Alexei Drummond

Port sequencing error models from BEAST1 to BEAST2

13

Sequencing error models (and post-mortem damage models) are implemented in BEAST1. See http://mbe.oxfordjournals.org/content/26/2/245.full Users would like this ported to BEAST2.

https://github.com/CompEvol/beast2/blob/b7a1c6d9920ca2847207aa3c1c155a74393456d9/src/beast/evolution/operators/DeltaExchangeOperator.java#L39

This will allow the possibility of applying a ScaledDirichlet prior to a multi-dimensional relative rates parameter and then slicing out the components to the site models of each partition. https://github.com/CompEvol/beast2/blob/b7a1c6d9920ca2847207aa3c1c155a74393456d9/src/beast/evolution/sitemodel/SiteModel.java#L54

Adding a line like: `xml = xml.replaceAll("\\n\\s*\\n", "\n");` directly after `xml = findPlates(xml2)` should fix it.

Right now there is no (single-liner) way of checking if a Parameter is being estimated or not. A default method in Parameter like: ``` default boolean isEstimated() { // loop...

TreeDistribution should be abstract.

1

The TreeDistribution is a super class for all concrete TreeDistribution classes such as YuleModel and Coalescent. Instantiating TreeDistribution directly is an error, so it should be declared abstract.

StarBEAST has baked in prior on population sizes. There is no need for this.

1

SpeciesTreePrior currently takes a gammaParameter that is used to parameterize the prior on popSize parameters. It would be simple to make gammaParameter optional, and then the user would be free...

GTR should be able to be parameterized by a single 6-dimensional rate parameter

1

This would facilitate the easy application of a Dirichlet prior to the relative rates, which is the standard approach used by the majority of Bayesian phylogenetic inference packages and is...

Path sampling model comparison tutorial in BEAST2

2

Model comparison is a fundamental step in most Bayesian phylogenetic analyses. We need a full tutorial for this.

Implement the free-rate model of rates across sites

3

The developers may consider incorporating the free-rate model, which relaxes the assumption of Gamma rate heterogeneity without many more parameters (https://doi.org/10.1038/nmeth.4285). One can for example use Dirichlet prior for rate...