Coalescence-aware Alignment-based Species Tree EstimatoR (CASTER)

June 25, 2026 · View on GitHub

Genome-wide data have the promise of dramatically improving phylogenetic inferences. Yet, inferring the true phylogeny remains a challenge, mainly because the evolutionary histories of different genomic regions differ. The traditional concatenation approach ignores such differences, resulting in both theoretical and empirical shortcomings. In response, many discordance-aware inference methods have been developed. Yet, all have their own weaknesses. Many methods rely on short recombination-free genomic segments to build gene trees and thus suffer from a lack of signals for gene tree reconstruction, resulting in poor species tree. Some methods wrongly assume that the rate of evolution is uniform across the species tree. Yet, others lack enough scalability to analyze phylogenomic data.

We introduce a new site-based species tree inference method that seeks to address these challenges without reconstructing gene trees. Our method, called CASTER (Coalescence-aware Alignment-based Species Tree EstimatoR), has two flavors: CASTER-site and CASTER-pair. The first is based on patterns in individual sites and the second is based on pairs of sites.

CASTER has several outstanding features:

CASTER introduces two new optimization objectives based on genomic site patterns of four species; we show that optimizing these objectives produces two estimators: CASTER-site is statistically consistent under MSC+F84 model while allowing mutation rate to change across sites and across species tree branches; CASTER-pair is statistically consistent under MSC+LM model under further assumptions.
CASTER comes with a scalable algorithm to optimize the objectives summed over all species quartets. Remarkably, its time complexity is linear to the number of sites and at most quasi-quadratic with respect to the number of species.
CASTER can handle multiple samples per species, and CASTER-site specifically can work with allele frequencies of unphased multiploid.
CASTER is extremenly memory efficent, requiring <1 byte per SNP per sample

Under extensive simulation of genome-wide data, including recombination, we show that both CASTER-site and CASTER-pair out-perform concatenation using RAxML-ng, as well as discordance-aware methods SVDQuartets and wASTRAL in terms of both accuracy and running time. Noticeably, CASTER-site is 60–150X faster than the alternative methods. It reconstructs an Avian tree of 51 species from aligned genomes with 254 million SNPs in only 3.5 hours on an 8-core desktop machine with 32 GB memory. It can also reconstruct a species tree of 201 species with approximately 2 billion SNPs using a server of 256 GB memory.

Our results suggest that CASTER-site and CASTER-pair can fulfill the need for large-scale phylogenomic inferences.

Publication

Chao Zhang, Rasmus Nielsen, Siavash Mirarab, CASTER: Direct species tree inference from whole-genome alignments. Science (2025) https://www.science.org/doi/10.1126/science.adk9688

Notice

Since CASTER-site and CASTER-pair assume different models, please run both and choose the result that makes more sense if you can.

CASTER-site sliding-window analysis uses anthor stand-alone tool called dstar.

Announcements

Integrated in Phylosuite (NEW)

Many ASTER tools have been integrated in PhyloSuite, an integrated and scalable desktop platform for streamlined molecular sequence data management and evolutionary phylogenetics studies.

GUI for Windows users

Please check out our software with GUI. Simply download the zip file, extract the contents, enter exe folder, and click aster-gui.exe.

Bug Reports

Contact chaozhang@pku.edu.cn, aster-users@googlegroups.com, or post on ASTER issues page.

Documentations

The rest of this TUTORIAL file
Forums (feel free to ask questions or ask for help running ASTER):
- User group discussions
- ASTER issues page
- QQ group: 130635706

INSTALLATION

For most users, installing ASTER is very easy! Install using one of three approaches:

You simply need to download the zip file for Windows/MacOS ARM/MacOS X86/Linux and extract the contents to a folder of your choice.
Alternatively, you can clone the github repository and checkout the branch named Windows/MacOS/Linux.
For MacOS/Linux users, you can also install ASTER using conda:
```
conda install aster
```

Binary files should be in the exe folder for Windows or bin folder otherwise. If you are lucky, these may just work as is and you may not need to build at all.

For Linux/MacOS/WSL users

In terminal, cd into the downloaded directory and run make.

If you see *** Installation complete! *** then you are done!
If you see Command 'g++' not found then before rerunning make,
- Debian (Ubuntu) users try
```
sudo apt update
sudo apt install g++
```
- CentOS (RedHat) users try
```
sudo yum update
sudo yum install gcc-c++
```
- Unix (MacOS) users should be prompted for installing g++ and please click "install". If no prompt, try g++. Please ensure that Clang version is at least 14.
If you see "error" when running make, please try make caster-site instead and file a bug report.

Binary files should be in the bin folder.

For Windows users

Executables for x86-64 are available in exe folder and it is very likely that they already work.
Windows Subsystem for Linux (WSL) is HIGHLY recommanded if you need to install on your own! Please follow instructions in "For Linux/Unix/WSL users" section.
To compile windows excutables:
1. Download MinGW and install posix version for your architecture (eg. x86-64)
2. Add path to bin folder of MinGW to system environment variable PATH
3. Double click make.bat inside the downloaded directory

GUI for Windows users (NEW)

Please check out our software with GUI. Simply download the zip file, extract the contents, enter exe folder, and click aster-gui.exe.

STOP!

Please make sure you removed paralogous alignment regions using RepeatMasker or alike. This will improve the accuracy of CASTER on biological datasets!

A detailed walkthrough is also available (English/中文).

A walkthrough for sliding window analysis is also available.

INPUT

The input by default is a single MSA in Fasta format
The input can also be a text file containing a list of Fasta files (one file per line) if you add -f list to input arguments
The input can also be a single Phylip files or vertically concatenated Phylip files in one file by adding -f phylip to input arguments
The input files can have missing taxa and multiple individuals/copies per species.
When individuals/copies/genes from the same species are available, you need to let CASTER to know that they are from the same species. You can do this in two ways.
1. You can give multiple individuals/copies/genes from the same species the same name in the input alignment.
2. OR, a mapping file needs to be provided using the -a option.

individual_A1 species_name_A
individual_A2 species_name_A
individual_B1 species_name_B
individual_B2 species_name_B
individual_B3 species_name_B
...

gene_A1 species_name_A
gene_A2 species_name_A
gene_B1 species_name_B
gene_B2 species_name_B
gene_B3 species_name_B
...

Examples:

Single Fasta file:

>species_A
AAA
>species_C
CCC
>species_G
GGG
>species_T
TTT

Single Phylip file, multiple individuals with mapping file:

5 3
individual_A1 AAA
individual_A2 AAA
species_C CCC
species_G GGG
species_T TTT

Mapping file:

individual_A1 species_A
individual_A2 species_A

Multiple Fasta file:

gene1.fasta
gene2.fasta

In gene1.fasta:

>species_A
AAA
>species_C
CCC
>species_G
GGG
>species_T
TTT

In gene2.fasta (order can change, fasta files can have missing taxa):

>species_C
CC
>species_A
AA
>species_T
TT

Multiple Phylip file, multiploid without mapping file (if using CASTER-pair, genes must be phased; if using CASTER-site, you can arbitrarily phase them):

6 3
species_A AAA
species_A AAA
species_A AAA
species_A AAA
species_C CCC
species_C CCC
4 2
species_A AA
species_A AA
species_T TT
species_T TT

Notice: only CASTER-site works on unphased SNPs, you can translate VCF files into Fasta (or Phylip) in the following way.

VCF:

species_A
A/A/C/T
A/A/G/G

Fasta (order and phasing do not matter):

>species_A
AA
>species_A
AA
>species_A
CG
>species_A
TG

OUTPUT

The output in is Newick format and gives:

the species tree topology
branch supports measured as local bootstrap support (>95.0 means good)
It can also annotate branches with other quantities, such as quartet scores and local bootstraps for all three topologies.

EXECUTION

ASTER currently has no GUI. You need to run it through the command-line. In a terminal/PowerShell, go to the directory (location) where you have downloaded ASTER and issue the following command:

bin/caster-site

This will give you a list of options available. If you are using Windows, please replace bin/caster-site with .\exe\caster-site.exe.

To find the species tree with input from in a file called INPUT_FILE, use:

bin/caster-site INPUT_FILE

bin/caster-site -i INPUT_FILE

In the first case, INPUT_FILE is hard-coded to be the last argument for backward compatibility.

For example if you want to run caster-site with input example/genetrees.tre_1.fas, then run

bin/caster-site example/genetrees.tre_1.fas

bin/caster-site -i example/genetrees.tre_1.fas

The results will be outputted to the standard output. To save the results in a file use the -o OUTPUT_FILE option before INPUT_FILE(Strongly recommended):

bin/caster-site -o OUTPUT_FILE INPUT_FILE

bin/caster-site -i INPUT_FILE -o OUTPUT_FILE

With -i INPUT_FILE option, the order does not matter anymore. For brevity, from here on we will not demonstrate -i INPUT_FILE cases.

To save the logs (also recommended), run:

bin/caster-site -o OUTPUT_FILE INPUT_FILE 2>LOG_FILE

For example, you can run

bin/caster-site -o example/genetrees.tre_1.fas.stree example/genetrees.tre_1.fas 2>example/genetrees.tre_1.fas.log

ASTER supports multi-threading. To run program with 4 threads, add -t 4 before INPUT_FILE:

bin/caster-site -t 4 -o OUTPUT_FILE INPUT_FILE 2>LOG_FILE

ASTER has very good parrallel efficiency up to 64 cores when input data is large. In fact, it often experiences super-linear speedup with 16 cores or more. So feel free to use as many cores as you want.

ASTER also allows rooting at an given outgroup:

bin/caster-site --root YOUR_OUTGROUP INPUT_FILE

Advanced Options

ASTER algorithm first performs R (4 by default) rounds of search and then repeatedly performs S (4 by default) rounds of subsampling and exploration until no improvement found.

bin/caster-site -r R -s S -o OUTPUT_FILE INPUT_FILE 2>LOG_FILE

If you want to run with more rounds of placement for ensured optimality, then you can run with

bin/caster-site -r 16 -s 16 -o OUTPUT_FILE INPUT_FILE 2>LOG_FILE

or simply

bin/caster-site -R -o OUTPUT_FILE INPUT_FILE 2>LOG_FILE

If you want to place taxa on an existing fully resolved species tree, you can use -c SPECIES_TREE_IN_NEWICK_FORMAT before INPUT_FILE:

bin/caster-site -o OUTPUT_FILE -c SPECIES_TREE_IN_NEWICK_FORMAT INPUT_FILE

Specifically, you can score and annotate a fully resolved species tree containing all taxa with -c SPECIES_TREE_IN_NEWICK_FORMAT. If want to score a species tree or you want to place only one taxon onto the tree, you can use

bin/caster-site -r 1 -s 0 -o OUTPUT_FILE -c SPECIES_TREE_IN_NEWICK_FORMAT INPUT_FILE

or simply,

bin/caster-site -C -o OUTPUT_FILE -c SPECIES_TREE_IN_NEWICK_FORMAT INPUT_FILE

If you want to give hints by providing candidate species trees or trees similar to the species tree, you can use -g SPECIES_TREES_IN_NEWICK_FORMAT before INPUT_FILE:

bin/caster-site -o OUTPUT_FILE -g SPECIES_TREES_IN_NEWICK_FORMAT INPUT_FILE