Cultivate your MSA to get better trees

# Arbow: cultivate your multiple sequence aligment to get better trees

## Name

We named this tool arbow as that would be the phonetic pronounciation of the short, endearing, term for an arborist in Australia.

## What it does

The goal of arbow is to automate and simplify the production of trees from multiple sequence alignments. The tool has been developed in the context of viral phylogenomics.

NOTE SOME DEFAULTS HAVE CHANGED in v0.6.*: - iqtree 2 is assumed by default, with an executable called iqtree (these come with new options to set your iqtree environment: -iv or --iqtree-version to specify the version (either 1 or 2), -ip or --iqtree-path to specify the path to the executable (assumes \$PATH by default), and -ie or --iqtree-exec to define the executable (iqtree2 by default)) - no sites with missing data are filtered by default (use -mp or -mc to change this now) - constant sites are by default those that have 100% of a single base (use -x or -c to change this)

In the current version (0.6.*) it:

1. Reads an alignment in multiFASTA format
2. Calculates stats for each sequence in the alignment
3. Trims 5/3 prime UTR regions --- defaults set to SARS-COV-2 (Genbank accession: NC_045512.2)
4. Calculates stats per column in the alignment
5. Allows the user to set a threshold of tolerable missing data in a column, and removes all non-conforming columns from the alignment. This can be defined in two ways:
• using the -mp flag which defines a proportion of missing sites (e.g., 0.01 means at most 1% missing data)
• using the -mc flag which defines the count of missing sites (e.g., 2 means at most 2 missing sites)
• one can still use -mm option, but that has been deprecated in favour of the -mc option. -mm will be removed in version v0.8.0.
6. From the remaining columns, arbow finds all the constant columns according to two user defined criteria: allow missing data (i.e., a column with missing data can still count to towards constant sites if it meets other criteria), and a defined threshold of variation. This process one to smooth over any potential sequencing errors. This threshold can be defined in two ways:
• the frequency of the major allow is equal to or larger than a threshold (i.e., if the threshold is set to 0.99 and there are 100 samples, 99 of which are A and one is G, that column would be counted as a constant A) --- use the -x option.
• if the total count of minor alleles is equal to or smaller than a threshold (i.e., if the threshold is 2 and there are 100 samples, 98 of which are A, 1 is C and 1 is T, that column would be considered as a constant A) --- use the -c option.
7. It then filters out all the variable columns, and outputs the variable alignment as a multiFASTA alignment.
8. It runs IQTree with a few sensible presets --- it is now possible to run both IQTree1 and IQTree2.

Currently, in step 4 above, columns that have a single observed nucleotide (e.g., C) but still have missing data that were not filtered out in step 3 are counted towards the overall frequency of that base in the alignment. In other words, if a user specifies a maximum number of 20 missing bases, and a column with 5 missing bases but with A in all other samples, that column will count towards the overall frequency of A in the alignment (i.e., majority consensus imputation). This assumptions is less risky the larger the number of samples in the alignment.

For step 5, missing data (i.e., - and N) are all codes as N.

Tests are underway to figure out how these assumptions might affect the output.

## Dependencies

1. Python >=3.6
2. IQTree 1.6+ or 2.0.4+
3. BioPython
4. Pandas
5. NumPy

## Installation

### Brew

brew install iqtree
pip<3> install arbow


### Conda

conda install -c bioconda iqtree
pip<3> install arbow


## Running

1. Generate a mulitple sequence alignment with your favourite aligner (e.g., MAFTT). Output a multiFASTA file.
2. Run arbow <options> <aln.fa>
3. Open tree-YYYY-MM-DD_HHMMSS.treefile in your favourite tree viewer (e.g, FigTree)
4. Open tree-YYYY-MM-DD_HHMMSS_bb.treefile or tree-YYYY-MM-DD_HHMMSS_alrt.treefile for branches with ultra-fast bootstrap support or SH-aLRT support only, respectively.

## Data stream

When running arbow, by default a stream is output to the console (stdout).

Data about the each sequence in the alignment is prefixed with [SEQ], and is followed by:

1. Count of each base (A, C, G, T, and NN is any character other than ACGT)
2. Percent missing data
3. A status column that has 0, 1, 2, or 3 * depending on whether the percent missing data is <0.5, >=0.5 and <1.0, >=1.0 and <5.0, or >=5, respectively.

Data about each column in the alignment is prefixed with [ALN], and is followed by:

1. Position in the alignment
2. Count of each base (bases counted will depend on whether all IUPAC codes are allowed or not - see below in usage)

## Usage

Usage: arbow [OPTIONS] ALN

Options:
--version
-i, --all-iupac               Print count of all IUPAC code for column
stats?

-s, --no-stream               Stop streaming stats to console

-mm, --max-missing INTEGER    Remove sites with 'mm' missing sites or more
[default: 20]

-x, --major-allele-freq FLOAT  If major allele frequency is equal or larger
than consider the site constant.  [default:
0.99]

-o, --out-var-aln TEXT        Filename for alignment of variable sites.
[default: aln-2020-04-07-150443.aln]

-p, --prefix TEXT             Prefix to append to IQTree output files.
[default: tree-2020-04-07-150443]

-iv, --iqtree-version [1|2]   Version of IQTree to use.  [default: 2]

-ip, --iqtree-path PATH       Path to iqtree executable

-ie, --iqtree-exec TEXT       The IQTree executable  [default: iqtree2]

-t, --iqtree-threads INTEGER  Number of cores to run IQtree  [default: 4]

-m, --iqtree-models TEXT      Substitution models to test.  [default:
HKY,TIM2,GTR]

-f, --iqtree-freq TEXT        Base frequency models to test.  [default: F]
-r, --iqtree-rates TEXT       Rate category models to test.  [default: G,R]
-b, --iqtree-bb INTEGER       Maximum number of UltraFast Bootstrap
iterations to attempt.  [default: 1000]

-a, --iqtree-alrt INTEGER     Number of replicates to perform SH-aLRT.
[default: 1000]

---iqtree-cmax INTEGER        Maximum number of rate categories to test.
[default: 5]

-io, --iqtree-outgroup TEXT     ID(s) of samples to be used as outgroup in
IQTree. (e.g., single: sample_1, multiple:
'sample_2,sample_3'. This option is ignore
if --use-ref-as-outgroup is selected.

-r, --ref-id TEXT              Sequence ID of the reference  [default:
MN908947.3]

-u, --use-ref-as-outgroup       Use the reference sequence as the outgroup
in IQTree  [default: False]

--five-prime-end INTEGER       Last base of the 5' UTR region in 1-index in
the ref sequence  [default: 265]

--three-prime-start INTEGER    First base of the 3' UTR region in 1-index in
the ref sequence  [default: 29675]

--include-const                When outputting the clean alignment, leave
constant sites in the alignment. [default is
to remove]

-l, --log FILENAME             Log file to store output. Use '-' to log to
stdout  [default:
arbow-2020-05-05-153619.log]

--help                        Show this message and exit.


## Default behaviour explained

By default, arbow will trim the 5' and 3' UTR regions.

### Remove sites with any gaps in the alignment

Let us say that you wish to remove all sites in the alignment that have any missing data, and retain all complete columns:

arbow -x 1.0 -mc 0 <in.aln>


### Keep all sites in an alignment (i.e., skip any filtering)

Let us say that you wish to keep all sites in the alignment, and you have an alignment with 200 sequences:

arbow <in.aln>


### Keep constant sites in the clean alignment

arbow --include-const <in.aln>


### Get help

arbow <-h|--help>


### Get version

arbow --version


## Project details

Uploaded source
Uploaded py3