"""Supporting functions for the 'reference' command."""
from __future__ import annotations
import collections
import logging
from typing import TYPE_CHECKING, Optional, Union
import numpy as np
import pandas as pd
import pyfaidx
from skgenome import tabio, GenomicArray as GA
from . import core, fix, descriptives, params
from .cmdutil import read_cna
from .cnary import CopyNumArray as CNA
if TYPE_CHECKING:
from collections.abc import Iterator
from numpy import bool_, ndarray
[docs]
def do_reference_flat(
targets: str,
antitargets: Optional[str] = None,
fa_fname: None = None,
is_haploid_x_reference: bool = False,
diploid_parx_genome: Optional[str] = None,
) -> CNA:
"""Compile a neutral-coverage reference from the given intervals.
Combines the intervals, shifts chrX values if requested, and calculates GC
and RepeatMasker content from the genome FASTA sequence.
Parameters
----------
targets : str
Path to BED file with target intervals.
antitargets : str, optional
Path to BED file with antitarget intervals.
fa_fname : str, optional
Path to reference genome FASTA file for GC/RepeatMasker calculations.
is_haploid_x_reference : bool, optional
True if reference should use male (haploid X) chromosome coverage.
Default is False.
diploid_parx_genome : str, optional
Reference genome name for pseudo-autosomal region handling
(e.g., 'hg19', 'hg38', 'mm10').
Returns
-------
CopyNumArray
Neutral reference with flat (expected neutral) log2 coverage values
and calculated GC/RepeatMasker features.
"""
ref_probes = bed2probes(targets)
if antitargets:
ref_probes.add(bed2probes(antitargets))
# Set sex chromosomes by "reference" sex
ref_probes["log2"] = ref_probes.expect_flat_log2(
is_haploid_x_reference, diploid_parx_genome
)
ref_probes["depth"] = np.exp2(ref_probes["log2"]) # Shim
# Calculate GC and RepeatMasker content for each probe's genomic region
if fa_fname:
gc, rmask = get_fasta_stats(ref_probes, fa_fname)
ref_probes["gc"] = gc
ref_probes["rmask"] = rmask
else:
logging.info("No FASTA reference genome provided; skipping GC, RM calculations")
ref_probes.sort_columns()
return ref_probes
[docs]
def bed2probes(bed_fname: str) -> CNA:
"""Create a neutral-coverage CopyNumArray from a file of regions."""
regions = tabio.read_auto(bed_fname)
table = regions.data.loc[:, ("chromosome", "start", "end")]
table["gene"] = regions.data["gene"] if "gene" in regions.data else "-"
table["log2"] = 0.0
table["spread"] = 0.0
return CNA(table, {"sample_id": core.fbase(bed_fname)})
[docs]
def do_reference(
target_fnames: list[str],
antitarget_fnames: Optional[list[str]] = None,
fa_fname: Optional[str] = None,
is_haploid_x_reference: bool = False,
diploid_parx_genome: Optional[str] = None,
female_samples: Optional[bool] = None,
do_gc: bool = True,
do_edge: bool = True,
do_rmask: bool = True,
do_cluster: bool = False,
min_cluster_size: int = 4,
) -> CNA:
"""Compile a pooled coverage reference from normal samples.
Creates a reference profile by combining coverage data from multiple normal
(control) samples. The reference represents the expected neutral copy number
baseline and is used to normalize tumor/test sample coverage for CNV detection.
The reference includes per-bin coverage values, GC content, RepeatMasker
information, and other features used for bias correction. Sample sex is either
provided or automatically inferred from X/Y chromosome coverage.
Parameters
----------
target_fnames : list of str
Paths to target coverage files (.targetcoverage.cnn or .cnn) from normal
samples. These are typically generated by the `coverage` command on
on-target/baited regions.
antitarget_fnames : list of str, optional
Paths to antitarget coverage files (.antitargetcoverage.cnn) from the
same normal samples. Must match target_fnames in order and count.
If None, only targets are used (e.g., for amplicon sequencing).
Default: None
fa_fname : str, optional
Path to reference genome FASTA file. Required for GC content and
RepeatMasker calculations. If None, GC and RM corrections are skipped.
Default: None
is_haploid_x_reference : bool, optional
True if the reference samples are male (haploid X chromosome).
Affects expected X chromosome coverage normalization.
Default: False (assumes female reference or mixed)
diploid_parx_genome : str, optional
Reference genome name for pseudo-autosomal region handling
(e.g., 'hg19', 'hg38', 'mm10'). PAR regions on X/Y are treated as
diploid even in male samples.
Default: None
female_samples : bool, optional
Explicit sex designation for all samples. If None, sex is inferred
from X/Y chromosome coverage patterns automatically.
True = all female, False = all male.
Default: None (auto-infer)
do_gc : bool, optional
Calculate and include GC content for each bin from FASTA.
Requires fa_fname. Used for GC bias correction.
Default: True
do_edge : bool, optional
Calculate and include "edge effect" metric (bin distance from
interval edges). Used for edge bias correction.
Default: True
do_rmask : bool, optional
Calculate and include RepeatMasker content for each bin from FASTA.
Requires fa_fname. Used for repeat region bias correction.
Default: True
do_cluster : bool, optional
Apply hierarchical clustering to identify and exclude outlier samples.
Helps create a more robust reference when sample quality varies.
Default: False
min_cluster_size : int, optional
Minimum cluster size when do_cluster=True. Smaller clusters are
considered outliers and excluded.
Default: 4
Returns
-------
CopyNumArray
Pooled reference with combined coverage values, GC content, RepeatMasker
content, and other features. This .cnn file is used as input to the
`fix` command for normalizing tumor samples.
Notes
-----
The reference creation process:
1. Sex inference (if not provided): Examines X/Y chromosome coverage
ratios in each sample to determine chromosomal sex.
2. Coverage pooling: Combines coverage across samples using robust
statistics (typically median).
3. Bias feature calculation: Extracts GC%, RepeatMasker%, edge effects
from the reference genome FASTA.
4. Optional clustering: Identifies and removes outlier samples.
The resulting reference should be reused for all samples prepared with
the same capture kit, sequencing platform, and protocol.
Warnings are issued for bins with unusually low or zero coverage, which
may indicate problematic baited regions.
See Also
--------
combine_probes : Internal function that performs the actual reference creation
infer_sexes : Infers sample sex from X/Y chromosome coverage
warn_bad_bins : Warns about bins with problematic coverage
Examples
--------
Basic reference from normal samples:
>>> ref = do_reference(
... ['N1.targetcoverage.cnn', 'N2.targetcoverage.cnn'],
... ['N1.antitargetcoverage.cnn', 'N2.antitargetcoverage.cnn'],
... fa_fname='hg19.fa'
... )
Reference without FASTA (no GC/RM correction):
>>> ref = do_reference(
... ['N1.targetcoverage.cnn', 'N2.targetcoverage.cnn'],
... ['N1.antitargetcoverage.cnn', 'N2.antitargetcoverage.cnn']
... )
Reference with explicit male samples:
>>> ref = do_reference(
... target_fnames=['M1.cnn', 'M2.cnn'],
... fa_fname='hg38.fa',
... female_samples=False,
... diploid_parx_genome='hg38'
... )
"""
if antitarget_fnames:
core.assert_equal(
"Unequal number of target and antitarget files given",
targets=len(target_fnames),
antitargets=len(antitarget_fnames),
)
if not fa_fname:
logging.info("No FASTA reference genome provided; skipping GC, RM calculations")
if female_samples is None:
# NB: Antitargets are usually preferred for inferring sex, but might be
# empty files, in which case no inference can be done. Since targets are
# guaranteed to exist, infer from those first, then replace those
# values where antitargets are suitable.
logging.info("Sample sex not provided; inferring from samples. ")
sexes = infer_sexes(target_fnames, False, diploid_parx_genome)
if antitarget_fnames:
a_sexes = infer_sexes(antitarget_fnames, False, diploid_parx_genome)
for sid, a_is_xx in a_sexes.items():
t_is_xx = sexes.get(sid)
if t_is_xx is None:
sexes[sid] = a_is_xx
elif t_is_xx != a_is_xx and a_is_xx is not None:
logging.warning(
"Sample %s chromosomal X/Y ploidy looks "
"like %s in targets but %s in antitargets; "
"preferring antitargets",
sid,
"female" if t_is_xx else "male",
"female" if a_is_xx else "male",
)
sexes[sid] = a_is_xx
else:
# In this case the gender of the samples is provided and won't be inferred
logging.info(
f"Provided sample sex is {'female' if female_samples else 'male'}. "
)
sexes = dict()
for fname in target_fnames:
sexes[read_cna(fname).sample_id] = female_samples
# TODO - refactor/inline this func here, once it works
ref_probes = combine_probes(
target_fnames,
antitarget_fnames,
fa_fname,
is_haploid_x_reference,
diploid_parx_genome,
sexes,
do_gc,
do_edge,
do_rmask,
do_cluster,
min_cluster_size,
)
warn_bad_bins(ref_probes)
return ref_probes
[docs]
def infer_sexes(
cnn_fnames: list[str], is_haploid_x: bool, diploid_parx_genome: Optional[str]
) -> dict[str, bool_]:
"""Map sample IDs to inferred chromosomal sex, where possible.
For samples where the source file is empty or does not include either sex
chromosome, that sample ID will not be in the returned dictionary.
"""
sexes = {}
for fname in cnn_fnames:
cnarr = read_cna(fname)
if cnarr:
is_xx = cnarr.guess_xx(is_haploid_x, diploid_parx_genome)
if is_xx is not None:
sexes[cnarr.sample_id] = is_xx
return sexes
[docs]
def combine_probes(
filenames: list[str],
antitarget_fnames: Optional[list[str]],
fa_fname: Optional[str],
is_haploid_x: bool,
diploid_parx_genome: Optional[str],
sexes: dict[str, Union[bool_, bool]],
fix_gc: bool,
fix_edge: bool,
fix_rmask: bool,
do_cluster: bool,
min_cluster_size: int,
) -> CNA:
"""Calculate the median coverage of each bin across multiple samples.
Parameters
----------
filenames : list
List of string filenames, corresponding to targetcoverage.cnn and
antitargetcoverage.cnn files, as generated by 'coverage' or
'import-picard'.
fa_fname : str
Reference genome sequence in FASTA format, used to extract GC and
RepeatMasker content of each genomic bin.
is_haploid_x : bool
do_cluster : bool
fix_gc : bool
fix_edge : bool
fix_rmask : bool
Returns
-------
CopyNumArray
One object summarizing the coverages of the input samples, including
each bin's "average" coverage, "spread" of coverages, and GC content.
"""
# Special parameters:
# skip_low (when centering) = True for target; False for antitarget
# do_edge = as given for target; False for antitarget
# do_rmask = False for target; as given for antitarget
ref_df, all_logr, all_depths = load_sample_block(
filenames,
fa_fname,
is_haploid_x,
diploid_parx_genome,
sexes,
True,
fix_gc,
fix_edge,
False,
)
if antitarget_fnames:
# XXX TODO ensure ordering matches targets!
# argsort on both -> same?
anti_ref_df, anti_logr, anti_depths = load_sample_block(
antitarget_fnames,
fa_fname,
is_haploid_x,
diploid_parx_genome,
sexes,
False,
fix_gc,
False,
fix_rmask,
)
if not anti_ref_df.empty:
ref_df = pd.concat([ref_df, anti_ref_df], ignore_index=True)
all_logr = np.hstack([all_logr, anti_logr])
all_depths = np.hstack([all_depths, anti_depths])
stats_all = summarize_info(all_logr, all_depths)
ref_df = ref_df.assign(**stats_all)
if do_cluster:
# Get extra cols, concat axis=1 here (DATAFRAME v-concat)
sample_ids = [core.fbase(f) for f in filenames]
if len(sample_ids) != len(all_logr) - 1:
raise ValueError(
f"Expected {len(all_logr) - 1} target coverage files (.cnn), "
+ f"got {len(sample_ids)}"
)
clustered_cols = create_clusters(all_logr, min_cluster_size, sample_ids)
if clustered_cols:
try:
ref_df = ref_df.assign(**clustered_cols)
except ValueError as exc:
print("Reference:", len(ref_df.index))
for cl_key, cl_col in clustered_cols.items():
print(cl_key, ":", len(cl_col))
raise exc
else:
print("** Why weren't there any clustered cols?")
ref_cna = CNA(ref_df, meta_dict={"sample_id": "reference"})
# NB: Up to this point, target vs. antitarget bins haven't been row-sorted.
# Holding off until here ensures the cluster-specific log2 columns are
# sorted with the same row order as the rest of the CNA dataframe.
ref_cna.sort()
ref_cna.sort_columns()
# TODO figure out centering
# ref_probes.center_all(skip_low=True) # <-- on each cluster col, too?
# or just subtract the same amount from the cluster calls after figuring
# out the main one?
return ref_cna
[docs]
def load_sample_block(
filenames: list[str],
fa_fname: Optional[str],
is_haploid_x: bool,
diploid_parx_genome: Optional[str],
sexes: dict[str, Union[bool_, bool]],
skip_low: bool,
fix_gc: bool,
fix_edge: bool,
fix_rmask: bool,
) -> tuple[pd.DataFrame, ndarray, ndarray]:
r"""Load and summarize a pool of \*coverage.cnn files.
Run separately for the on-target and (optional) antitarget bins.
Returns
-------
ref_df : pandas.DataFrame
All columns needed for the reference CNA object, including
aggregate log2 and spread.
all_logr : numpy.ndarray
All sample log2 ratios, as a 2D matrix (rows=bins, columns=samples),
to be used with do_cluster.
"""
# Ensures samples' target and antitarget matrix columns are in the same
# order, so they can be concatenated.
# (We don't explicitly pair off each sample's target and antitarget .cnn
# files; as long as the filename prefixes match, the resulting columns will
# be consistent. Same is true for sample sex inference.)
filenames = sorted(filenames, key=core.fbase)
# Load coverage from target/antitarget files
logging.info("Loading %s", filenames[0])
cnarr1 = read_cna(filenames[0])
if len(cnarr1) == 0:
# Just create an empty array with the right columns
col_names = ["chromosome", "start", "end", "gene", "log2", "depth"]
if "gc" in cnarr1 or fa_fname:
col_names.append("gc")
if fa_fname:
col_names.append("rmask")
col_names.append("spread")
empty_df = pd.DataFrame.from_records([], columns=col_names)
empty_logr = np.array([[]] * (len(filenames) + 1))
empty_dp = np.array([[]] * len(filenames))
return empty_df, empty_logr, empty_dp
# Calculate GC and RepeatMasker content for each probe's genomic region
ref_columns = {
"chromosome": cnarr1.chromosome,
"start": cnarr1.start,
"end": cnarr1.end,
"gene": cnarr1["gene"],
}
if fa_fname and (fix_rmask or fix_gc):
gc, rmask = get_fasta_stats(cnarr1, fa_fname)
if fix_gc:
ref_columns["gc"] = gc
if fix_rmask:
ref_columns["rmask"] = rmask
elif "gc" in cnarr1 and fix_gc:
# Reuse .cnn GC values if they're already stored (via import-picard)
gc = cnarr1["gc"]
ref_columns["gc"] = gc
# Make the sex-chromosome coverages of male and female samples compatible
is_chr_x = cnarr1.chr_x_filter(diploid_parx_genome)
is_chr_y = cnarr1.chr_y_filter(diploid_parx_genome)
ref_flat_logr = cnarr1.expect_flat_log2(is_haploid_x, diploid_parx_genome)
ref_edge_bias = fix.get_edge_bias(cnarr1, params.INSERT_SIZE)
# Pseudocount of 1 "flat" sample
all_depths = [cnarr1["depth"] if "depth" in cnarr1 else np.exp2(cnarr1["log2"])]
all_logr = [
ref_flat_logr,
bias_correct_logr(
cnarr1,
ref_columns,
ref_edge_bias,
ref_flat_logr,
sexes,
is_chr_x,
is_chr_y,
fix_gc,
fix_edge,
fix_rmask,
skip_low,
diploid_parx_genome,
),
]
# Load only coverage depths from the remaining samples
for fname in filenames[1:]:
logging.info("Loading %s", fname)
cnarrx = read_cna(fname)
# Bin information should match across all files
if not np.array_equal(
cnarr1.data.loc[:, ("chromosome", "start", "end", "gene")].values,
cnarrx.data.loc[:, ("chromosome", "start", "end", "gene")].values,
):
raise RuntimeError(f"{fname} bins do not match those in {filenames[0]}")
all_depths.append(
cnarrx["depth"] if "depth" in cnarrx else np.exp2(cnarrx["log2"])
)
all_logr.append(
bias_correct_logr(
cnarrx,
ref_columns,
ref_edge_bias,
ref_flat_logr,
sexes,
is_chr_x,
is_chr_y,
fix_gc,
fix_edge,
fix_rmask,
skip_low,
diploid_parx_genome,
)
)
all_logr = np.vstack(all_logr)
all_depths = np.vstack(all_depths)
ref_df = pd.DataFrame.from_dict(ref_columns)
return ref_df, all_logr, all_depths
[docs]
def bias_correct_logr(
cnarr: CNA,
ref_columns: dict[str, Union[pd.Series, ndarray]],
ref_edge_bias: pd.Series,
ref_flat_logr: ndarray,
sexes: dict[str, Union[bool_, bool]],
is_chr_x: pd.Series,
is_chr_y: pd.Series,
fix_gc: bool,
fix_edge: bool,
fix_rmask: bool,
skip_low: bool,
diploid_parx_genome: Optional[str],
) -> pd.Series:
"""Perform bias corrections on the sample."""
cnarr.center_all(skip_low=skip_low, diploid_parx_genome=diploid_parx_genome)
shift_sex_chroms(cnarr, sexes, ref_flat_logr, is_chr_x, is_chr_y)
# Skip bias corrections if most bins have no coverage (e.g. user error)
if (
cnarr["log2"] > params.NULL_LOG2_COVERAGE - params.MIN_REF_COVERAGE
).sum() <= len(cnarr) // 2:
logging.warning(
"WARNING: most bins have no or very low coverage; "
"check that the right BED file was used"
)
else:
if "gc" in ref_columns and fix_gc:
logging.info(f"Correcting for GC bias for {cnarr.sample_id}...")
cnarr = fix.center_by_window(cnarr, 0.1, ref_columns["gc"])
if "rmask" in ref_columns and fix_rmask:
logging.info(f"Correcting for RepeatMasker bias for {cnarr.sample_id}...")
cnarr = fix.center_by_window(cnarr, 0.1, ref_columns["rmask"])
if fix_edge:
logging.info(f"Correcting for density bias for {cnarr.sample_id}...")
cnarr = fix.center_by_window(cnarr, 0.1, ref_edge_bias)
return cnarr["log2"]
[docs]
def shift_sex_chroms(
cnarr: CNA,
sexes: dict[str, Union[bool_, bool]],
ref_flat_logr: ndarray,
is_chr_x: pd.Series,
is_chr_y: pd.Series,
) -> None:
"""Shift sample X and Y chromosomes to match the reference sex.
Reference values::
XY: chrX -1, chrY -1
XX: chrX 0, chrY -1
Plan::
chrX:
xx sample, xx ref: 0 (from 0)
xx sample, xy ref: -= 1 (from -1)
xy sample, xx ref: += 1 (from 0) +1
xy sample, xy ref: 0 (from -1) +1
chrY:
xx sample, xx ref: = -1 (from -1)
xx sample, xy ref: = -1 (from -1)
xy sample, xx ref: 0 (from -1) +1
xy sample, xy ref: 0 (from -1) +1
"""
is_xx = sexes.get(cnarr.sample_id)
cnarr["log2"] += ref_flat_logr
if is_xx:
# chrX has same ploidy as autosomes; chrY is just unusable noise
cnarr[is_chr_y, "log2"] = -1.0 # np.nan is worse
else:
# 1/2 #copies of each sex chromosome
cnarr[is_chr_x | is_chr_y, "log2"] += 1.0
[docs]
def summarize_info(all_logr: ndarray, all_depths: ndarray) -> dict[str, ndarray]:
"""Average & spread of log2ratios and depths for a group of samples.
Can apply to all samples, or a given cluster of samples.
"""
logging.info("Calculating average bin coverages")
cvg_centers = np.apply_along_axis(descriptives.biweight_location, 0, all_logr)
depth_centers = np.apply_along_axis(descriptives.biweight_location, 0, all_depths)
logging.info("Calculating bin spreads")
spreads = np.array(
[
descriptives.biweight_midvariance(a, initial=i)
for a, i in zip(all_logr.T, cvg_centers, strict=False)
]
)
result = {
"log2": cvg_centers,
"depth": depth_centers,
"spread": spreads,
}
# TODO center the resulting log2
# ref_df = pd.DataFrame.from_dict(ref_columns)
# ref_cna = CNA.from_columns(ref_columns, {'sample_id': "reference"})
return result
[docs]
def create_clusters(logr_matrix, min_cluster_size, sample_ids):
"""Extract and summarize clusters of samples in logr_matrix.
1. Calculate correlation coefficients between all samples (columns).
2. Cluster the correlation matrix.
3. For each resulting sample cluster (down to a minimum size threshold),
calculate the central log2 value for each bin, similar to the full pool.
Also print the sample IDs in each cluster, if feasible.
Also recalculate and store the 'spread' of each cluster, though this might
not be necessary/good.
Return a DataFrame of just the log2 values. Column names are ``log2_i``
where i=1,2,... .
"""
# from .cluster import markov
from .cluster import kmeans
# Drop the pseudocount sample
logr_matrix = logr_matrix[1:, :]
print("Clustering", len(logr_matrix), "samples...")
# clusters = markov(logr_matrix)
clusters = kmeans(logr_matrix)
cluster_cols = {}
sample_ids = np.array(sample_ids) # For easy indexing
for i, clust_idx in enumerate(clusters):
i += 1
# print(len(clust_idx), clust_idx)
if len(clust_idx) < min_cluster_size:
logging.info(
"Skipping cluster #%d, size %d < min. %d",
i,
len(clust_idx),
min_cluster_size,
)
continue
logging.info("Summarizing cluster #%d of %d samples", i, len(clust_idx))
# List which samples are in each cluster
samples = sample_ids[clust_idx]
logging.info("\n".join(["\t" + s for s in samples]))
# Calculate each cluster's summary stats
clust_matrix = logr_matrix[clust_idx, :]
# XXX re-add the pseudocount sample to each cluster? need benchmark
clust_info = summarize_info(clust_matrix, [])
cluster_cols.update(
{
f"log2_{i}": clust_info["log2"],
f"spread_{i}": clust_info["spread"],
}
)
return cluster_cols
[docs]
def warn_bad_bins(cnarr: CNA, max_name_width: int = 50) -> None:
"""Warn about target bins where coverage is poor.
Prints a formatted table to stderr.
"""
bad_bins = cnarr[fix.mask_bad_bins(cnarr)]
fg_index = ~bad_bins["gene"].isin(params.ANTITARGET_ALIASES)
fg_bad_bins = bad_bins[fg_index]
if len(fg_bad_bins) > 0:
bad_pct = (
100 * len(fg_bad_bins) / sum(~cnarr["gene"].isin(params.ANTITARGET_ALIASES))
)
logging.info(
"Targets: %d (%s) bins failed filters (log2 < %s, log2 > %s, spread > %s)",
len(fg_bad_bins),
f"{bad_pct:.4}%",
params.MIN_REF_COVERAGE,
-params.MIN_REF_COVERAGE,
params.MAX_REF_SPREAD,
)
if len(fg_bad_bins) < 500:
gene_cols = min(max_name_width, max(map(len, fg_bad_bins["gene"])))
labels = fg_bad_bins.labels()
chrom_cols = max(labels.apply(len))
last_gene = None
for label, probe in zip(labels, fg_bad_bins, strict=False):
if probe.gene == last_gene:
gene = ' "'
else:
gene = probe.gene
last_gene = gene
if len(gene) > max_name_width:
gene = gene[: max_name_width - 3] + "..."
if "rmask" in cnarr:
logging.info(
" %s %s log2=%.3f spread=%.3f rmask=%.3f",
gene.ljust(gene_cols),
label.ljust(chrom_cols),
probe.log2,
probe.spread,
probe.rmask,
)
else:
logging.info(
" %s %s log2=%.3f spread=%.3f",
gene.ljust(gene_cols),
label.ljust(chrom_cols),
probe.log2,
probe.spread,
)
# Count the number of BG bins dropped, too (names are all "Antitarget")
bg_bad_bins = bad_bins[~fg_index]
if len(bg_bad_bins) > 0:
bad_pct = (
100 * len(bg_bad_bins) / sum(cnarr["gene"].isin(params.ANTITARGET_ALIASES))
)
logging.info(
"Antitargets: %d (%s) bins failed filters",
len(bg_bad_bins),
f"{bad_pct:.4}%",
)
[docs]
def get_fasta_stats(cnarr: CNA, fa_fname: str) -> tuple[ndarray, ndarray]:
"""Calculate GC and RepeatMasker content of each bin in the FASTA genome."""
logging.info("Calculating GC and RepeatMasker content in %s ...", fa_fname)
gc_rm_vals = [
calculate_gc_lo(subseq) for subseq in fasta_extract_regions(fa_fname, cnarr)
]
gc_vals, rm_vals = zip(*gc_rm_vals, strict=False)
return np.asarray(gc_vals, dtype=float), np.asarray(rm_vals, dtype=float)
[docs]
def calculate_gc_lo(subseq: str) -> tuple[float, float]:
"""Calculate the GC and lowercase (RepeatMasked) content of a string."""
cnt_at_lo = subseq.count("a") + subseq.count("t")
cnt_at_up = subseq.count("A") + subseq.count("T")
cnt_gc_lo = subseq.count("g") + subseq.count("c")
cnt_gc_up = subseq.count("G") + subseq.count("C")
tot = float(cnt_gc_up + cnt_gc_lo + cnt_at_up + cnt_at_lo)
if not tot:
return 0.0, 0.0
frac_gc = (cnt_gc_lo + cnt_gc_up) / tot
frac_lo = (cnt_at_lo + cnt_gc_lo) / tot
return frac_gc, frac_lo
[docs]
def reference2regions(refarr: CNA) -> tuple[GA, GA]:
"""Split reference into target and antitarget regions."""
is_bg = refarr["gene"].isin(params.ANTITARGET_ALIASES)
regions = GA(
refarr.data.loc[:, ("chromosome", "start", "end", "gene")],
{"sample_id": "reference"},
)
targets = regions[~is_bg]
antitargets = regions[is_bg]
return targets, antitargets