Skip to content

Latest commit

 

History

History
80 lines (60 loc) · 3.09 KB

README.md

File metadata and controls

80 lines (60 loc) · 3.09 KB
Raw

mandrake

Build and run tests Anaconda package Documentation Status

Fast visualisation of the population structure of pathogens using Stochastic Cluster Embedding.

Paper:

Lees JA, Tonkin-Hill G, Yang Z, Corander J. Mandrake: visualizing microbial population structure by embedding millions of genomes into a low-dimensional representation. Philosophical Transactions of The Royal Society B. 2022;377: 20210237.

https://doi.org/10.1098/rstb.2021.0237

Documentation available at: https://mandrake.readthedocs.io/en/latest/

Installation (briefly)

See https://mandrake.readthedocs.io/en/latest/installation.html for more details.

  1. Install miniconda.
  2. Run conda create -n mandrake_env mandrake to install into a clean environment.
  3. Run conda activate mandrake_env to use the environment.

Refer to the conda-forge documentation if you want to install a CUDA (GPU) enabled version.

Semi-manual

You will need some dependencies, which you can install through conda:

conda create -n mandrake_env python
conda env update -n mandrake_env --file environment.yml
conda activate mandrake_env

You can then clone this repository, and run:

python setup.py install

GPU acceleration

You will need the CUDA toolkit installed.

If you have the ability to compile CUDA (e.g. nvcc) you should see a message:

CUDA found, compiling both GPU and CPU code

otherwise only the CPU version will be compiled:

CUDA not found, compiling CPU code only

Usage

After installing, an example command would look like this:

mandrake --sketches sketchlib.h5 --kNN 500 --cpus 4 --maxIter 1000000

This would use a file sketchlib.h5 created by pp-sketchlib to calculate accessory distances using 500 nearest neighbours.

Output can be found in numerous files prefixed mandrake.embedding*.

Other useful arguments include:

  • --alignment use a fasta alignment to calculate distances
  • --accessory use a presence/absence file (Rtab or similar) to calculate distances
  • --distances use a .npz file from a previous run and skip straight to the embedding step
  • --labels give labels to colour the output by
  • --perplexity change the perplexity of the preprocessing (similar to t-SNE)
  • --animate produce a video of the optimisation
  • --use-gpu use a GPU for the run. Make sure to increase --n-workers.

See the documentation for more details.