We will be offering an R workshop December 18-20, 2019. Learn more.
The shhh.flows command is Pat Schloss's translation of Chris Quince's PyroNoise algorithm  from C to C++ with the incorporation of mothur's bells and whistles. Based on processing of test datasets provided by Quince, shhh.flows gives the same/similar output to AmpliconNoise. (Note: The AmpliconNoise pipeline includes a second algorithm, SeqNoise. SeqNoise does not operate on the underlying flowgrams. A mothur implementation of SeqNoise is available as shhh.seqs.) shhh.flows uses a expectation-maximization algorithm to correct flowgrams to identify the idealized form of each flowgram and translate that flowgram to a DNA sequence. Our testing has shown that when Titanium data are trimmed to 450 flows using trim.flows, shhh.flows provides the highest quality data for any other method available. In contrast, when we use the min/max number of flows suggested by Quince of 360/720, the error rate is not that great. This much improved error rate does come at a computational cost. Whereas the features in trim.seqs take on the order of minutes, shhh.flows can take on the order of hours. Running shhh.flows with large datasets without multiple processors or MPI is not suggested. You can obtain the appropriate version of MPI for your operating system at http://www.open-mpi.org/. You will also need a lookup file that tells shhh.flows the probability of observing an intensity value for a given homopolymer length. You can get mothur-compatible files here and will need to put these files either with your data or the mothur executable.
There are two ways to run shhh.flows. First, if you have single flow files that you generated in trim.flows and you want to process the through shhh.flows, you will need to use the flow option as follows:
mothur > shhh.flows(flow=GQY1XT001.A01.v35.flow)
This will generate several files including ...
- GQY1XT001.A01.v35.shhh.fasta - idealized fasta sequence data containing the de-noised sequences
- GQY1XT001.A01.v35.shhh.names - a names file that maps each read to an idealized fasta sequence
- GQY1XT001.A01.v35.shhh.qual - quality scores on a 100 point scale and should not be confused with the more conventional phred scores.
- GQY1XT001.A01.v35.shhh.groups - a group file indicating the group that each sequence in the names file comes from
- GQY1XT001.A01.v35.shhh.counts - a summary of the original translated sequences sorted with their idealized sequence counterpart
Alternatively, if you used multiple barcodes and or primers in trim.flows, then the names of the resulting flow files will be stored in a file ending in "flow.files". Using the files option will tell the shhh.flows command to process each of those flow files:
mothur > shhh.flows(files=GQY1XT001.flow.files)
This will create the 5 files from above for each barcode / primer combination plus concatenated GQY1XT001.shhh.fasta and GQY1XT001.shhh.names files that can be used as input to trim.seqs.
Although MPI is not required to run shhh.flows, to get all you can out of all of your processors, you really need to use MPI and the MPI version of mothur. To run shhh.flows without MPI, but still get a small boost in performance you can do the following:
mothur > shhh.flows(files=GQY1XT001.flow.files, processors=8)
This will allow you to use 8 processors (for mac and linux boxes) in the initial distance calculation step, but will only use 1 processor during the expeectation-maximization step. Alternatively, you can get the full boost in both stages with mpi by running the following from the command line:
mpirun -np 8 mothurMPI "#shhh.flows(file=GQY1XT001.flow.files)"
A lookup file is required to run shhh.flows and it needs to be located either in the same folder as your data, next to your executable, or in the path you give this option. You can obtain the various lookup files that are compatible with mothur.
The maxiter option tells shhh.flows the maximum iterations to run if the delta value does not first drop below the mindelta value. The minimum number of iterations is 10. By default maxiter is set to 1000. If you set maxiter to 0, then the number of iterations do not matter and the mindelta criteria will be used. To change the value of maxiter, do so as follows:
mothur > shhh.flows(flow=GQY1XT001.A01.v35.flow, maxiter=1000)
The mindelta sets a threshold for determining how much change in the flowgram correction is allowed before saying that the job is done. By default this is set to 0.000001 (i.e. 10^-6). We took this default value from Chris Quince. It can be changed as folllows:
mothur > shhh.flows(flow=GQY1XT001.A01.v35.flow, mindelta=0.000001)
The cutoff option is used in the initial clustering step to seed the expectation-maximizaton step. Quince suggests a default value of 0.01 and we have no reason to suggest otherwise:
mothur > shhh.flows(flow=GQY1XT001.A01.v35.flow, cutoff=0.01)
The sigma option is used to set the dispersion of the data in the expectation-maximization step of the algorithm. Quince suggests a default value of 0.06 and we have no reason to suggest otherwise:
mothur > shhh.flows(flow=GQY1XT001.A01.v35.flow, sigma=0.06)
On some of the older setups a different flow order was used than what is currently used. To recreate the default you would do the following:
mothur > trim.flows(fasta=GQY1XT001.flow, order=TACG)
The next thing you will likely want to do is remove the barcode and primer sequences from each idealized fasta sequence and generate a group file like you would if you had just used the raw fasta data. You, of course, can do this with trim.seqs. For example, we might use the following command:
mothur > trim.seqs(fasta=GQY1XT001.shhh.fasta, name=GQY1XT001.shhh.names, oligos=GQY1XT001.oligos, pdiffs=2, bdiffs=1, flip=T, processors=8)
From this output you are ready to carry out the rest of your pipeline.
- First introduced - version 1.22.0.
- 1.25.0 - with outputdir didn't put all of the files to the output directory