fastmixture is a new software for estimating ancestry proportions in unrelated individuals. It is significantly faster than previous model-based software while providing accurate and robust ancestry estimates.
Please cite our paper in Peer Community Journal.
Preprint also available on BioRxiv.
To run the fastmixture software, you have a few options depending on your environment and preference:
-
Installing fastmixture via PyPI or Source Code
# Option 1: Build and install via PyPI pip install fastmixture # Option 2: Download source and install via pip git clone https://github.com/Rosemeis/fastmixture.git cd fastmixture pip install . # Option 3: Download source and install in a new Conda environment git clone https://github.com/Rosemeis/fastmixture.git conda env create -f fastmixture/environment.yml conda activate fastmixture
You can now run the program with the
fastmixturecommand. For more details on running it, see the Usage section. -
Using the fastmixture docker image with Docker or Apptainer
If you prefer or need to use a containerized setup (especially useful in HPC environments), a pre-built fastmixture container image is available on Docker Hub.
A. Using Docker
- Pull the image from Docker Hub
# Docker command docker pull albarema/fastmixture- Run the
fastmixturecontainer
# Mount the directory containing the PLINK files using --volume flag (e.g. `pwd`/project-data/) # Indicate the cpus available for the container to run # e.g. data prefix is 'toy.data' and output prefix is 'toy.fast' docker run --cpus=8 -v `pwd`/project-data/:/data/ albarema/fastmixture fastmixture --bfile data/toy.data --K 3 --out data/toy.fast --threads 8
B. Using Apptainer (formerly Singularity)
For Apptainer/Singularity users, please take a look at your HPC system's documentation for guidance. Apptainer will create the .sif image in your current working directory (pwd) by default. You will later use this image to run the software. If needed, specify a different directory and filename to store the image. Remember to bind the directories where the data is stored (
--bind).- Pull
fastmixturecontainer image into a .sif file that Apptainer can use
# Singularity/Apptainer apptainer pull <fastmixture.sif> docker://albarema/fastmixture
- Run
fastmixturecontainer
# Singularity/Apptainer apptainer run <fastmixture.sif> fastmixture --bfile data/toy.data --K 3 --out data/toy.fast --threads 8
If you run into issues with your installation on a HPC system, it could be due to a mismatch of CPU architectures between login and compute nodes (illegal instruction). You can try and remove every instance of the march=native compiler flag in the setup.py file which optimizes fastmixture to your specific hardware setup. Another alternative is to use the uv package manager, where you can run fastmixture in a temporary and isolated environment by simply adding uvx in front of the fastmixture command.
# uv tool run example
uvx fastmixture --bfile data/toy.data --K 3 --out data/toy.fast --threads 8fastmixture requires input data in binary PLINK format.
- Choose the value of
Kthat best fits your data. We recommend performing principal component analysis (PCA) first as an exploratory analysis before runningfastmixture. - Use multiple seeds for your analysis to ensure robust and reliable results (e.g. ≥ 3).
# Using binary PLINK files for K=3
fastmixture --bfile data --K 3 --threads 32 --seed 1 --out test
# Outputs Q and P files (test.K3.s1.Q and test.K3.s1.P)A supervised mode is available in fastmixture using --supervised. Provide a file of population assignments for individuals as integers in a single column file. Unknown or admixed individuals must be given a value of '0'.
# Using binary PLINK files for K=3
fastmixture --bfile data --K 3 --threads 32 --seed 1 --out super.test --supervised data.labels
# Outputs Q and P files (super.test.K3.s1.Q and super.test.K3.s1.P)A projection mode is available in fastmixture using --projection. Provide a file of pre-computed ancestral allele frequencies (P-file) from a previous fastmixture run based on a reference dataset. Only ancestry proportions (Q-file) are estimated in a new dataset. SNPs must be strictly overlapping between the datasets.
# fastmixture in reference dataset
fastmixture --bfile ref --K 3 --threads 32 --seed 1 --out ref.test
# fastmixture using projection mode in new dataset
fastmixture --bfile new --K 3 --threads 32 --seed 1 --out new.test --projection ref.test.K3.s1.P
# Outputs Q file (new.test.K3.s1.Q)--iter, specify maximum number of iterations for EM algorithm (1000)--tole, specify tolerance for convergence in EM algorithm (0.5)--batches, specify number of initial mini-batches (32)--check, specify number of iterations performed before convergence check (5)--power, spe 709E cify number of power iterations in SVD (11)--chunk, number of SNPs to process at a time in randomized SVD (8192)--als-iter, specify maximum number of iterations in ALS procedure (1000)--als-tole, specify tolerance for convergence in ALS procedure (1e-4)--no-freqs, do not save ancestral allele frequencies (P-file)--random-init, random parameter initialization instead of SVD
This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details
- Jonas Meisner, Novo Nordisk Foundation Center for Basic Metabolic Research, University of Copenhagen
- Cindy Santander, Computational and RNA Biology, University of Copenhagen
- Alba Refoyo Martinez, Center for Health Data Science, University of Copenhagen