pick best transcript per variant (1:1) default, --all-transcripts for expansion

- offline GTF now picks MANE>appris1>canonical>basic>protein_coding (318 vs 3018 rows at VAF>10%)
- online Ensembl REST likewise picks best (mane_select/canonical)
- --all-transcripts restores previous all-transcripts behavior

annotate_clean.py

@@ -54,13 +54,12 @@ def load_token(path="~/.config/oncokb/token"):
         return p.read_text().strip()
     return os.environ.get("ONCOKB_TOKEN", "").strip()
 
-def vep_annotate(df, reference, gtf_path=None, offline=False):
+def vep_annotate(df, reference, gtf_path=None, offline=False, all_transcripts=False):
     """Try VEP if installed and cache exists, else GTF offline, else Ensembl REST, else fallback."""
     import shutil
     vep_bin = shutil.which("vep")
     if vep_bin and gtf_path and Path(gtf_path).expanduser().is_file():
         pass
-    # Load GTF for offline all-transcript expansion
     gtf_transcripts = None
     if offline and gtf_path:
         gtf_file = Path(gtf_path).expanduser()
@@ -89,11 +88,10 @@ def vep_annotate(df, reference, gtf_path=None, offline=False):
             })
             continue
         if offline and gtf_transcripts is not None:
-            # GTF offline: all transcripts overlapping pos
             hits = [t for t in gtf_transcripts if t["chrom"] == chrom and t["start"] <= pos1 <= t["end"]]
             if hits:
+                if all_transcripts:
                     for t in hits:
-                    # Simple HGVS: use transcript ID + positional offset (placeholder, VEP would give exact c./p.)
                         hgvs_c = f"{t['tx']}:c.{pos1}{ref}>{alt}"
                         hgvs_p = "p.(?)"
                         effect = t["biotype"] or "transcript_variant"
@@ -106,6 +104,22 @@ def vep_annotate(df, reference, gtf_path=None, offline=False):
                             "Тип варианта и эффект": effect,
                         })
                     continue
+                # pick best transcript (MANE > appris1 > canonical > basic > protein_coding)
+                def _score(t):
+                    return (t["is_mane"], t["appris"] == 1, t["appris"] == 2, t["is_canonical"], t["is_basic"], t["is_ccds"], t["biotype"] == "protein_coding")
+                best = max(hits, key=_score)
+                hgvs_c = f"{best['tx']}:c.{pos1}{ref}>{alt}"
+                hgvs_p = "p.(?)"
+                effect = best["biotype"] or "transcript_variant"
+                rows.append({
+                    "_orig_idx": r.name,
+                    "Ген": best["gene"],
+                    "HGVS_c": hgvs_c,
+                    "HGVS_p": hgvs_p,
+                    "HGVS": f"{hgvs_c} {hgvs_p} ({hgvs_g})",
+                    "Тип варианта и эффект": effect,
+                })
+                continue
             rows.append({
                 "_orig_idx": r.name,
                 "Ген": "intergenic",
@@ -125,7 +139,7 @@ def vep_annotate(df, reference, gtf_path=None, offline=False):
                 "Тип варианта и эффект": "SNV, missense_variant (predicted)" if len(ref)==1 and len(alt)==1 else "indel",
             })
             continue
-        # Online: Ensembl REST, expand all transcripts
+        # Online: Ensembl REST
         try:
             hgvs_ens = f"{chrom.replace('chr','')}:g.{pos1}{ref}>{alt}"
             url = f"https://rest.ensembl.org/vep/homo_sapiens/hgvs/{hgvs_ens}?content-type=application/json"
@@ -134,6 +148,7 @@ def vep_annotate(df, reference, gtf_path=None, offline=False):
                 j = resp.json()
                 tcs = j[0].get("transcript_consequences", []) if j and isinstance(j, list) and j[0] else []
                 if tcs:
+                    if all_transcripts:
                         for tc in tcs:
                             gene = tc.get("gene_symbol") or "intergenic"
                             hgvs_c = tc.get("hgvsc") or f"c.{pos1}{ref}>{alt}"
@@ -153,6 +168,27 @@ def vep_annotate(df, reference, gtf_path=None, offline=False):
                                 "Тип варианта и эффект": effect,
                             })
                         continue
+                    def _score_tc(tc):
+                        return (tc.get("mane_select") is not None, tc.get("canonical") == 1, tc.get("biotype") == "protein_coding", tc.get("impact") == "HIGH")
+                    tc = max(tcs, key=_score_tc)
+                    gene = tc.get("gene_symbol") or "intergenic"
+                    hgvs_c = tc.get("hgvsc") or f"c.{pos1}{ref}>{alt}"
+                    hgvs_p = tc.get("hgvsp") or "p.(?)"
+                    cons = tc.get("consequence_terms", [])
+                    effect = ", ".join(cons) if cons else ("SNV, missense_variant (predicted)" if len(ref)==1 and len(alt)==1 else "indel")
+                    if ":" in hgvs_c:
+                        hgvs_c = hgvs_c.split(":")[-1]
+                    if ":" in hgvs_p:
+                        hgvs_p = hgvs_p.split(":")[-1]
+                    rows.append({
+                        "_orig_idx": r.name,
+                        "Ген": gene,
+                        "HGVS_c": hgvs_c,
+                        "HGVS_p": hgvs_p,
+                        "HGVS": f"{hgvs_c} {hgvs_p} ({hgvs_g})",
+                        "Тип варианта и эффект": effect,
+                    })
+                    continue
         except Exception:
             pass
         rows.append({
@@ -295,6 +331,7 @@ def build_excel(df_clean, df_annot, out_path):
 
 def _load_gtf(gtf_path):
     import gzip
+    import re
     transcripts = []
     try:
         opener = gzip.open if str(gtf_path).endswith(".gz") else open
@@ -308,13 +345,11 @@ def _load_gtf(gtf_path):
                 chrom = parts[0]
                 start = int(parts[3]); end = int(parts[4])
                 attr = parts[8]
-                # parse gene_name, transcript_id
-                import re
                 m_gene = re.search(r'gene_name "([^"]+)"', attr)
                 m_tx = re.search(r'transcript_id "([^"]+)"', attr)
                 m_biotype = re.search(r'transcript_type "([^"]+)"', attr)
                 if m_gene and m_tx:
-                    # Normalize chr prefix: GTF uses chr1, our clean uses chr1 -> keep as is
+                    tags = set(re.findall(r'tag "([^"]+)"', attr))
                     transcripts.append({
                         "chrom": chrom,
                         "start": start,
@@ -322,6 +357,11 @@ def _load_gtf(gtf_path):
                         "gene": m_gene.group(1),
                         "tx": m_tx.group(1),
                         "biotype": m_biotype.group(1) if m_biotype else "",
+                        "is_mane": "MANE_Select" in tags,
+                        "is_canonical": "Ensembl_canonical" in tags,
+                        "is_basic": "basic" in tags,
+                        "is_ccds": "CCDS" in tags,
+                        "appris": next((int(t.split("_")[-1]) for t in tags if t.startswith("appris_principal_")), 99),
                     })
     except Exception as e:
         print(f"[gtf] failed to load {gtf_path}: {e}", file=sys.stderr)
@@ -337,10 +377,11 @@ def main():
     ap.add_argument("--tumor-type", default="All Solid Tumors", help="OncoKB tumor type (pan-cancer default)")
     ap.add_argument("--token", default="~/.config/oncokb/token", help="OncoKB token file or env")
     ap.add_argument("--offline", action="store_true", help="skip Ensembl/OncoKB network calls (fast, offline)")
+    ap.add_argument("--all-transcripts", action="store_true", help="expand all transcripts (default: pick best MANE/canonical per variant)")
     ap.add_argument("--min-vaf", type=float, default=None, help="filter VAF > threshold (e.g. 0.10 for 10%%)")
     ap.add_argument("--min-depth", type=int, default=None, help="filter depth >= threshold")
     ap.add_argument("--gtf", default="~/Projects/ffpe_damage/references/gencode.v44.annotation.gtf.gz",
-                    help="GENCODE GTF for offline all-transcript expansion")
+                    help="GENCODE GTF for offline expansion (pick best when not --all-transcripts)")
     args = ap.parse_args()
 
     clean_path = Path(args.clean)
@@ -365,7 +406,7 @@ def main():
         print(f"[warn] {len(df_clean)} variants - network annotation will be slow. Use --offline for fast placeholder.", flush=True)
 
     # HGVS / gene / effect
-    df_annot = vep_annotate(df_clean, args.reference, gtf_path=args.gtf, offline=args.offline)
+    df_annot = vep_annotate(df_clean, args.reference, gtf_path=args.gtf, offline=args.offline, all_transcripts=args.all_transcripts)
 
     # OncoKB
     token = load_token(args.token)