Methylation data tools. Primarily classes and functions to be built upon by other Python tools. Arrow based tables for efficiently storing and processing Bismark BAMs. Module for producing pile-up images of regions for CNNs.
Explore alignments.Alignment and classes.*
Script: jtm-write-alignment-data
Outputs two tables, one with locus level (nucleotide, individual CpG, etc.) information, and one with read level information, plus metadata.
To load the data in R:
library(arrow)
library(jsonlite)
locusTable <- read_parquet("dataset/locus-table.parquet")
readTable <- read_parquet("dataset/read-table.parquet")
metadata <- fromJSON("dataset/metadata.json")
chrm_ids <- metadata['locus']['chrm_map']Most text data is encoded into integers. These mappings are recorded in the metadata.
2D pileups, as binary arrays with values between 0 & 1 representing different sequence features such as methylation state, mapping quality and nucleotide sequence.
Paired-end BAMs should be sorted by query name (preferably) or coordinate (may take more memory). Unsorted BAMs will probably work but use a lot of memory.
After installation use jtm-generate-images run --help for arguments. Available layers (that can then be passed
to the --layer option) can be printed using jtm-generate-images layers. run produces gzipped TAR files that
contain the binary array and a metadata JSON file, specifying the shape of the array and other things.
jtm-generate-images invokes the script generate_images.py.
import jtmethtools as jtm
fn = 'image.region_name.layer.tar.gz'
array, metadata = jtm.images.read_array(fn)
import matplotlib.pyplot as plt
plt.imshow(array, interpolation='nearest', cmap='gray')