Matiq/ffpe_damage
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

filter damaged variants: clean output + VCF filtering

Browse Source 
- flag C>T/G>A at read ends (end<=10) as FFPE-suspect, strand bias as supporting evidence
- outputs: candidate_variants.tsv [is_ffpe_suspect,ffpe_reason], clean_variants.tsv,
  ffpe_suspect_variants.tsv, candidate_CtoT/GtoA_clean.bed
- optional --vcf: produce .clean.vcf (without damaged) and .ffpe_flagged.vcf;
  with --keep-damaged mark FILTER=FFPE instead of removing
- auto-create BAM/FASTA index if missing, improved error messages
This commit is contained in:
Matiq
2026-09-06 18:18:29 +03:00
parent 5fe6298926
commit 5385f76447
2 changed files with 181 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.gitignore
+11
View File
@@ -3,3 +3,14 @@ __pycache__/
*.pyc
test_data/
FFPE_QC/
FFPE_sample/
*.bam
*.bai
*.fai
hg38.fa
hg38.fa.gz
*.vcf.gz
*.tbi
*.zip
README.html
124/
ffpe_damage_v2.py
+168 -4
View File
@@ -63,6 +63,17 @@ def parse_args():
ap.add_argument("--min-depth", type=int, default=20)
ap.add_argument("--min-alt-count", type=int, default=5)
ap.add_argument("--include-duplicates", action="store_true")
ap.add_argument("--vcf", default=None,
help="optional VCF to filter: FFPE-suspect variants are "
"marked FILTER=FFPE or removed")
ap.add_argument("--keep-damaged", action="store_true",
help="when --vcf is given, keep damaged variants with "
"FILTER=FFPE instead of removing them")
ap.add_argument("--filter-end", type=int, default=10,
help="mean read-position distance from either end that "
"counts as end-associated (default 10)")
ap.add_argument("--filter-strand-p", type=float, default=0.05,
help="Fisher p-value threshold for strand bias (default 0.05)")
return ap.parse_args()
@@ -317,6 +328,23 @@ def write_end_enrichment(d, args):
return df
def _is_ffpe_suspect(row, end_thresh, strand_p_thresh):
sub = f"{row['ref']}>{row['alt']}"
if sub not in ("C>T", "G>A"):
return False, ""
is_end = (row["mean_read_position_5prime"] <= end_thresh or
row["mean_read_position_3prime"] <= end_thresh)
if not is_end:
return False, ""
reasons = [f"end<={end_thresh}"]
frac_plus = row["alt_fraction_plus"]
if frac_plus <= 0.1 or frac_plus >= 0.9:
reasons.append(f"strand_frac={frac_plus:.2f}")
if row["strand_bias_pvalue_fisher"] < strand_p_thresh:
reasons.append(f"strand_p={row['strand_bias_pvalue_fisher']:.2g}")
return True, ";".join(reasons)
def write_beds_and_variants(d, args):
ct_lines = []
ga_lines = []
@@ -380,10 +408,47 @@ def write_beds_and_variants(d, args):
with open(os.path.join(args.outdir, "candidate_GtoA.bed"), "w") as f:
f.writelines(ga_lines)
if var_rows:
pd.DataFrame(var_rows).to_csv(
os.path.join(args.outdir, "candidate_variants.tsv"),
df = pd.DataFrame(var_rows)
flags = df.apply(lambda r: _is_ffpe_suspect(r, args.filter_end,
args.filter_strand_p), axis=1)
df["is_ffpe_suspect"] = [x[0] for x in flags]
df["ffpe_reason"] = [x[1] for x in flags]
df.to_csv(os.path.join(args.outdir, "candidate_variants.tsv"),
index=False, sep="\t")
clean = df[~df["is_ffpe_suspect"]]
damaged = df[df["is_ffpe_suspect"]]
clean.to_csv(os.path.join(args.outdir, "clean_variants.tsv"),
index=False, sep="\t")
damaged.to_csv(os.path.join(args.outdir, "ffpe_suspect_variants.tsv"),
index=False, sep="\t")
for name, subset, fname in [
("clean", clean, "candidate_CtoT_clean.bed"),
("clean", clean, "candidate_GtoA_clean.bed"),
]:
pass
clean_ct = []
clean_ga = []
for _, r in clean.iterrows():
line = (f"{r['chrom']}\t{r['position']}\t{r['position']+1}\t"
f"{r['ref']}>{r['alt']};depth={int(r['depth'])};"
f"alt={int(r['alt_count'])};VAF={r['VAF']:.3f}\n")
if r["ref"] == "C" and r["alt"] == "T":
clean_ct.append(line)
if r["ref"] == "G" and r["alt"] == "A":
clean_ga.append(line)
with open(os.path.join(args.outdir, "candidate_CtoT_clean.bed"), "w") as f:
f.writelines(clean_ct)
with open(os.path.join(args.outdir, "candidate_GtoA_clean.bed"), "w") as f:
f.writelines(clean_ga)
return df
else:
open(os.path.join(args.outdir, "clean_variants.tsv"), "w").close()
open(os.path.join(args.outdir, "ffpe_suspect_variants.tsv"), "w").close()
return pd.DataFrame()
def write_summary(d, args):
ct = d["sub_counts"]["C>T"]
@@ -454,15 +519,106 @@ def plot_cg_profiles(cg5, cg3, outdir):
plt.close()
def _filter_vcf(args, flagged_df):
vcf_in = args.vcf
outdir = args.outdir
base = os.path.splitext(os.path.basename(vcf_in))[0]
if base.endswith(".vcf"):
base = base[:-4]
clean_path = os.path.join(outdir, f"{base}.clean.vcf")
flagged_path = os.path.join(outdir, f"{base}.ffpe_flagged.vcf")
flagged_keys = set()
for _, r in flagged_df.iterrows():
if r.get("is_ffpe_suspect"):
flagged_keys.add((str(r["chrom"]), int(r["position"]) + 1,
str(r["ref"]), str(r["alt"])))
n_total = n_kept = n_flagged = 0
with open(vcf_in) as fin, \
open(clean_path, "w") as fout_clean, \
open(flagged_path, "w") as fout_flagged:
ffpe_header = '##FILTER=<ID=FFPE,Description="FFPE deamination suspect (end-associated C>T/G>A, strand-biased)">'
header_done = False
for line in fin:
if line.startswith("##FILTER=<ID=FFPE"):
fout_clean.write(line)
fout_flagged.write(line)
header_done = True
continue
if line.startswith("#CHROM"):
if not header_done:
fout_clean.write(ffpe_header + "\n")
fout_flagged.write(ffpe_header + "\n")
fout_clean.write(line)
fout_flagged.write(line)
continue
if line.startswith("#"):
fout_clean.write(line)
fout_flagged.write(line)
continue
n_total += 1
parts = line.rstrip("\n").split("\t")
if len(parts) < 8:
fout_clean.write(line)
continue
chrom, pos_s, _, ref, alt_s = parts[0], parts[1], parts[2], parts[3], parts[4]
try:
pos = int(pos_s)
except ValueError:
fout_clean.write(line)
continue
alts = alt_s.split(",")
is_flagged = any((chrom, pos, ref, a) in flagged_keys for a in alts)
filt_col = parts[6] if len(parts) > 6 else "."
if is_flagged:
n_flagged += 1
if args.keep_damaged:
new_filt = "FFPE" if filt_col in (".", "PASS", "") else filt_col + ";FFPE"
parts[6] = new_filt
fout_clean.write("\t".join(parts) + "\n")
fout_flagged.write(line)
else:
fout_flagged.write(line)
else:
n_kept += 1
fout_clean.write(line)
print(f"[vcf] total {n_total}, kept {n_kept}, "
f"flagged {n_flagged} -> {clean_path} / {flagged_path}")
def main():
args = parse_args()
os.makedirs(args.outdir, exist_ok=True)
bam = pysam.AlignmentFile(args.bam, "rb")
if not bam.has_index():
try:
print(f"[info] BAM index not found, creating {args.bam}.bai ...")
pysam.index(args.bam)
bam.close()
raise RuntimeError("BAM index (.bai) not found.")
bam = pysam.AlignmentFile(args.bam, "rb")
except Exception as e:
bam.close()
raise RuntimeError(
f"BAM index (.bai) not found and auto-creation failed: {e}. "
f"Run: samtools index {args.bam}") from e
try:
fasta = pysam.FastaFile(args.reference)
except Exception as e:
bam.close()
if "fai" in str(e).lower() or "index" in str(e).lower():
try:
print(f"[info] FASTA index not found, creating {args.reference}.fai ...")
pysam.faidx(args.reference)
fasta = pysam.FastaFile(args.reference)
except Exception as e2:
raise RuntimeError(
f"FASTA index (.fai) not found and auto-creation failed: {e2}. "
f"Run: samtools faidx {args.reference}") from e2
else:
raise
d = process_bam(bam, fasta, args)
bam.close()
@@ -473,11 +629,14 @@ def main():
cg5, cg3 = write_cg_profiles(d, args)
write_strand_profiles(d, args)
write_end_enrichment(d, args)
write_beds_and_variants(d, args)
flagged_df = write_beds_and_variants(d, args)
write_summary(d, args)
plot_all12(sub_df, args.outdir)
plot_cg_profiles(cg5, cg3, args.outdir)
if args.vcf:
_filter_vcf(args, flagged_df)
print(f"total reads : {d['total_reads']}")
print(f"used reads : {d['used_reads']}")
print(f"usable bases : {d['usable_bases']}")
@@ -488,6 +647,11 @@ def main():
f"{d['sub_counts']['G>A']} / {d['ref_base_counts']['G']} = "
f"{d['sub_counts']['G>A'] / d['ref_base_counts']['G'] * 100:.3f}%")
print(f"outputs : {args.outdir}/")
if not flagged_df.empty:
n_flag = int(flagged_df["is_ffpe_suspect"].sum())
n_clean = len(flagged_df) - n_flag
print(f"variants flagged FFPE : {n_flag} / {len(flagged_df)} "
f"-> ffpe_suspect_variants.tsv / clean_variants.tsv")
if __name__ == "__main__":