tascCODA

Tree-aggregated compositional analysis for high-throughput sequencing data

tascCODA extends the scCODA model (Büttner, Ostner et al., 2021) with a method to perform sparse, tree-aggregated modeling of high-throughput sequencing data. Hereby, tascCODA can infer credible changes on the features (i.e. cell types/ASVs/taxa) of a high-throughput sequencing dataset, as well as effects on subgroups of the set of features, which are defined by a tree structure, for example a cell lineage hierarchy or a taxonomic tree.

The statistical methodology and benchmarking performance are described here:

Ostner, J., Carcy, S., and Müller, C. L. (2021). tascCODA: Bayesian Tree-Aggregated Analysis of Compositional Amplicon and Single-Cell Data. Front. Genet. 12, 766405.

Code for reproducing the analysis from the paper is available here.

Installation

Running the package requires a working Python environment (>=3.8).

This package uses the tensorflow (>=2.8) and tensorflow-probability (>=0.16) packages. The GPU computation features of these packages have not been tested with tascCODA and are thus not recommended.

To install tascCODA via pip, call:

pip install tasccoda

To install tascCODA from source:

• Navigate to the directory that you want to install tascCODA in

• Clone the repository from Github:

  git clone https://github.com/bio-datascience/tascCODA

• Navigate to the root directory of tascCODA:

  cd tascCODA

• Install dependencies::

  pip install -r requirements.txt

• Install the package:

  python setup.py install

Usage

Import tascCODA in a Python session via:

import tasccoda

You can then import a dataset in the same way as scCODA (see here for an instruction on scCODA's data structure) Once you imported your dataset, add a toytree tree object, for example generated from a Newick string, as data.uns["phylo_tree"].

Then, initialize your analysis object, together with your formula and reference feature (see the scCODA documentation for explanations). To set the aggregation bias, pass "phi" as a key in the pen_args parameter

`model = tasccoda.tree_ana.CompositionalAnalysisTree(
data,
reference_cell_type="9",
formula="x_0",
pen_args={"phi":phi},
)
`

Then, run HMC sampling with dual-averaging step size adaptation by calling:

`result = model.sample_hmc_da()`

Finally, you can look at result.node_df to find credible effects of covariates on tree nodes or plot a tree with indicators for credible effects with result.draw_tree_effects()