Running large and long command line jobs - using shmlast!

Running large and long command line jobs - using shmlast!

Our goal for this hands-on session is for you become more familiar with running longer programs on the command line. You’ll be introduced to shmlast, which implements an algorithm for discovering potential orthologs between an RNA-seq assembly and a protein database.

1. Start up a Jetstream instance

If you don’t already have one running then please start up an m1.medium instance running Ubuntu 16.04 on Jetstream by following these instructions from before.

2. Install shmlast

Copy and paste the following commands into your jetstream instance command line to updates the software list and install the base packages we will need for this session:

sudo apt-get -y update 
sudo apt-get install -y python3.5-dev python3.5-venv make libc6-dev g++ zlib1g-dev last-align parallel

Then create a Python environment with virtualenv, which will isolate your python packages:

python3.5 -m venv ~/py3
. ~/py3/bin/activate
pip install -U pip

And now install shmlast 1.2:

pip install shmlast>=1.2

This downloads and installs the latest version of shmlast.

3. Download and format sequence data

Next we need some data! Here we’re going to grab one of the three mouse RNA data sets,

curl -O ftp://ftp.ncbi.nih.gov/refseq/M_musculus/mRNA_Prot/mouse.1.rna.fna.gz

and all 8 of the cow protein data sets.

for i in 1 2 3 4 5 6 7 8
do
   curl -O ftp://ftp.ncbi.nih.gov/refseq/B_taurus/mRNA_Prot/cow.$i.protein.faa.gz
done

shmlast wants one query database (here, we’ll use mouse) and one database to be searched (here, cow) - but first we have to combine all of the databases into one:

gunzip -c cow.*.faa.gz > cow.faa

4. Run shmlast!

Now run shmlast:

shmlast crbl -q mouse.1.rna.fna.gz -d cow.faa --n_threads=6

this will take 16 minutes (!!) and produce some rather large files.

SideNote: What is shmlast doing? shmlast is going to compute putative orthologs between mouse transcripts and cow proteins. Orthologs are genes that duplicated from speciation (i.e. are the “same” gene in cow and mouse) and are presumed to have the same function, although that is a computational inference that needs to be treated with care.

As we’ll see later in the course, the computation of orthologs (or homologs more generally) is a core step in annotating genomes and transcriptomes. The reason is that you don’t automatically get gene assignments when you build a new genome or transcriptome - you just get unidentified DNA or RNA sequence! And then you have to name each transcript or gene, and generally people want that name to be the same across species. And that involves computing orthologs.

One of the most common ways to compute ortholog assignments is to use reciprocal-best-hit BLAST, in which you use BLAST to find the two sequences that match each other best in the database. However, reciprocal best hit has a few problems in the face of complicated evolutionary scenarios or deep RNA sequencing; from the supp. material of Aubry et al., 2014,

[... reciprocal best hit ] perform[s] well when sequences are present
as single copy genes in the datasets being compared and perform[s] less
well when trying to distinguish highly similar sequence groups (such
as multi-copy genes) from each other.  The issue of multiple-copy
genes and near identical gene-groups is particularly relevant for the
analysis of transcriptome data. It is to be expected that following de
novo assemblies of RNAseq data, most gene loci will be represented by
multiple assembled transcript variants.

Basically, this is saying that in many realistic scenarios (most especially multi-copy genes, but also multiple isoforms) reciprocal best hit is too conservative and will ignore real orthologs.

So Aubry et al. invent conditional reciprocal best hit, which tries to find close homolog groupings that can deal with multi-copy genes. shmlast is a reimplementation of that.

5. Why does shmlast take “so long”?

Well, we’re calculating all pairwise matches between 36,000 mouse transcripts and 64,000 cow proteins! So frankly it’s amazing it works so fast in the first place!


6. Looking at the output

Like most bioinformatics software, shmlast produces a lot of output. How do we explore the results?

The main output is mouse.1.rna.fna.gz.x.cow.faa.crbl.csv, which is a Comma-Separated Value file that you can load into any spreadsheet program. If we look at the file by typing head mouse.1.rna.fna.gz.x.cow.faa.crbl.csv we should see something like this:

E,EG2,E_scaled,ID,bitscore,q_aln_len,q_frame,q_len,q_name,q_start,q_strand,s_aln_len,s_len,s_name,s_start,s_strand,score
6.6e-24,9.8e-16,23.18045606445813,641897,109.65804469295703,89,1,390,"ref|NM_001013372.2| Mus musculus neural regeneration protein (Nrp), mRNA",64,+,89,389,ref|XP_005212262.1| PREDICTED: DNA oxidative demethylase ALKBH1 isoform X1 [Bos taurus],0,+,241.0
5.4e-194,4.4e-165,193.26760624017703,719314,605.7589445367834,313,0,331,"ref|NM_207235.1| Mus musculus olfactory receptor 358 (Olfr358), mRNA",0,+,313,313,ref|XP_607965.3| PREDICTED: olfactory receptor 1361 [Bos taurus],0,+,1365.0
2.8e-188,5e-160,187.5528419686578,423289,588.9868500580775,307,0,323,"ref|NM_146368.1| Mus musculus olfactory receptor 361 (Olfr361), mRNA",0,+,307,313,ref|XP_607965.3| PREDICTED: olfactory receptor 1361 [Bos taurus],0,+,1327.0
6.6e-183,5.6e-155,182.18045606445813,725159,572.2147555793716,307,0,318,"ref|NM_146622.1| Mus musculus olfactory receptor 360 (Olfr360), mRNA",0,+,307,313,ref|XP_607965.3| PREDICTED: olfactory receptor 1361 [Bos taurus],0,+,1289.0
5.4e-194,4.4e-165,193.26760624017703,719315,605.7589445367834,313,0,331,"ref|NM_207235.1| Mus musculus olfactory receptor 358 (Olfr358), mRNA",0,+,313,313,ref|XP_002691614.1| PREDICTED: olfactory receptor 1361 [Bos taurus],0,+,1365.0
2.8e-188,5e-160,187.5528419686578,423290,588.9868500580775,307,0,323,"ref|NM_146368.1| Mus musculus olfactory receptor 361 (Olfr361), mRNA",0,+,307,313,ref|XP_002691614.1| PREDICTED: olfactory receptor 1361 [Bos taurus],0,+,1327.0
6.6e-183,5.6e-155,182.18045606445813,725160,572.2147555793716,307,0,318,"ref|NM_146622.1| Mus musculus olfactory receptor 360 (Olfr360), mRNA",0,+,307,313,ref|XP_002691614.1| PREDICTED: olfactory receptor 1361 [Bos taurus],0,+,1289.0
4.8e-183,5.6e-155,182.3187587626244,373474,572.2147555793716,266,0,310,"ref|XR_001782298.1| PREDICTED: Mus musculus predicted gene 4786 (Gm4786), misc_RNA",29,+,266,266,ref|NP_001035610.1| 60S ribosomal protein L7a [Bos taurus],0,+,1289.0
3.2e-153,3.1e-138,152.4948500216801,643504,516.6020212552417,246,1,659,"ref|NR_003628.1| Mus musculus predicted gene 5766 (Gm5766), non-coding RNA",357,+,246,266,ref|NP_001035610.1| 60S ribosomal protein L7a [Bos taurus],0,+,1163.0

Note that you can scroll to the right within the text to see all the output.

Here the columns are helpfully labeled, but it’s still kind of a mess to look at - we’ll look at it in more detail in R, instead of using the command line. The key bits are the q_name_ and s_name column, which tell you which query and which subject sequences match each other.

How big is this file? Big! You can calculate how many lines are present by using the wc command, which will tell you how many lines, words, and characters are in the file:

  59130  1295041 16880227 mouse.1.rna.fna.gz.x.cow.faa.crbl.csv

(A precomputed version of this file is available for download here

Some points for discussion

  • What evolutionary scenarios are there that complicate orthology and homology assignment?

> Next…

Next we’re going to do some exploratory data analysis of this file using RStudio.