TaxUMAP

Visualize structure in large microbiome datasets. Implements a microbiome-research-focused extension of the Uniform Manifold Approximation and Projection (UMAP) by calculating microbiome sample-by-sample distances at different taxonomic aggregations of taxon abundances, and allowing for custom weighting of aggregates.

Installation

Notice: TaxUMAP will be made available on both PyPi and Bioconda for installation via pip and conda. But until then, please use pip install -e . as described below to install in developer mode.

git clone https://github.com/jsevo/taxumap.git
pip install -e .

Quickstart: Notebook Example

• An interactive Jupyter notebook file is provided in examples/taxumap_example.ipynb. This demonstrates TaxUMAP on an example dataset we provide. See below for more information.

• Additionally, two other notebooks are provided to cover both adjusting the TaxUMAP default agg_levels and weights parameters (examples/adjusting_taxumap_parameters.ipynb), as well as cleaning the taxonomy table (examples/cleaning_taxonomy_table.ipynb).

Data required

Two tables are required: the microbiota data and a taxonomy table.

The microbiota data file (e.g., examples/example_data/microbiota_table.csv) must have a column with sample indices labeled 'index_column'. The remaining columns are expected to be amplicon sequence variant (ASV) labels or operational taxonomic unit (OTU) labels, i.e., the lowest level of taxonomy:

The taxonomy table (e.g., examples/example_data/taxonomy.csv) is expected to resolve higher taxonomic groups for each ASV/OTU in the microbiota table. The index of the taxonomy table should be ASV/OTU labels, while the columns of the taxonomy table should be higher taxonomic categories (e.g., kingdom, phylum, etc.). The columns must be ordered from left to right in decreasing taxonomic hierarchy, e.g.:

In the above tables, the '' designates strings. Any UNKNOWN taxonomic levels (e.g., 'unknown species') should be set to np.nan or the string 'nan'. For more information on how to properly resolve unknown taxonomic levels for TaxUMAP, please see the notebook examples/cleaning_taxonomy_table.ipynb. Finally, the taxonomy table should be monophyletic.

Usage

Command line:

run_taxumap.py -t examples/example_data/taxonomy.csv -m examples/example_data/microbiota_table.csv

The embedding will be saved in the current working folder, or to a location with the -o path/to/folder/ flag. Additionally, for best results, the flag -n should be folllowed by the number of unique patients in your dataset (see Optional flag information below for more details).

Python:

from taxumap.taxumap_base import Taxumap

##### Initialize Taxumap object #####

# From file
tu = Taxumap(taxonomy='path/to/taxonomy.csv',
            microbiome_data='path/to/microbiota_table.csv')

##### Run the transformation and look at the results #####

# Transform the data (an inplace function)
tu.transform_self()

# Raw embedding dataframe
tu.df_embedding

# "Which taxon dominate each sample?" dataframe
tu.df_dominant_taxon

# Visualize the embedding (will save to present working directory as "taxumap_scatterplot.pdf")
tu.scatter(save=True)

# Save the embedding (will save to present working directory as "taxumap_embedding.csv" if no parameter passed)
tu.save_embedding('path/to/embedding.csv')

Flags for Command-line Interface (CLI)

Required

• -t or --taxonomy: filepath to your taxonomy.csv file
• -m or --microbiota: filepath to your microbiota_table.csv file

Optional, but recommended

• -a or --agg_levels: Which taxonomic levels to aggregate, in the form of a /-delimined string (e.g. Phylum/Family, Family, Phylum/Order/Genus). Defaults to Phylum/Family.
• -w or --weights: Weights to give to each taxonomic level defined in --agg_levels, in the form of a /-delimined string (e.g. 5/6, 0.5/2, 6/2/1, etc). Defaults to 1 for each specified aggregation level.
• -n or --neigh: Change the n_neighbors parameter passed to the UMAP algorithm. See documentation here.

Optional, change default behavior

• -o or --outdir: Where to save embedding. Defaults to the present working directory.
• -s or --save: Set to False to not save the embedding. Defaults to True.
• -b or --min_dist: Change the min_dist parameter passed to the UMAP algorithm. See documentation here.

Example notebook (with example data)

A dataset provided by Olin et al. can be used to try out features of TaxUMAP and how to format new data properly.

• Link to original publication
• Link to the dataset

Publication

Olin A, Henckel E, Chen Y, et al. Stereotypic Immune System Development in Newborn Children. Cell. 2018;174(5):1277-1292.e14. doi:10.1016/j.cell.2018.06.045

Dataset

Olin, Axel (2018), “Stereotypic Immune System Development in Newborn Children”, Mendeley Data, v1

For convenience, we are providing in the taxumap/examples/example_data directory a pre-cleaned version of this dataset, as allowed under the CC BY 4.0 license. An accompanying Jupyter Notebook is provided to demonstrated how the data was cleaned.

generated summary

This code defines a class called Taxumap, which is used for running the taxUMAP algorithm. The class constructor takes several arguments, including agg_levels, weights, microbiota_data, taxonomy, name, and random_state. These arguments are used to initialize attributes of the Taxumap object.

Some of the methods of the Taxumap class include transform_self(), scatter(), save_embedding(), and df_dominant_taxon(). These methods are used to perform the taxUMAP transformation, generate a scatter plot of the transformed data, save the embedding to a file, and get the dominant taxonomic group based on the maximum abundance in each sample, respectively.

Overall, this code provides a framework for running the taxUMAP algorithm and visualizing the results.

License

MIT