GTFast

Installation

To install via pip:

pip install gtfast

To install the development version:

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

cd gtfast; pip install -e .

Example usage

Parsing a .gtf file

import gtfast

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 gtfast, run:

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

Interfacing with AnnData and updating an adata.var table.

If you're workign with single-cell data, you can easily lift annotations from a gtf to your adata object.

from anndata import read_h5ad
import gtfast

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

gtfast.add(adata, gtf)

Since the gtfast 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 gtfast
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 = gtfast.load()
stop = time.time()

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

baseline loading time: 87.54s

GTFast 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 and it's an amazing tool for python developers!

Contact

If you have suggestions, questions, or comments, please reach out to Michael Vinyard via email