Tutorial with Docker¶
- What do you want to simulate? How many simulations?
- Create your model.csv and param.txt input files.
- Perform a small test simulation.
- Perform high-throughput simulations.
1. Define your simulation¶
- What do you want to simulate? How many simulations?
- Suppose we want to simulate a full chromosome with a locus size of 200Mb, with a two population split model, with one population size change, where there first population has a sample size of 10 diploid individuals and the second population has a sample size of 70 diploid individuals. And we want to use priors on all of parameters. Suppose we want a total of 50,000 simulations.
- Draw your model.
- Do this on paper first.
____
| |
| 1 |
|_ _|-en
| |____
| ___ |-ej
| | | |
| | | |
1 2
2. Create input files¶
Create the model_file_tutorial.csv and param_file_tutorial.txt files.
model_file_tutorial.csv
# Use the simulator MaCS
-macs,./bin/macs,
# Simulate a locus size of 10kb. Start with 10kb, then increase to 200Mb
-length,10000,
# Use a mutation rate of 2.5e-8
-t,2.5e-8,
# Use a recombination rate of 1e-8
-r,1e-8,
# Tell MaCS to retain 1e5 previous base pairs
-h,1e5,
# Two populations, first with sample size 10 diploid individuals and 2nd with sample size 70 diploid individuals
-I,2,20,140,
# Effective population size of population 1 defined as A
-n,1,A,
# Effective population size of population 2 defined as B
-n,2,B,
# Divergence event from 1 to 2
-ej,AB_t,2,1,
# Population size change in population 1 to size AN
-en,AN_t,1,AN
param_file_tutorial.txt
A = (1e3:1e4.0)
B = (1e3:1e4.0)
AB_t = (1000:4000)
AN_t = (0:4000)
AN = (1e4:1e5.0)
3. Perform test simulation¶
Docker requires sudo privileges. If you do not have sudo, use Singularity.
Check that Docker is installed:
sudo docker run hello-world
Quick and easy install script provided by Docker:
curl -sSL https://get.docker.com/ | sh
See Developer documentation for more information on Docker.
a. Pull Docker image¶
Pull the latest SimPrily Docker image:
sudo docker pull agladstein/simprily
Once you have successfully pulled the image you will see something like this:
Using default tag: latest
latest: Pulling from agladstein/simprily
f49cf87b52c1: Pull complete
7b491c575b06: Pull complete
b313b08bab3b: Pull complete
51d6678c3f0e: Pull complete
09f35bd58db2: Pull complete
f7e0c30e74c6: Pull complete
c308c099d654: Pull complete
339478b61728: Pull complete
d16221c2883e: Pull complete
df211aed0ee8: Pull complete
94afb574a896: Pull complete
b253919783b5: Pull complete
45cb233ca3a5: Pull complete
Digest: sha256:1de7a99a23264caa22143db2a63794fa34541ccaf9155b9fb50488b5949a9d7d
Status: Downloaded newer image for agladstein/simprily:latest
Next, double check the images you’ve pulled:
sudo docker image ls
You should see something like this:
REPOSITORY TAG IMAGE ID CREATED SIZE
agladstein/simprily latest 1d3fbe956b00 5 hours ago 938MB
b. Run SimPrily¶
Run one small example with the Docker container
sudo docker run -t -i --mount type=bind,source="$(pwd)",target=/app/tutorial agladstein/simprily python /app/simprily.py -p /app/tutorial/param_file_tutorial.txt -m /app/tutorial/model_file_tutorial.csv -i tutorial_1 -o /app/tutorial/output_dir -v
You should see something like this:
debug-1: Debug on: Level 1
JOB tutorial_1
Current Seed: 19924
debug-1: name total panel genotyped
debug-1: A 20 0 20
debug-1: B 140 0 140
debug-1: total samples: 160
debug-1: Perform simulation and get sequences
debug-1: Number of sites in simulation: 3071
debug-1: Calculating summary statistics
#########################
### PROGRAM COMPLETED ###
#########################
Then, you should see a new directory created "$(pwd)"/output_dir
.
In that directory, you should see the directories
sim_data
germline_out
results
and the directory results
should have the file results_tutorial_1.txt
, which should look something like this:
A B AN AB_t AN_t SegS_A_CGI Sing_A_CGI Dupl_A_CGI TajD_A_CGI SegS_B_CGI Sing_B_CGI Dupl_B_CGI TajD_B_CGI FST_AB_CGI
6803.19290799 5631.76173775 907706.772716 2253.4362688 1707.92117592 2490 500 193 0.648468498628 2210 37 2 2.26242085379 0.146122749866
4. Perform HTC simulations¶
a. Estimate the required resources¶
- Compare to provided benchmarking
First we should compare our model to the benchmark. Our model is “simple”, like Model 1, we have 160 diploid samples, and want to simulate 200Mb. So according to the benchmarking, we can expect our model to use approximately 1GB of memory and take about 5 min to run.
1Gb is a reasonable amount of memory for most CPUs.
50,000 simulations X 5 min / 60 per simulation = 4,167 hrs for all simulations.
- Profile the Simulation
- After performing the test simulation and before starting high-throughput simulations, the memory use and run time of this model should be assessed.
- Edit model_file_tutorial.csv so it has the desired length of 200Mb. Change it to:
-length,200000000,
- Run the simulation and time it with
time
:
time sudo docker run -t -i --mount type=bind,source="$(pwd)",target=/app/tutorial agladstein/simprily python /app/simprily.py -p /app/tutorial/param_file_tutorial.txt -m /app/tutorial/model_file_tutorial.csv -i tutorial_1 -o /app/tutorial/output_dir -v
We expect the simulation to take about 5 minutes, but the time depends on the parameters randomly chosen from the priors, so it could take less or more time.
- While that is running check
top
to see how much memory is being used.
top
You should see something like this:
top - 18:29:42 up 1:59, 3 users, load average: 1.00, 0.63, 0.28
Tasks: 142 total, 2 running, 140 sleeping, 0 stopped, 0 zombie
%Cpu(s): 98.7 us, 1.3 sy, 0.0 ni, 0.0 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
KiB Mem : 8175420 total, 5298016 free, 237632 used, 2639772 buff/cache
KiB Swap: 0 total, 0 free, 0 used. 7599724 avail Mem
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
20053 root 20 0 26492 16596 2988 R 97.3 0.2 4:35.89 macs
20022 root 20 0 172476 34128 11044 S 2.3 0.4 0:07.51 python
11488 root 20 0 442116 27524 13588 S 0.3 0.3 0:12.62 docker-containe
1 root 20 0 37952 6128 4104 S 0.0 0.1 0:08.21 systemd
2 root 20 0 0 0 0 S 0.0 0.0 0:00.00 kthreadd
In this case we see that Python is using about 170Mb of virtual memory.
b. Decide where to run your simulations¶
Depending on how fast we want all the runs to finish, we pick the number of cores we want to run on. In this case, since we expect it to only take about 4,000 hrs we could run this on the Open Science Grid or a smaller HPC, server, or cloud.
c. Run in parallel on large server¶
If we have a server with at least 100 cores, we could run the simulations in about 2 days with parallel
:
seq 1 50000 | parallel -j 100 sudo docker run -t -i --mount type=bind,source="$(pwd)",target=/app/tutorial agladstein/simprily python /app/simprily.py -p /app/tutorial/param_file_tutorial.txt -m /app/tutorial/model_file_tutorial.csv -i tutorial_{} -o /app/tutorial/output_dir
d. Run as workflow on Open Science Grid¶
Or we can use the Pegasus workflow on the Open Science Grid.
- Log onto Open Science Grid Connect
ssh user-name@login01.osgconnect.net
- Clone the entire repository. We only need the pegasus_workflow directory
git clone https://github.com/agladstein/SimPrily.git
- Go into the
pegasus_workflow
directory:
cd SimPrily/pegasus_workfow
- Copy or create the model_file_tutorial.csv and param_file_tutorial.txt from above.
- Submit a small test workflow:
./submit -p param_file_tutorial.txt -m model_file_tutorial.csv -j 10
We should see something like:
2018.06.25 11:02:08.849 CDT: -----------------------------------------------------------------------
2018.06.25 11:02:08.855 CDT: File for submitting this DAG to HTCondor : simprily-0.dag.condor.sub
2018.06.25 11:02:08.860 CDT: Log of DAGMan debugging messages : simprily-0.dag.dagman.out
2018.06.25 11:02:08.865 CDT: Log of HTCondor library output : simprily-0.dag.lib.out
2018.06.25 11:02:08.870 CDT: Log of HTCondor library error messages : simprily-0.dag.lib.err
2018.06.25 11:02:08.876 CDT: Log of the life of condor_dagman itself : simprily-0.dag.dagman.log
2018.06.25 11:02:08.881 CDT:
2018.06.25 11:02:08.886 CDT: -no_submit given, not submitting DAG to HTCondor. You can do this with:
2018.06.25 11:02:08.897 CDT: -----------------------------------------------------------------------
2018.06.25 11:02:11.948 CDT: Your database is compatible with Pegasus version: 4.8.0
2018.06.25 11:02:12.078 CDT: Submitting to condor simprily-0.dag.condor.sub
2018.06.25 11:02:12.174 CDT: Submitting job(s).
2018.06.25 11:02:12.180 CDT: 1 job(s) submitted to cluster 19334.
2018.06.25 11:02:12.185 CDT:
2018.06.25 11:02:12.190 CDT: Your workflow has been started and is running in the base directory:
2018.06.25 11:02:12.196 CDT:
2018.06.25 11:02:12.201 CDT: /local-scratch/agladstein/workflows/simprily_1529942525/workflow/simprily_1529942525
2018.06.25 11:02:12.206 CDT:
2018.06.25 11:02:12.212 CDT: *** To monitor the workflow you can run ***
2018.06.25 11:02:12.217 CDT:
2018.06.25 11:02:12.222 CDT: pegasus-status -l /local-scratch/agladstein/workflows/simprily_1529942525/workflow/simprily_1529942525
2018.06.25 11:02:12.227 CDT:
2018.06.25 11:02:12.233 CDT: *** To remove your workflow run ***
2018.06.25 11:02:12.238 CDT:
2018.06.25 11:02:12.243 CDT: pegasus-remove /local-scratch/agladstein/workflows/simprily_1529942525/workflow/simprily_1529942525
2018.06.25 11:02:12.248 CDT:
2018.06.25 11:02:12.760 CDT: Time taken to execute is 5.657 seconds
We can monitor the workflow by using the command given in the printed statement. In this case:
pegasus-status -l /local-scratch/agladstein/workflows/simprily_1529942525/workflow/simprily_1529942525
Which outputs:
STAT IN_STATE JOB
Run 02:11 simprily-0 ( /local-scratch/agladstein/workflows/simprily_1529942525/workflow/simprily_1529942525 )
Run 00:58 ┗━run-sim.sh_ID0000009
Summary: 2 Condor jobs total (R:2)
UNRDY READY PRE IN_Q POST DONE FAIL %DONE STATE DAGNAME
4 0 0 1 0 12 0 70.6 Running *simprily-0.dag
Summary: 1 DAG total (Running:1)
This means that the simprily
workflow is running (in the directory shown),
and currently one simulation is running (the 9th simulation).
We see that 12 processes have completed and the entire workflow is 70.6% done.
The email that was used to create the OSG Connect account will receive an email when the workflow is complete.
In this case the results will be written to
/local-scratch/agladstein/workflows/simprily_1529942525/outputs/
[agladstein@login01 pegasus_workfow]$ ls /local-scratch/agladstein/workflows/simprily_1529942525/outputs/
final_results.txt
final_results.txt
contains the parameter values used to run the simulations and summary statistics calculated from the simulations
for all of the simulations from the workflow.
For example:
[agladstein@login01 pegasus_workfow]$ head /local-scratch/agladstein/workflows/simprily_1529942525/outputs/final_results.txt
A B AN AB_t AN_t SegS_A_CGI Sing_A_CGI Dupl_A_CGI TajD_A_CGI SegS_B_CGI Sing_B_CGI Dupl_B_CGI TajD_B_CGI FST_AB_CGI
5344.42290079 8823.11026958 23042.8392599 1621.37069753 3576.63582673 66 14 5 0.612131940133 76 11 2 2.36972132977 0.08462380112
8626.21444024 4432.98604274 18027.9154874 3139.75475448 2737.45325718 69 24 9 -0.313129414676 41 22 2.80590304723 0.163579588626
1204.0872668 2992.49797803 61262.9413354 2599.93721571 438.20762215 243 13 33 1.72441317331 201 00 3.24176945597 0.124526585551
3696.12346086 5279.87024578 28842.820112 2699.2693586 3226.68324657 74 25 1 0.381684017383 95 63 1.33636580198 0.129678414366
5456.92727156 4876.88978622 45379.0305936 1742.81453106 620.930047461 182 38 21 0.498313835616 196 25 6 0.762716222206 0.324577494188
4506.50442054 7227.8502438 92238.4468821 2039.10543287 2451.45021427 208 31 2 1.43332513774 361 37 44 0.37981118879 0.294256698297
6305.87444634 8167.64822508 96815.9966437 3923.83195464 2226.21980366 287 76 17 0.619786936616 284 72 2.60377068841 0.140691639063
1559.12937959 4924.54508638 56393.1667057 2534.56840169 2250.6811166 55 2 1 3.13808514057 154 212 1.43704802413 0.107439420978
3781.32638852 4356.79072056 33283.5573331 1325.91016803 531.570175129 95 18 18 0.21637274122 96 410 1.03611653739 0.0946289249489
- Once the previous test workflow completes, we can scale up incrementally.
Next run 100 jobs, then run 1000 jobs. If everything there are no errors, we can run the full workflow of 50,000 jobs:
./submit -p param_file_tutorial.txt -m model_file_tutorial.csv -j 50000
Since we estimated it should take about 4000 CPU hrs to run, we can expect the OSG to finish this full workflow in a day or less, depending on the OSG’s current load.