TensorFlow I/O

TensorFlow is back at Google I/O on May 14! Register now

View on TensorFlow.org

Run in Google Colab

View source on GitHub

Download notebook

Overview

This tutorial demonstrates the tfio.genome package that provides commonly used genomics IO functionality--namely reading several genomics file formats and also providing some common operations for preparing the data (for example--one hot encoding or parsing Phred quality into probabilities).

This package uses the Google Nucleus library to provide some of the core functionality.

Setup

try:
  %tensorflow_version 2.x
except Exception:
  pass
!pip install -q tensorflow-io

import tensorflow_io as tfio
import tensorflow as tf

FASTQ Data

FASTQ is a common genomics file format that stores both sequence information in addition to base quality information.

First, let's download a sample fastq file.

# Download some sample data:
curl -OL https://raw.githubusercontent.com/tensorflow/io/master/tests/test_genome/test.fastq

% Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100   407  100   407    0     0   2035      0 --:--:-- --:--:-- --:--:--  2035

Read FASTQ Data

Now, let's use tfio.genome.read_fastq to read this file (note a tf.data API coming soon).

fastq_data = tfio.genome.read_fastq(filename="test.fastq")
print(fastq_data.sequences)
print(fastq_data.raw_quality)

tf.Tensor(
[b'GATTACA'
 b'CGTTAGCGCAGGGGGCATCTTCACACTGGTGACAGGTAACCGCCGTAGTAAAGGTTCCGCCTTTCACT'
 b'CGGCTGGTCAGGCTGACATCGCCGCCGGCCTGCAGCGAGCCGCTGC' b'CGG'], shape=(4,), dtype=string)
tf.Tensor(
[b'BB>B@FA'
 b'AAAAABF@BBBDGGGG?FFGFGHBFBFBFABBBHGGGFHHCEFGGGGG?FGFFHEDG3EFGGGHEGHG'
 b'FAFAF;F/9;.:/;999B/9A.DFFF;-->.AAB/FC;9-@-=;=.' b'FAD'], shape=(4,), dtype=string)

As you see, the returned fastq_data has fastq_data.sequences which is a string tensor of all sequences in the fastq file (which can each be a different size) along with fastq_data.raw_quality which includes Phred encoded quality information about the quality of each base read in the sequence.

Quality

You can use a helper op to convert this quality information into probabilities if you are interested.

quality = tfio.genome.phred_sequences_to_probability(fastq_data.raw_quality)
print(quality.shape)
print(quality.row_lengths().numpy())
print(quality)

WARNING:tensorflow:From /tmpfs/src/tf_docs_env/lib/python3.6/site-packages/tensorflow/python/util/deprecation.py:574: calling map_fn_v2 (from tensorflow.python.ops.map_fn) with dtype is deprecated and will be removed in a future version.
Instructions for updating:
Use fn_output_signature instead
(4, None, 1)
[ 7 68 46  3]
<tf.RaggedTensor [[[0.0005011872854083776], [0.0005011872854083776], [0.0012589251855388284], [0.0005011872854083776], [0.0007943279924802482], [0.00019952621369156986], [0.0006309572490863502]], [[0.0006309572490863502], [0.0006309572490863502], [0.0006309572490863502], [0.0006309572490863502], [0.0006309572490863502], [0.0005011872854083776], [0.00019952621369156986], [0.0007943279924802482], [0.0005011872854083776], [0.0005011872854083776], [0.0005011872854083776], [0.0003162277571391314], [0.0001584893325343728], [0.0001584893325343728], [0.0001584893325343728], [0.0001584893325343728], [0.0010000000474974513], [0.00019952621369156986], [0.00019952621369156986], [0.0001584893325343728], [0.00019952621369156986], [0.0001584893325343728], [0.00012589251855388284], [0.0005011872854083776], [0.00019952621369156986], [0.0005011872854083776], [0.00019952621369156986], [0.0005011872854083776], [0.00019952621369156986], [0.0006309572490863502], [0.0005011872854083776], [0.0005011872854083776], [0.0005011872854083776], [0.00012589251855388284], [0.0001584893325343728], [0.0001584893325343728], [0.0001584893325343728], [0.00019952621369156986], [0.00012589251855388284], [0.00012589251855388284], [0.0003981070767622441], [0.0002511885541025549], [0.00019952621369156986], [0.0001584893325343728], [0.0001584893325343728], [0.0001584893325343728], [0.0001584893325343728], [0.0001584893325343728], [0.0010000000474974513], [0.00019952621369156986], [0.0001584893325343728], [0.00019952621369156986], [0.00019952621369156986], [0.00012589251855388284], [0.0002511885541025549], [0.0003162277571391314], [0.0001584893325343728], [0.015848929062485695], [0.0002511885541025549], [0.00019952621369156986], [0.0001584893325343728], [0.0001584893325343728], [0.0001584893325343728], [0.00012589251855388284], [0.0002511885541025549], [0.0001584893325343728], [0.00012589251855388284], [0.0001584893325343728]], [[0.00019952621369156986], [0.0006309572490863502], [0.00019952621369156986], [0.0006309572490863502], [0.00019952621369156986], [0.002511885715648532], [0.00019952621369156986], [0.03981072083115578], [0.003981071058660746], [0.002511885715648532], [0.050118714570999146], [0.003162277629598975], [0.03981072083115578], [0.002511885715648532], [0.003981071058660746], [0.003981071058660746], [0.003981071058660746], [0.0005011872854083776], [0.03981072083115578], [0.003981071058660746], [0.0006309572490863502], [0.050118714570999146], [0.0003162277571391314], [0.00019952621369156986], [0.00019952621369156986], [0.00019952621369156986], [0.002511885715648532], [0.06309572607278824], [0.06309572607278824], [0.0012589251855388284], [0.050118714570999146], [0.0006309572490863502], [0.0006309572490863502], [0.0005011872854083776], [0.03981072083115578], [0.00019952621369156986], [0.0003981070767622441], [0.002511885715648532], [0.003981071058660746], [0.06309572607278824], [0.0007943279924802482], [0.06309572607278824], [0.001584893325343728], [0.002511885715648532], [0.001584893325343728], [0.050118714570999146]], [[0.00019952621369156986], [0.0006309572490863502], [0.0003162277571391314]]]>

One hot encodings

You may also want to encode the genome sequence data (which consists of A T C G bases) using a one hot encoder. There's a built in operation that can help with this.

one_hot = tfio.genome.sequences_to_onehot(fastq_data.sequences)
print(one_hot)
print(one_hot.shape)

<tf.RaggedTensor [[[0, 0, 1, 0], [1, 0, 0, 0], [0, 0, 0, 1], [0, 0, 0, 1], [1, 0, 0, 0], [0, 1, 0, 0], [1, 0, 0, 0]], [[0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1], [0, 0, 0, 1], [1, 0, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0], [1, 0, 0, 0], [0, 0, 1, 0], [0, 0, 1, 0], [0, 0, 1, 0], [0, 0, 1, 0], [0, 0, 1, 0], [0, 1, 0, 0], [1, 0, 0, 0], [0, 0, 0, 1], [0, 1, 0, 0], [0, 0, 0, 1], [0, 0, 0, 1], [0, 1, 0, 0], [1, 0, 0, 0], [0, 1, 0, 0], [1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 0, 1], [0, 0, 1, 0], [0, 0, 1, 0], [0, 0, 0, 1], [0, 0, 1, 0], [1, 0, 0, 0], [0, 1, 0, 0], [1, 0, 0, 0], [0, 0, 1, 0], [0, 0, 1, 0], [0, 0, 0, 1], [1, 0, 0, 0], [1, 0, 0, 0], [0, 1, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1], [1, 0, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1], [1, 0, 0, 0], [1, 0, 0, 0], [1, 0, 0, 0], [0, 0, 1, 0], [0, 0, 1, 0], [0, 0, 0, 1], [0, 0, 0, 1], [0, 1, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 1, 0, 0], [0, 0, 0, 1], [0, 0, 0, 1], [0, 0, 0, 1], [0, 1, 0, 0], [1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 0, 1]], [[0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 0, 0, 1], [0, 0, 1, 0], [0, 0, 1, 0], [0, 0, 0, 1], [0, 1, 0, 0], [1, 0, 0, 0], [0, 0, 1, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 0, 0, 1], [0, 0, 1, 0], [1, 0, 0, 0], [0, 1, 0, 0], [1, 0, 0, 0], [0, 0, 0, 1], [0, 1, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 1, 0, 0], [0, 0, 0, 1], [0, 0, 1, 0], [0, 1, 0, 0], [1, 0, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 0, 1, 0], [1, 0, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 1, 0, 0], [0, 0, 0, 1], [0, 0, 1, 0], [0, 1, 0, 0]], [[0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 1, 0]]]>
(4, None, 4)

print(tfio.genome.sequences_to_onehot.__doc__)

Convert DNA sequences into a one hot nucleotide encoding.

    Each nucleotide in each sequence is mapped as follows:
    A -> [1, 0, 0, 0]
    C -> [0, 1, 0, 0]
    G -> [0 ,0 ,1, 0]
    T -> [0, 0, 0, 1]

    If for some reason a non (A, T, C, G) character exists in the string, it is
    currently mapped to a error one hot encoding [1, 1, 1, 1].

    Args:
        sequences: A tf.string tensor where each string represents a DNA sequence

    Returns:
        tf.RaggedTensor: The output sequences with nucleotides one hot encoded.