SMASH

This is a collection of the elements needed from Peter Andrews' SMASH, which requires elements from Peter Andrews' MUMdex code.

All content under peter/ belongs to Peter Andrews. Source code under peter/ may have been modified by Ryan A. Hagenson for the purposes of the Sheltzer Lab.

Dependencies

• mumdex/smash
  • build-essential
  • libx11-dev
  • libgsl-dev
  • ghostscript
• MISC
  • R
    • BiocManager
    • DNAcopy

Process

The general process for a single SMASH run starting from this repository is as follows:

1. Unpack peter/bins-n-ref.tar.gz to a central location
   • This contains two subdirectories, bins/ and ref/, both of which will be used later
2. Change directory into peter/ and build smash, extract_cn_segments, and mummer
3. Build the genome reference index via mummer
   • Run mummer [-rcref] [-verbose] /absolute/path/to/reference.fa
   • -rcref is "high memory" index, ~120GB for the reference unpacked above
   • Not using -rcref is "low memory" index, ~32GB for the reference unpacked above
   • This process will build a subdirectory /absolute/path/to/reference.fa.bin/ containing the associated index files
4. The easiest way to run SMASH is to use one of the pre-built BASH scripts
   • peter/cn/smash.sh is the "standard" run
     • Usage: smash.sh /path/to/peter/ /path/to/reference.fa /path/to/bins/ bins,to,use ID001 /path/to/ID001/R1.fq.gz /path/to/ID001/R2.fq.gz [num_threads]
   • peter/cn/smash_marvel.sh adds additional steps
     • Usage: smash_marvel.sh /path/to/peter/ /path/to/reference.fa /path/to/bins/ bins,to,use ID001 /path/to/ID001/R1.fq.gz /path/to/ID001/R2.fq.gz [num_threads]

Benchmarking

SMASH has diminishing returns on thread usage as most time is spent in IO across the memory-mapped reference.

On a benchmarked example, the following results were recieved via hyperfine where only the [num_threads] parameter was changed.

parameter_num_threads	mean	stddev	median	user	system	min	max
1	14918.95017	646.3076615	14733.67796	1251.462838	1656.75334	14177.56567	16035.03518
2	7722.782798	210.7823508	7732.975703	1202.794143	1648.677958	7467.368446	8102.611258
4	4914.293215	144.497086	4856.840803	1201.658335	1691.165742	4755.719154	5205.411549
6	4221.590413	90.91750457	4187.44144	1250.378439	1801.309565	4145.375666	4462.437155
8	3991.534113	36.32553087	3980.843808	1294.940436	1841.52768	3954.125145	4057.709754
12	3957.081538	177.6688093	3881.406046	1382.837122	1713.667732	3823.216936	4370.94731
16	3817.09813	23.53336835	3818.519433	1366.036457	1693.253564	3771.39473	3852.890088

The baseline command used was:

/home/ryan/mumdex/cn/smash.sh \
  /home/ryan/mumdex/ \
  /home/ryan/ref/chrAll-peter.fa \
  /home/ryan/bins/ \
  20000,50000 \
  A20581 \
  /home/ryan/smash/testing/data/Sample_A20581qC20PXT1M1/A20581qC20PXT1M1_S7_L001_R1_001.fastq.gz \
  /home/ryan/smash/testing/data/Sample_A20581qC20PXT1M1/A20581qC20PXT1M1_S7_L001_R2_001.fastq.gz \
  {num_threads}

From the above, we can see that despite 16 threads being 2.02 times faster than 2 threads this means ~4 times the total resource usage per unit of work. For example, if we assume a 2 thread run takes 4 hours to complete we would expend 2CPU x 4h or 8 CPU-hours; the same run on 16 threads would complete in roughly half the time or 2 hours however we would expend 16CPU x 2h or 32 CPU-hours. The optimal threading is 4 threads.

Assorted Notes on Usage

• smash.sh and smash_marvel.sh wrap the R1 and R2 files in a call to zcat before passing them to smash so both files must be GZipped
• If running SMASH on a cluster, be sure to compile against the oldest nodes to ensure functional ASM on any node, at YSM this is done via srun -C oldest
• It is highly encouraged to use mummer's -rcref if possible as it will greatly speed up later steps -- if desired, both mummer with and without -rcref can be run

Plans

There are current plans to create a Nextflow script which will take any number of SMASH runs and automatically run them locally or via a cluster from a single workflow definition. This script will use the 4 thread optimum and 32GB "low memory" reference -- working under the assumption that efficient resource usage is more important than walltime.

Alternatives

NYGC has prepared a Docker image of SMASH prior to any edits made by Ryan A. Hagenson for application in the Sheltzer Lab. This is the same baseline Ryan A. Hagenson began his edits from and should match the original content under peter/.