Introduction

Based on this suggestion by Remco, we can correctly account for constant sites in a BEAST2 analysis by adding the following data tag below your current data tag:

<data id="xyz" spec="FilteredAlignment" filter="-" data="@xyzOriginal" constantSiteWeights="100 200 300 400"/>

This assumes that your original <data> tag had id=xyz and was renamed to id=xyzOriginal, and that you have 1000 constant sites that were removed from the alignment, with:

• 100 As

• 200 Cs

• 300 Gs

• 400 Ts

What does this do?

This script will take a FASTA file with a single DNA sequence (e.g., a bacterial chromosome), a VCF file containing the position of SNPs along the FASTA file (e.g., as outputted from snippy-core) and the XML file produced by BEAUTi containing only variable sites. It will output a new XML file named <original_xml_name>_plus_const.xml with the added information to account for constant sites. There is nothing else you need to do but run BEAST2.

It will optionally also take a BED file with positions to mask (e.g., positions of phage).

How to install?

pip3 install b2constsites

How to run it?

If installed correctly, a scripted called run_b2cs should be available in your path:

run_b2cs --help

At a minimum, you need to supply a sequence file with your reference (assuming it has a single chromosome entry — this was designed for bacterial genomics, but may work with viral too), a VCF file with variants, and the XML output from BEAUTi.

run_b2cs myref.fasta myvar.vcf myxml.xml

A new file called myxml_plus_const.xml will be created in the same folder as myxml.xml.

ASSUMPTIONS and CAVEATS

This script will:

• Only take in to account SNPs and MNPs annotated in the VCF. Other variant types will be ignored.

• Will only take into consideration A, C, G, and T bases in your reference sequence. All other characters will be ignored.

• Has not been tested with BEAST1.8, and as far as I know it will not work with that version of BEAST. This was designed for use with BEAST2.

The output will be, therefore, an approximation. However, it should be a close enough approximation that it will provide a better inference from BEAST2 than if one uses only variable sites, and then corrects in some post hoc manner.

Authors

Maintainer

Anders Gonçalves da Silva

Issues or Questions

GitHub Issues