## Cookbook: Parallelized STRUCTURE analyses on unlinked SNPs¶

As part of the ipyrad.analysis toolkit we've created convenience functions for easily distributing STRUCTURE analysis jobs on an HPC cluster, and for doing so in a programmatic and reproducible way. Importantly, our workflow allows you to easily sample different distributions of unlinked SNPs among replicate analyses, with the final inferred population structure summarized from a distribution of replicates. We also provide some simple examples of interactive plotting functions to make barplots.

### Why structure?¶

Although there are many newer and faster implementations of STRUCTURE, such as faststructure or admixture, the original structure works much better with missing data, which is of course a common feature of RAD-seq data sets.

### A note on Jupyter/IPython¶

This is a Jupyter notebook, a reproducible and executable document. The code in this notebook is Python (2.7), and should be executed either in a jupyter-notebook, like this one, or in an IPython terminal. Execute each cell in order to reproduce our entire analysis. We make use of the ipyparallel Python library to distribute STRUCTURE jobs across processers in parallel (more can be found about that here). The example data set used in this analysis is from the empirical example ipyrad tutorial.

### Required software¶

You can easily install the required software for this notebook with a locally installed conda environment. Just run the commented code below in a terminal. If you are working on an HPC cluster you do not need administrator privileges to install the software in this way, since it is only installed locally.

In [1]:
## conda install ipyrad -c ipyrad
## conda install structure -c ipyrad
## conda install clumpp -c ipyrad
## conda install toytree -c eaton-lab


### Import Python libraries¶

In [1]:
import ipyrad.analysis as ipa      ## ipyrad analysis toolkit
import ipyparallel as ipp          ## parallel processing
import toyplot                     ## plotting library


### Parallel cluster setup¶

Start an ipcluster instance in a separate terminal. An easy way to do this in a jupyter-notebook running on an HPC cluster is to go to your Jupyter dashboard, and click [new], and then [terminal], and run 'ipcluster start' in that terminal. This will start a local cluster on the compute node you are connected to. See our [ipyparallel tutorial] (coming soon) for further details.

In [2]:
##
## ipcluster start --n=4
##

In [3]:
## get parallel client
ipyclient = ipp.Client()
print "Connected to {} cores".format(len(ipyclient))

Connected to 4 cores


## Quick guide (tldr;)¶

The following cell shows the quickest way to results. Further explanation of all of the features and options is provided further below.

In [5]:
## set N values of K to test across
kvalues = [2, 3, 4, 5, 6]

In [5]:
## init an analysis object
s = ipa.structure(
name="quick",
workdir="./analysis-structure",
)

## set main params (use much larger values in a real analysis)
s.mainparams.burnin = 1000
s.mainparams.numreps = 5000

## submit N replicates of each test to run on parallel client
for kpop in kvalues:
s.run(kpop=kpop, nreps=4, ipyclient=ipyclient)

## wait for parallel jobs to finish
ipyclient.wait()

submitted 4 structure jobs [quick-K-2]
submitted 4 structure jobs [quick-K-3]
submitted 4 structure jobs [quick-K-4]
submitted 4 structure jobs [quick-K-5]
submitted 4 structure jobs [quick-K-6]

Out[5]:
True
In [16]:
## return the evanno table (deltaK) for best K
etable = s.get_evanno_table(kvalues)
etable

Out[16]:
Nreps deltaK estLnProbMean estLnProbStdev lnPK lnPPK
2 4 0.000 -337925.500 374729.242 0.000 0.00
3 4 67.961 -16777.575 4697.207 321147.925 319225.45
4 4 12.556 -14855.100 1505.766 1922.475 18906.25
5 4 9.997 -31838.875 35252.979 -16983.775 352432.15
6 4 0.000 -401254.800 307272.233 -369415.925 0.00
In [10]:
## get admixture proportion tables avg'd across reps
tables = s.get_clumpp_table(kvalues, quiet=True)

In [14]:
## plot bars for a k-test in tables w/ hover labels
table = tables[3].sort_values(by=[0, 1, 2])

toyplot.bars(
table,
width=500,
height=200,
title=[[i] for i in table.index.tolist()],
xshow=False,
);


## Full guide¶

### Enter input and output file locations¶

The .str file is a structure formatted file output by ipyrad. It includes all SNPs present in the data set. The .snps.map file is an optional file that maps which loci each SNP is from. If this file is used then each replicate analysis will randomly sample a single SNP from each locus in reach rep. The results from many reps therefore will represent variation across unlinked SNP data sets, as well as variation caused by uncertainty. The workdir is the location where you want output files to be written and will be created if it does not already exist.

In [4]:
## the structure formatted file

## an optional mapfile, to sample unlinked SNPs

## the directory where outfiles should be written
workdir = "./analysis-structure/"


### Create a Structure Class object¶

Structure is kind of an old fashioned program that requires creating quite a few input files to run, which makes it not very convenient to use in a programmatic and reproducible way. To work around this we've created a convenience wrapper object to make it easy to submit Structure jobs and to summarize their results.

In [5]:
## create a Structure object
struct = ipa.structure(name="structure-test",
data=strfile,
mapfile=mapfile,
workdir=workdir)


### Set parameter options for this object¶

Our Structure object will be used to submit jobs to the cluster. It has associated with it a name, a set of input files, and a large number of parameter settings. You can modify the parameters by setting them like below. You can also use tab-completion to see all of the available options, or print them like below. See the full structure docs here for further details on the function of each parameter. In support of reproducibility, it is good practice to print both the mainparams and extraparams so it is clear which options you used.

In [27]:
## set mainparams for object
struct.mainparams.burnin = 10000
struct.mainparams.numreps = 100000

## see all mainparams
print struct.mainparams

## see or set extraparams
print struct.extraparams

burnin             10000
extracols          0
label              1
locdata            0
mapdistances       0
markernames        0
markovphase        0
missing            -9
notambiguous       -999
numreps            100000
onerowperind       0
phased             0
phaseinfo          0
phenotype          0
ploidy             2
popdata            0
popflag            0
recessivealleles   0

alpha               1.0
alphamax            10.0
alphapriora         1.0
alphapriorb         2.0
alphapropsd         0.025
ancestdist          0
ancestpint          0.9
computeprob         1
echodata            0
fpriormean          0.01
fpriorsd            0.05
freqscorr           1
gensback            2
inferalpha          1
inferlambda         0
intermedsave        0
lambda_             1.0
locispop            0
locprior            0
locpriorinit        1.0
log10rmax           1.0
log10rmin           -4.0
log10rpropsd        0.1
log10rstart         -2.0
maxlocprior         20.0
metrofreq           10
migrprior           0.01
numboxes            1000
onefst              0
pfrompopflagonly    0
popalphas           0
popspecificlambda   0
printlambda         1
printlikes          0
printnet            1
printqhat           0
printqsum           1
randomize           0
reporthitrate       0
seed                12345
sitebysite          0
startatpopinfo      0
unifprioralpha      1
updatefreq          10000
usepopinfo          0



### Submit jobs to run on the cluster¶

The function run() distributes jobs to run on the cluster and load-balances the parallel workload. It takes a number of arguments. The first, kpop, is the number of populations. The second, nreps, is the number of replicated runs to perform. Each rep has a different random seed, and if you entered a mapfile for your Structure object then it will subsample unlinked snps independently in each replicate. The seed argument can be used to make the replicate analyses reproducible. The extraparams.seed parameter will be generated from this for each replicate. And finally, provide it the ipyclient object that we created above. The structure object will store an asynchronous results object for each job that is submitted so that we can query whether the jobs are finished yet or not. Using a simple for-loop we'll submit 20 replicate jobs to run at four different values of K.

In [28]:
## a range of K-values to test
tests = [3, 4, 5, 6]

In [9]:
## submit batches of 20 replicate jobs for each value of K
for kpop in tests:
struct.run(
kpop=kpop,
nreps=20,
seed=12345,
ipyclient=ipyclient,
)

submitted 20 structure jobs [structure-test-K-3]
submitted 20 structure jobs [structure-test-K-4]
submitted 20 structure jobs [structure-test-K-5]
submitted 20 structure jobs [structure-test-K-6]


### Track progress until finished¶

You can check for finished results by using the get_clumpp_table() function, which tries to summarize the finished results files. If no results are ready it will simply print a warning message telling you to wait. If you want the notebook to block/wait until all jobs are finished then execute the wait() function of the ipyclient object, like below.

In [11]:
## see submitted jobs (we query first 10 here)
struct.asyncs[:10]

Out[11]:
[<AsyncResult: _call_structure>,
<AsyncResult: _call_structure>,
<AsyncResult: _call_structure>,
<AsyncResult: _call_structure>,
<AsyncResult: _call_structure>,
<AsyncResult: _call_structure>,
<AsyncResult: _call_structure>,
<AsyncResult: _call_structure>,
<AsyncResult: _call_structure>,
<AsyncResult: _call_structure>]
In [12]:
## query a specific job result by index
print struct.asyncs[0].result()

In [ ]:
## block/wait until all jobs finished
ipyclient.wait()


### Summarize replicates with CLUMPP¶

We ran 20 replicates per K-value hypothesis. We now need to concatenate and purmute those results so they can be summarized. For this we use the software clumpp. The default arguments to clumpp are generally good, but you can modify them the same as structure params, by accessing the .clumppparams attribute of your structure object. See the clumpp documentation for more details. If you have a large number of samples (>50) you may wish to use the largeKgreedy algorithm (m=3) for faster runtimes. Below we run clumpp for each value of K that we ran structure on. You only need to tell the get_clumpp_table() function the value of K and it will find all of the result files given the Structure object's name and workdir.

In [8]:
## set some clumpp params
struct.clumppparams.m = 3               ## use largegreedy algorithm
struct.clumppparams.greedy_option = 2   ## test nrepeat possible orders
struct.clumppparams.repeats = 10000     ## number of repeats
struct.clumppparams

Out[8]:
datatype                  0
every_permfile            0
greedy_option             2
indfile                   0
m                         3
miscfile                  0
order_by_run              1
outfile                   0
override_warnings         0
permfile                  0
permutationsfile          0
permuted_datafile         0
popfile                   0
print_every_perm          0
print_permuted_data       0
print_random_inputorder   0
random_inputorderfile     0
repeats                   10000
s                         2
w                         1                   
In [31]:
## run clumpp for each value of K
tables = struct.get_clumpp_table(tests)

[K3] 20/20 results permuted across replicates (max_var=0).
[K4] 20/20 results permuted across replicates (max_var=0).
[K5] 20/20 results permuted across replicates (max_var=0).
[K6] 20/20 results permuted across replicates (max_var=0).

In [29]:
## return the evanno table w/ deltaK
struct.get_evanno_table(tests)

Out[29]:
Nreps deltaK estLnProbMean estLnProbStdev lnPK lnPPK
3 20 0.000 -146836.325 445572.806 0.000 0.000
4 20 0.218 -151762.425 316342.669 -4926.100 69040.195
5 20 0.288 -225728.720 242763.400 -73966.295 70036.825
6 20 0.000 -369731.840 300321.531 -144003.120 0.000

### Sort the table order how you like it¶

This can be useful if, for example, you want to order the names to be in the same order as tips on your phylogeny.

In [32]:
## custom sorting order
myorder = [
"32082_przewalskii",
"33588_przewalskii",
"41478_cyathophylloides",
"41954_cyathophylloides",
"29154_superba",
"30686_cyathophylla",
"33413_thamno",
"30556_thamno",
"35236_rex",
"40578_rex",
"35855_rex",
"39618_rex",
"38362_rex",
]

print "custom ordering"
print tables[4].ix[myorder]

custom ordering
0      1          2      3
32082_przewalskii       1.0  0.000  0.000e+00  0.000
33588_przewalskii       1.0  0.000  0.000e+00  0.000
41478_cyathophylloides  0.0  0.005  9.948e-01  0.000
41954_cyathophylloides  0.0  0.005  9.948e-01  0.000
29154_superba           0.0  0.019  6.731e-01  0.308
30686_cyathophylla      0.0  0.020  6.820e-01  0.298
33413_thamno            0.0  0.845  0.000e+00  0.155
30556_thamno            0.0  0.892  7.000e-04  0.107
35236_rex               0.0  0.908  1.000e-04  0.092
40578_rex               0.0  0.989  2.000e-04  0.010
35855_rex               0.0  0.990  0.000e+00  0.010
39618_rex               0.0  1.000  0.000e+00  0.000
38362_rex               0.0  1.000  0.000e+00  0.000


### A function for adding an interactive hover to our plots¶

The function automatically parses the table above for you. It can reorder the individuals based on their membership in each group, or based on an input list of ordered names. It returns the table of data as well as a list with information for making interactive hover boxes, which you can see below by hovering over the plots.

In [33]:
def hover(table):
hover = []
for row in range(table.shape[0]):
stack = []
for col in range(table.shape[1]):
label = "Name: {}\nGroup: {}\nProp: {}"\
.format(table.index[row],
table.columns[col],
table.ix[row, col])
stack.append(label)
hover.append(stack)
return list(hover)


### Visualize population structure in barplots¶

Hover over the plot to see sample names and info in the hover box.

In [37]:
for kpop in tests:
## parse outfile to a table and re-order it
table = tables[kpop]
table = table.ix[myorder]

## plot barplot w/ hover
canvas, axes, mark = toyplot.bars(
table,
title=hover(table),
width=400,
height=200,
xshow=False,
style={"stroke": toyplot.color.near_black},
)


### Make a slightly fancier plot and save to file¶

In [37]:
## save plots for your favorite value of K
table = struct.get_clumpp_table(kpop=3)
table = table.ix[myorder]

mean scores across 20 replicates.

In [38]:
## further styling of plot with css
style = {"stroke":toyplot.color.near_black,
"stroke-width": 2}

## build barplot
canvas = toyplot.Canvas(width=600, height=250)
axes = canvas.cartesian(bounds=("5%", "95%", "5%", "45%"))
axes.bars(table, title=hover(table), style=style)

## add names to x-axis
ticklabels = [i for i in table.index.tolist()]
axes.x.ticks.locator = toyplot.locator.Explicit(labels=ticklabels)
axes.x.ticks.labels.angle = -60
axes.x.ticks.show = True
axes.x.ticks.labels.offset = 10
axes.x.ticks.labels.style = {"font-size": "12px"}
axes.x.spine.style = style
axes.y.show = False

## options: uncomment to save plots. Only html retains hover.
import toyplot.svg
import toyplot.pdf
import toyplot.html
toyplot.svg.render(canvas, "struct.svg")
toyplot.pdf.render(canvas, "struct.pdf")
toyplot.html.render(canvas, "struct.html")

## show in notebook
canvas

Out[38]:

### Calculating the best K¶

Use the .get_evanno_table() function.

In [6]:
struct.get_evanno_table([3, 4, 5, 6])

Out[6]:
Nreps deltaK estLnProbMean estLnProbStdev lnPK lnPPK
3 20 0.000 -146836.325 445572.806 0.000 0.000
4 20 0.218 -151762.425 316342.669 -4926.100 69040.195
5 20 0.288 -225728.720 242763.400 -73966.295 70036.825
6 20 0.000 -369731.840 300321.531 -144003.120 0.000

### Testing for convergence¶

The .get_evanno_table() and .get_clumpp_table() functions each take an optional argument called max_var_multiple, which is the max multiple by which you'll allow the variance in a 'replicate' run to exceed the minimum variance among replicates for a specific test. In the example below you can see that many reps were excluded for the higher values of K, such that fewer reps were analyzed for the final results. By excluding the reps that had much higher variance than other (one criterion for asking if they converged) this can increase the support for higher K values. If you apply this method take care to think about what it is doing and how to interpret the K values. Also take care to consider whether your replicates are using the same input SNP data but just different random seeds, or if you used a map file, in which case your replicates represent different sampled SNPs and different random seeds. I'm of the mind that there is no true K value, and sampling across a distribution of SNPs across many replicates gives you a better idea of the variance in population structure in your data.

In [9]:
struct.get_evanno_table([3, 4, 5, 6], max_var_multiple=50.)

[K3] 4 reps excluded (not converged) see 'max_var_multiple'.
[K4] 11 reps excluded (not converged) see 'max_var_multiple'.
[K5] 1 reps excluded (not converged) see 'max_var_multiple'.
[K6] 17 reps excluded (not converged) see 'max_var_multiple'.

Out[9]:
Nreps deltaK estLnProbMean estLnProbStdev lnPK lnPPK
3 16 0.000 -13292.675 183.466 0.000 0.000
4 9 8.886 -24653.167 17188.640 -11360.492 152730.426
5 19 1.759 -188744.084 182567.986 -164090.918 321182.668
6 3 0.000 -31652.333 19012.073 157091.751 0.000

### Copying this notebook to your computer/cluster¶

You can easily copy this notebook and then just replace my file names with your filenames to run your analysis. Just click on the [Download Notebook] link at the top of this page. Then run jupyter-notebook from a terminal and open this notebook from the dashboard.