def _write_mtx(unidata: UnimodalData, output_dir: str, precision: int):
""" Write Unimodal data to mtx
"""
try:
from pegasusio.cylib.io import write_mtx
except ModuleNotFoundError:
print("No module named 'pegasusio.cylib.io'")
if not os.path.isdir(output_dir):
os.mkdir(output_dir)
for key in unidata.list_keys():
matrix = unidata.matrices[key]
mtx_file = os.path.join(output_dir, ("matrix" if key == "X" else key) + ".mtx.gz")
fifo_file = mtx_file + ".fifo"
if os.path.exists(fifo_file):
os.unlink(fifo_file)
os.mkfifo(fifo_file)
pobj = subprocess.Popen(f"gzip < {shlex.quote(fifo_file)} > {shlex.quote(mtx_file)}", shell = True)
write_mtx(fifo_file, matrix.data, matrix.indices, matrix.indptr, matrix.shape[0], matrix.shape[1], precision = precision) # matrix is cell x gene csr_matrix, will write as gene x cell
assert pobj.wait() == 0
os.unlink(fifo_file)
logger.info(f"{mtx_file} is written.")
unidata.barcode_metadata.to_csv(os.path.join(output_dir, "barcodes.tsv.gz"), sep = '\t')
logger.info("barcodes.tsv.gz is written.")
unidata.feature_metadata.to_csv(os.path.join(output_dir, "features.tsv.gz"), sep = '\t')
logger.info("features.tsv.gz is written.")
logger.info(f"Mtx for {unidata.get_uid()} is written.")
def deseq2(
pseudobulk: UnimodalData,
design: str,
contrast: Tuple[str, str, str],
de_key: str = "deseq2",
replaceOutliers: bool = True,
) -> None:
"""Perform Differential Expression (DE) Analysis using DESeq2 on pseduobulk data. This function calls R package DESeq2, requiring DESeq2 in R installed.
DE analysis will be performed on all pseudo-bulk matrices in pseudobulk.
Parameters
----------
pseudobulk: ``UnimodalData``
Pseudobulk data with rows for samples and columns for genes. If pseudobulk contains multiple matrices, DESeq2 will apply to all matrices.
design: ``str``
Design formula that will be passed to DESeq2
contrast: ``Tuple[str, str, str]``
A tuple of three elements passing to DESeq2: a factor in design formula, a level in the factor as numeritor of fold change, and a level as denominator of fold change.
de_key: ``str``, optional, default: ``"deseq2"``
Key name of DE analysis results stored. For cluster.X, stored key will be cluster.de_key
replaceOutliers: ``bool``, optional, default: ``True``
If execute DESeq2's replaceOutliers step. If set to ``False``, we will set minReplicatesForReplace=Inf in ``DESeq`` function and set cooksCutoff=False in ``results`` function.
Returns
-------
``None``
Update ``pseudobulk.varm``:
``pseudobulk.varm[de_key]``: DE analysis result for pseudo-bulk count matrix.
``pseudobulk.varm[cluster.de_key]``: DE results for cluster-specific pseudo-bulk count matrices.
Examples
--------
>>> pg.deseq2(pseudobulk, '~gender', ('gender', 'female', 'male'))
"""
try:
import rpy2.robjects as ro
from rpy2.robjects import pandas2ri, numpy2ri, Formula
from rpy2.robjects.packages import importr
from rpy2.robjects.conversion import localconverter
except ModuleNotFoundError as e:
import sys
logger.error(f"{e}\nNeed rpy2! Try 'pip install rpy2'.")
sys.exit(-1)
try:
deseq2 = importr('DESeq2')
except ModuleNotFoundError:
import sys
text = """Please install DESeq2 in order to run this function.\n
To install this package, start R and enter:\n
if (!require("BiocManager", quietly = TRUE))
install.packages("BiocManager")
BiocManager::install("DESeq2")"""
logger.error(text)
sys.exit(-1)
import math
to_dataframe = ro.r('function(x) data.frame(x)')
for mat_key in pseudobulk.list_keys():
with localconverter(ro.default_converter + numpy2ri.converter +
pandas2ri.converter):
dds = deseq2.DESeqDataSetFromMatrix(
countData=pseudobulk.get_matrix(mat_key).T,
colData=pseudobulk.obs,
design=Formula(design))
if replaceOutliers:
dds = deseq2.DESeq(dds)
res = deseq2.results(dds, contrast=ro.StrVector(contrast))
else:
dds = deseq2.DESeq(dds, minReplicatesForReplace=math.inf)
res = deseq2.results(dds,
contrast=ro.StrVector(contrast),
cooksCutoff=False)
with localconverter(ro.default_converter + pandas2ri.converter):
res_df = ro.conversion.rpy2py(to_dataframe(res))
res_df.fillna(
{
'log2FoldChange': 0.0,
'lfcSE': 0.0,
'stat': 0.0,
'pvalue': 1.0,
'padj': 1.0
},
inplace=True)
de_res_key = de_key if mat_key.find(
'.') < 0 else f"{mat_key.partition('.')[0]}.{de_key}"
pseudobulk.varm[de_res_key] = res_df.to_records(index=False)