Internode

Very fast and accurate species tree inference despite ILS. Somewhat a successor to ASTRID-2 and a competitor to ASTER.

Quick Start

Installing the binary

A single binary called wastrid can be downloaded from our recommended release, currently only containing binaries for x86_64 Linux and arm64 macOS. You can put it in PATH or simply in the directory that you wish to run it from.

Preparing data

If you are familiar with ASTRAL/ASTER, you can skip this section because we almost use the same input format.

wastrid accepts the common newline delimited Newick format for the input gene trees, where each line contains a gene tree in Newick format. We packed some example data¹ containing 200 gene trees on 101 taxa that you can download to try out.

In addition, the gene trees can be annotated with branch support (the example data given above is supplied with bootstrap support from 0 to 100) for better accuracy² using wASTRID-s.

Inferring a Species Tree

Say, you have prepared this genes.tre file containing all your gene trees separated by newlines. Now we have two differing scenarios:

1. The gene trees have been annotated by support values that you trust (e.g., bootstrap support, aBayes support). Moreover, you know the upper-bound and lower-bound of these support. Continuing using our example data mentioned above, the lower bound is 0 and the upper bound is 100.
2. In the other scenario, the gene trees don't have support you trust.

In the first scenario, the command goes something like this:

# `-b 0-100` means that the support values are in the range of [0, 100]
wastrid -i genes.tre -b 0-100 -o output_stree.tre

and output_stree.tre is the output species tree path; -b specifies our bounds for the support (lower bound 0, upper bound 100). As another example, for aBayes support that can have lower bound 0.333 and upper bound 1, the bounds can be specified as -b 0.333-1 instead of what we have here.

In the second scenario (you don't have support/you don't trust the accuracy of such support), you really just want to run ASTRID-3 (not weighted ASTRID), ASTID-2 but faster, in which case just do

wastrid -i genes.tre --preset vanilla -o output_stree.tre

where preset can preconfigure flags for you. Other presets include:

• --preset abayes, equivalent to -m support -b 0.333-1
• --preset hundred-bootstrap, equivalent to -m support -b 0-100

After running the appropriate command, the output species tree topology is at output_stree.tre. Note that the branch lengths of the species tree are not biologically meaningful.

Examples

ASTRID-3

Emulation of ASTRID: output is printed to STDOUT

wastrid -i gtrees.tre --preset vanilla

Weighted ASTRID by support on IQTree aBayes support: output is written to a file specified by the path species.tre ([] denotes optional arguments)

wastrid -i gtrees.tre -b 0.333-1 [-m support] -o species.tre

Compilation

See also prebuilt binaries (located in Releases).

internode is developed with Rust, so compiling it from scratch requires a proper installation of the Rust toolchain. However, the FastME bindings used internally make the compiling process a bit harder than simply cargo build.

The FastME bindings used internally were generated via bindgen, which requires libclang. After a proper installation of libclang for bindgen, running the usual pipeline works for building the wastrid binary:

cargo build --release

For maximum speed that reduces cross-machine compatibility, the usual tricks apply:

RUSTFLAGS="-C target-cpu=native" cargo build --release

Notes

• This implementation of ASTRID is faster than the original implementation (of ASTRID-2). That is, wastrid --preset vanilla is speed-wise a better ASTRID.
• Missing data imputation is implemented (and automatically turned on), but alpha quality, using the original procedure of ASTRID.
• ASTRID-multi (see also DISCO) is still not implemented

Acknowledgments

The code contains translated parts from ASTRID-2 and TreeSwift. Due to ASTRID-2's GPLv2 and TreeSwift's GPLv3 license, this project (except in third_party, see below) is licensed as GPLv3.

FastME (residing in third_party) is copied from its original source code. We use it in our repository under their CeCILL-C license. FastME does not fall under the GPLv3 license of this project.

internode is a derivation of ASTRID-2, which is a variant of NJst, developed by Liang Liu and Scott Edwards. See these papers for a start.

Portions of this code were developed with the assistance of AI tools, such as GitHub Copilot. This is as of 2023 a common practice to ensure efficient and effective software development.

Footnotes

1. From ASTRAL-III S100, generated by Zhang and Mirarab. ↩

2. Note that usually the more accurate the support is, the more accurate the final result. FastTree default support might even degrade the accuracy a bit. ↩