anngtf 0.0.11

anngtf - lift annotations from a `.gtf` file to your AnnData object.

anngtf

Lift annotations from a gtf to your adata object.

Installation

To install via pip:

pip install anngtf

To install the development version:

git clone https://github.com/mvinyard/anngtf.git

cd anngtf; pip install -e .

Example usage

Parsing a .gtf file

import anngtf

gtf_filepath = "/path/to/ref/hg38/refdata-cellranger-arc-GRCh38-2020-A-2.0.0/genes/genes.gtf"

If this is your first time using anngtf, run:

gtf = anngtf.parse(path=gtf_filepath, genes=False, force=False, return_gtf=True)

Running this function will create two .csv files from the given .gtf files - one containing all feature types and one containing only genes. Both of these files are smaller than a .gtf and can be loaded into memory much faster using pandas.read_csv() (shortcut implemented in the next function). Additionally, this function leaves a paper trail for anngtf to find the newly-created .csv files again in the future such that one does not need to pass a path to the gtf.

In the scenario in which you've already run the above function, run:

gtf = anngtf.load() # no path necessary! 

Updating the adata.var table.

import anndata as a
import anngtf

adata = anndata.read_h5ad("/path/to/singlecell/data/adata.h5ad")
gtf = anngtf.load(genes=True)

anngtf.add(adata, gtf)

Since the anngtf distribution already knows where the .csv / .gtf files are, we could directly annotate adata without first specifcying gtf as a DataFrame, saving a step but I think it's more user-friendly to see what each one looks like, first.

Working advantage

Let's take a look at the time difference of loading a .gtf into memory as a pandas.DataFrame:

import anngtf
import gtfparse
import time

start = time.time()
gtf = gtfparse.read_gtf("/home/mvinyard/ref/hg38/refdata-cellranger-arc-GRCh38-2020-A-2.0.0/genes/genes.gtf")
stop = time.time()

print("baseline loading time: {:.2f}s".format(stop - start), end='\n\n')

start = time.time()
gtf = anngtf.load()
stop = time.time()

print("anngtf loading time: {:.2f}s".format(stop - start))

baseline loading time: 87.54s

anngtf loading time: 12.46s

~ 7x speed improvement.

• Note: This is not meant to criticize or comment on anything related to gtfparse - in fact, this library relies solely on gtfparse for the actual parsing of a .gtf file into memory as pandas.DataFrame.