shhh.flows denoising

Hello

I would like to know how mothur can determine the noise level in a flowgram
what is the calculation process

thanks in advance

I’m not sure what you mean… Perhaps you’re referring to this:

http://www.mothur.org/wiki/Trim.flows#signal_.26_noise

Hi

Does that mean the more signal under 0.70, longer will be the denoising ?

And I got two another tiny question …

Let’s say you have a flow.files containing 8 flows file (made by trim.flows with an oligos file)
is the error modeling will be applied on each flow file independently (is the error modeling reset after each files)
or is the error modeling is updated by each files, so all files can use it ?

And is there a way to evaluate the quality of a denoising (like a function in mothur or another program)

Thanks

Does that mean the more signal under 0.70, longer will be the denoising ?

No, I’d have to double check but I think anything less than 0.7 is considered a 0 and anything above is a real intensity.

is the error modeling will be applied on each flow file independently (is the error modeling reset after each files)
or is the error modeling is updated by each files, so all files can use it ?

The model is fixed- that’s the look up file. We then denoise each sample separately.

And is there a way to evaluate the quality of a denoising (like a function in mothur or another program)

Sequence and process a mock community in parallel :slight_smile:


Also, I should add that for more details about the algorithm, I'd strongly encourage you to look at the original Quince papers. The only difference between what we do and he does is that we trim all of the flow grams to a common number of flows. We do this because we found that the denoising was pretty lame if you allow a range of flow numbers (e.g. 360 to 720) and we got a much lower error rate when we fixed it at a single number (e.g. 450).

Pat

So if I take a sample of 25000 sequences and I cut it in 2 part and then I put them in a flow.files, the denoising will be less good then if I did’nt cut them , Right ?

Thanks

P.S I’m gonna check the Quince Paper too

Correct.

Just to be sure

If I got 3 samples
If I launch 3 instances of the program on each samples, the denoising will be the same then using 1 instance of the program with a flow.files containing the 3 samples

Am I right ?

And if I use the large=1000 parameter, will it affect the denoising ?
If I use the Large parameter, if I’m right, mothur cut the flow in many pieces, but I want to know, are these pieces are considered different samples (Different error modeling), or the same sample (using the same error modeling)

Thanks

If I launch 3 instances of the program on each samples, the denoising will be the same then using 1 instance of the program with a flow.files containing the 3 samples

Am I right ?

Correct, within some margin for randomness.


And if I use the large=1000 parameter, will it affect the denoising ?
If I use the Large parameter, if I’m right, mothur cut the flow in many pieces, but I want to know, are these pieces are considered different samples (Different error modeling), or the same sample (using the same error modeling)

Different samples and then they’re brought back together at the end and uniqued.

Pat