Biological TFDS Datasets

Some datasets that are useful for machine learning in biology. We use TensorFlow Datasets to download and access the datasets.

Usage

Downloading a dataset

Just run the download_and_prepare method on the dataset builder. Note that some of the datasets are very big and this can take some time.

Here is an example for the UniRef50 dataset:

from bio_tfds.protein import uniref

ds = uniref.UniRef50(data_dir="<data_dir>")
ds.download_and_prepare(download_dir="<download_dir>")

If you are on longleaf, the datasets have already been downloaded and prepared in the /proj/craffel/datasets/tfds directory.

Accessing a dataset

In order to access a dataset that has already been prepared, use its as_dataset method.

Here is an example for the UniRef50 dataset on longleaf:

import itertools
from bio_tfds.protein import uniref

ds = uniref.UniRef50().as_dataset(split="train")

for x in itertools.islice(ds, 10):
    print(x)

This is very fast.