Genomic Arrays based on TileDB

GenomicArrays is a Python package for converting genomic data from BigWig format to TileDB arrays.

Installation

Install the package from PyPI

pip install genomicarrays

Quick Start

Build a GenomicArray

Building a GenomicArray generates 3 TileDB files in the specified output directory:

• feature_annotation: A TileDB file containing input feature intervals.
• sample_metadata: A TileDB file containing sample metadata, each BigWig file is considered a sample.
• A matrix TileDB file named by the layer_matrix_name parameter. This allows the package to store multiple different matrices, e.g. 'coverage', 'some_computed_statistic', for the same interval, and sample metadata attributes.

The organization is inspired by the SummarizedExperiment data structure. The TileDB matrix file is stored in a features X samples orientation.

To build a GenomicArray from a collection of BigWig files:

import numpy as np
import tempfile
import genomicarrays as garr

# Create a temporary directory, this is where the
# output files are created. Pick your location here.
tempdir = tempfile.mkdtemp()

# List BigWig paths
bw_dir = "your/biwig/dir"
files = os.listdir(bw_dir)
bw_files = [f"{bw_dir}/{f}" for f in files]

features = pd.DataFrame({
     "seqnames": ["chr1", "chr1"],
     "starts": [1000, 2000],
     "ends": [1500, 2500]
})

# Build GenomicArray
dataset = garr.build_genomicarray(
     files=bw_files,
     output_path=tempdir,
     features=features,
     # Specify a fasta file to extract sequences
     # for each region in features
     genome_fasta="path/to/genome.fasta",
     # agg function to summarize mutiple values
     # from bigwig within an input feature interval.
     feature_annotation_options=garr.FeatureAnnotationOptions(
        aggregate_function = np.nanmean
     ),
     # for parallel processing multiple bigwig files
     num_threads=4
)

Note

• The aggregate function is expected to return either a scalar value or a 1-dimensional NumPy ndarray. If the later, users need to specify the expected dimension of the return array. e.g.

        feature_annotation_options=garr.FeatureAnnotationOptions(
              aggregate_function = my_custom_func,
              expected_agg_function_length = 10,
         ),

• The build process stores missing intervals from a bigwig file as np.nan. The default is to choose an aggregate functions that works with np.nan.

Query a GenomicArrayDataset

Users have the option to reuse the dataset object retuned when building the arrays or by creating a GenomicArrayDataset object by initializing it to the path where the files were created.

# Create a GenomicArrayDataset object from the existing dataset
dataset = GenomicArrayDataset(dataset_path=tempdir)

# Query data for the first 10 regions across all samples
coverage_data = dataset[0:10, :]

print(expression_data.matrix)
print(expression_data.feature_annotation)

 ## output 1
 array([[1. , 0.5],
      [1. , 0.5],
      [1. , 0.5],
      [1. , 0.5],
      [1. , 0.5],
      [1. , 0.5],
      [1. , 0.5],
      [1. , 0.5],
      [1. , 0.5],
      [1. , 0.5],
      [1. , nan]], dtype=float32)

 ## output 2
 seqnames  starts  ends  genarr_feature_index
 0      chr1     300   315                     0
 1      chr1     320   335                     1
 2      chr1     340   355                     2
 3      chr1     360   375                     3
 4      chr1     380   395                     4
 5      chr1     400   415                     5
 6      chr1     420   435                     6
 7      chr1     440   455                     7
 8      chr1     460   475                     8
 9      chr1     480   495                     9
 10     chr1     500   515                    10

Note

This project has been set up using PyScaffold 4.6. For details and usage information on PyScaffold see https://pyscaffold.org/.