity, because human DNA is amplified in the same or higher factor than the bacterial DNA while preparing the library. With classic PCR only bacDNA is amplified. (NGS-Experts: please tell me, if I'm wrong.)

Thanks, Dr. Gudrun, How to calculate how many bacterias must be "catched" in the analysed bit of tissue to get a positive result? I mean how many volume (weight) of biopsy sample you need to collecct the bacteria DNA? Thanks! 

With calculate I rather ment estimate ; ) .

In my case we looked at the number of bacterias in histological sections and learned that in cases with less than 5 bacterias in the section the PCR-MCA was not successful. There had to be a sufficient number of bacterias to get a positive result. Maybe this could also be helpful for your research to get first a "picture" of the amount of tested bacterias.

I think the amount of your input-DNA is restricted by the used detection-method.

Another thread of researchgate dealt with the question about enrichment of bacterial DNA in human background.

I don't think that with NGS the human DNA will get amplified during lib preparation since you will have 2 options:

1) 16S sequencing: you are using universal primers for bacteria, so no DNA from host will be amplified. With regards to depth of sequencing to catch the rare biosphere, usually 300k-500k reads/sample (which is cheap and easily accessible bi illumina Hiseq platforms) suffice and there is not much more you can get with higher depths. The problem would be that 16S sequencing doesn't provide you the resolution to claim about species or strains. So, not a good choice for detecting pathogens.

2) shotgun sequencing: here you can avoid PCR amplification and its related bias. There will be human genomic DNA in your library that you can remove through bioinformatics (there are even some approaches that enrich bacterial DNA during extraction protocol). This approach will provide the resolution at species level (or even strains). The problem would be the required depth of sequencing that you need. no one can give you exact number, the best approach would be to test one of your samples at a very high depth (lets say 20m) reads and then do a step wise test to see at what depth you've reached the highest coverage (lets say for example after 6m reads your coverage estimates have reached the peak) and then apply that depth to the rest of your samples to avoid extra charges.