pathogen-embed 2.2.0

Reduced dimension embeddings for pathogen sequences

pathogen-embed

Create reduced dimension embeddings for pathogen sequences

pathogen-embed is an open-source software to run reduced dimension embeddings (PCA, MDS, t-SNE, and UMAP) on viral populations. For more details, read Nanduri et al. and check out the corresponding GitHub repository.

• Change log
• Documentation
• Source Code
• Bug reports

Installation

With pip

pip install pathogen-embed

With Conda

conda install -c conda-forge -c bioconda pathogen-embed

Quickstart

The following commands show an example of how to apply the pathogen-embed tools to a small set of seasonal influenza A/H3N2 hemagglutinin (HA) sequences. To start, calculate the distance matrix between each pair of sequences in the dataset.

pathogen-distance \
  --alignment tests/data/h3n2_ha_alignment.fasta \
  --output distances.csv

(Optional) For faster distance calculations without indel support, use snp-dists. This command converts the default tab-delimited output of snp-dists into the comma-delimited format expected by pathogen-embed.

snp-dists -c -b tests/data/h3n2_ha_alignment.fasta > distances.csv

Create a t-SNE embedding from the distance matrix. Note that the perplexity is the number of nearest neighbors to consider in the embedding calculations, so this value has to be less than or equal to the total number of samples in the input (N=50, here).

pathogen-embed \
  --alignment tests/data/h3n2_ha_alignment.fasta \
  --distance-matrix distances.csv \
  --output-dataframe tsne.csv \
  --output-figure tsne.pdf \
  --output-pairwise-distance-figure tsne_pairwise_distances.pdf \
  t-sne \
    --perplexity 45.0

The following figure shows the resulting embedding.

The following figure shows the distribution of pairwise Euclidean distances by corresponding pairwise genetic distance. The equation in the figure title shows how genetic distances (x in the equation) scale to Euclidean distances (y) in the embedding.

Find clusters in the embedding.

pathogen-cluster \
  --embedding tsne.csv \
  --label-attribute tsne_label \
  --output-dataframe tsne_with_clusters.csv \
  --output-figure tsne_with_clusters.pdf

The following image shows the t-SNE embedding colored by clusters. Note that the underlying clustering algorithm, HDBSCAN, allows samples to not be assigned to any cluster if there isn't a reliable cluster to place them in. These unassigned samples receive a cluster label of "-1".

If you know the minimum genetic distance you want to require between clusters, you can use the equation from the pairwise distance figure above to determine the corresponding minimum Euclidean distance to pass to pathogen-cluster's --distance-threshold argument.

Example: Identify reassortment groups from multiple gene alignments

To identify potential reassortment groups for viruses with segmented genomes, you can calculate one distance matrix per gene and pass multiple distance matrices to pathogen-embed. Internally, pathogen-embed sums the given distances matrices into a single matrix to use for an embedding. The clusters in the resulting embedding represent genetic diversity in each gene individually and potential reassortment between genes. The following example shows how to apply this approach to alignments for seasonal influenza A/H3N2 HA and NA.

Calculate a separate distance matrix per gene alignment for HA and NA.

pathogen-distance \
  --alignment tests/data/h3n2_ha_alignment.fasta \
  --output ha_distances.csv

pathogen-distance \
  --alignment tests/data/h3n2_na_alignment.fasta \
  --output na_distances.csv

Create a t-SNE embedding using the HA/NA alignments and distance matrices. The t-SNE embedding gets initialized by a PCA embedding from the alignments.

pathogen-embed \
  --alignment tests/data/h3n2_ha_alignment.fasta tests/data/h3n2_na_alignment.fasta \
  --distance-matrix ha_distances.csv na_distances.csv \
  --output-dataframe tsne.csv \
  --output-figure tsne.pdf \
  --output-pairwise-distance-figure tsne_pairwise_distances.pdf \
  t-sne \
    --perplexity 45.0

Finally, find clusters in the embedding which represent the within and between diversity of the given HA and NA sequences.

pathogen-cluster \
  --embedding tsne.csv \
  --label-attribute tsne_label \
  --output-dataframe tsne_with_clusters.csv \
  --output-figure tsne_with_clusters.pdf

Compare the resulting embedding and clusters to the embedding above from only HA sequences, to get a sense of how including the NA sequences affects the results.

Build documentation

Build the Documentation:

make -C /docs html

Clean the docs.

make -C /docs clean

Releasing a new version on PyPI

1. Update CHANGES.md to reflect changes since the last release and set the correct version number for the new release at the top of the file.
2. Update the version number in setup.py.
3. Create a new GitHub release, setting the same version number as above as the release tag and, optionally, generating release notes automatically. When you select "Publish release", GitHub Actions will run the publish workflow which uses the pypi-publish action to create a new release on PyPI for you.
4. Create a pull request to bump the version of the corresponding Bioconda package.

Run tests during development

Clone this repository locally.

git clone https://github.com/blab/pathogen-embed.git
cd pathogen-embed

Install an editable version of the package into a custom Conda environment.

conda create -n pathogen-embed python=3.11
conda activate pathogen-embed
python3 -m pip install -e '.[dev]'

Run tests with cram.

cram --shell=/bin/bash tests