merge-cli 3.5.0__tar.gz → 3.6__tar.gz

{merge_cli-3.5.0 → merge_cli-3.6}/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: merge-cli
-Version: 3.5.0
-Summary: MERGE 变异致病性预测 CLI（服务器固定，集成模型内嵌，无需手动配置）
+Version: 3.6
+Summary: MERGE variant pathogenicity prediction CLI (fixed server, integrated model embedded, no manual configuration required)
 Project-URL: Homepage, https://merge.fanglab.cn
 Requires-Python: >=3.11
 Requires-Dist: click>=8.1

{merge_cli-3.5.0 → merge_cli-3.6}/merge_cli/cli.py

@@ -7,7 +7,8 @@ Changes (v3.5.0):
   3. Added --genome-ref option to specify reference genome FASTA path
   4. Added --no-shap option for skipping SHAP analysis
 """
-import os, sys, time
+import os, sys, time, subprocess
+import requests
 import click
 from rich.console import Console
 from rich.table import Table
@@ -251,46 +252,62 @@ def predict(chrom, pos, ref, alt, genome, fmt, no_ensemble,
             skip_shap = not click.confirm("\nGenerate SHAP analysis plot?", default=True)
         variant_info = f"{chrom}:{pos} {ref}>{alt}"
 
-        # ── Determine variant type ────────────────────────────────────
-        # Priority: CLI --ensemble-type > prediction["_variant_type"] > remote ANNOVAR API
-        effective_ensemble_type = ensemble_type
-        if not effective_ensemble_type:
-            effective_ensemble_type = prediction.get("_variant_type")
-        if not effective_ensemble_type:
-            with console.status("[bold cyan]Detecting variant type...[/bold cyan]"):
+        # ── Pre-computed fast path ────────────────────────────────
+        # Both remote (/predict/ returns prediction["precomputed"]) and local
+        # (predict_local returns prediction["precomputed"]) use the same key.
+        precomp = prediction.get("precomputed")
+        if precomp:
+            model_key = precomp.get("model_used", "MERGE_Precomputed")
+            ensemble = {
+                "success":        True,
+                "ensemble_results": {model_key: precomp},
+                "raw_features":   {},
+                "from_cache":     True,
+            }
+            prediction["_variant_type"] = precomp.get("variant_type")
+            console.print("[dim]⚡ Pre-computed score retrieved from cache[/dim]")
+        # ─────────────────────────────────────────────────────────
+        else:
+            # ── Determine variant type ────────────────────────────────────
+            # Priority: CLI --ensemble-type > prediction["_variant_type"] > remote ANNOVAR API
+            effective_ensemble_type = ensemble_type
+            if not effective_ensemble_type:
+                effective_ensemble_type = prediction.get("_variant_type")
+            if not effective_ensemble_type:
+                with console.status("[bold cyan]Detecting variant type...[/bold cyan]"):
+                    try:
+                        from . import local_engine as _le
+                        effective_ensemble_type = _le.get_variant_type(chrom, pos, ref, alt, genome)
+                    except Exception:
+                        effective_ensemble_type = None
+            if effective_ensemble_type:
+                prediction["_variant_type"] = effective_ensemble_type
+
+            with console.status("[bold cyan]Computing MERGE ensemble score...[/bold cyan]"):
                 try:
-                    from . import local_engine as _le
-                    effective_ensemble_type = _le.get_variant_type(chrom, pos, ref, alt, genome)
-                except Exception:
-                    effective_ensemble_type = None
-        if effective_ensemble_type:
-            prediction["_variant_type"] = effective_ensemble_type
-
-        with console.status("[bold cyan]Computing MERGE ensemble score...[/bold cyan]"):
-            try:
-                if is_local:
-                    from . import local_engine
-                    dbnsfp      = prediction.get("dbnsfp") or {}
-                    transcripts = dbnsfp.get("transcripts") or ([dbnsfp] if dbnsfp else [])
-                    ens_result, ens_err = local_engine.run_local_ensemble(
-                        prediction, transcripts, ensemble_type=effective_ensemble_type,
-                        skip_shap=skip_shap, variant_info=variant_info)
-                    if ens_err:
-                        console.print(f"[yellow]Ensemble model warning: {ens_err}[/yellow]")
-                    ensemble = {"success": True, **ens_result}
-                else:
-                    dbnsfp      = prediction.get("dbnsfp") or {}
-                    transcripts = [dbnsfp] if dbnsfp else []
-                    ens_resp = api.predict_ensemble(prediction, transcripts,
-                                                    ensemble_type=effective_ensemble_type)
-                    if ens_resp.get("success"):
-                        ensemble = ens_resp
-                        for _v in (ensemble.get("ensemble_results") or {}).values():
-                            if isinstance(_v, dict) and _v.get("variant_type"):
-                                prediction["_variant_type"] = _v["variant_type"]
-                                break
-            except Exception as e:
-                console.print(f"[yellow]Ensemble score failed (raw scores unaffected): {e}[/yellow]")
+                    if is_local:
+                        from . import local_engine
+                        dbnsfp      = prediction.get("dbnsfp") or {}
+                        transcripts = dbnsfp.get("transcripts") or ([dbnsfp] if dbnsfp else [])
+                        ens_result, ens_err = local_engine.run_local_ensemble(
+                            prediction, transcripts, ensemble_type=effective_ensemble_type,
+                            skip_shap=skip_shap, variant_info=variant_info)
+                        if ens_err:
+                            console.print(f"[yellow]Ensemble model warning: {ens_err}[/yellow]")
+                        ensemble = {"success": True, **ens_result}
+                    else:
+                        dbnsfp      = prediction.get("dbnsfp") or {}
+                        transcripts = [dbnsfp] if dbnsfp else []
+                        ens_resp = api.predict_ensemble(prediction, transcripts,
+                                                        ensemble_type=effective_ensemble_type)
+                        if ens_resp.get("success"):
+                            ensemble = ens_resp
+                            for _v in (ensemble.get("ensemble_results") or {}).values():
+                                if isinstance(_v, dict) and _v.get("variant_type"):
+                                    prediction["_variant_type"] = _v["variant_type"]
+                                    break
+                except Exception as e:
+                    console.print(f"[yellow]Ensemble score failed (raw scores unaffected): {e}[/yellow]")
 
     output.render_single(prediction, ensemble, fmt=fmt,
                          errors=prediction.get("errors"))
@@ -428,15 +445,33 @@ def local_setup():
 # ── merge local download ──────────────────────────────────────
 @local.command("download")
 @click.option("--file", "file_type",
-              type=click.Choice(["all", "dbnsfp", "gpn-msa", "popeve"]),
+              type=click.Choice(["all", "dbnsfp", "gpn-msa", "popeve", "precomputed"]),
               default="all",
-              help="Show download instructions for specific file")
-def local_download(file_type):
-    """Show official download URLs for pre-computed data files."""
+              help="Show download instructions for specific file (or auto-download precomputed VCF)")
+@click.option("--genome", default="all", type=click.Choice(["all", "hg38", "hg19"]),
+              help="Genome version for precomputed VCF download (default: both)")
+@click.option("--variant-type", "variant_type", default="all",
+              type=click.Choice(["all", "coding", "splicing"]),
+              help="Variant type for precomputed VCF download (default: both)")
+def local_download(file_type, genome, variant_type):
+    """Show official download URLs for pre-computed data files.
+
+    \b
+    The precomputed VCF cache files are auto-downloaded from merge.fanglab.cn:
+      merge local download --file precomputed              # download all 4 files
+      merge local download --file precomputed --genome hg38
+      merge local download --file precomputed --genome hg38 --variant-type coding
+    """
     cfg = get_local_config()
     data_dir = cfg["data_dir"]
     os.makedirs(data_dir, exist_ok=True)
 
+    # ── Auto-download precomputed VCF files ───────────────────────
+    if file_type == "precomputed":
+        _download_precomputed_vcfs(data_dir, genome, variant_type)
+        return
+    # ─────────────────────────────────────────────────────────────
+
     INSTRUCTIONS = {
         "dbnsfp": {
             "label": "dbNSFP (ESM1b + AlphaMissense)",
@@ -472,6 +507,8 @@ def local_download(file_type):
         "\n[bold green]✓ Ensemble models (pkl files)[/bold green] are bundled with pip install, "
         "no manual download needed.\n"
         "  If model files are missing, run: [bold]pip install --force-reinstall merge-cli[/bold]\n"
+        "\n[bold yellow]⚡ Pre-computed VCF cache[/bold yellow] can be auto-downloaded:\n"
+        "  [bold]merge local download --file precomputed[/bold]\n"
     )
 
     show_keys = (list(INSTRUCTIONS.keys()) if file_type == "all"
@@ -487,6 +524,129 @@ def local_download(file_type):
         console.print()
 
 
+def _download_precomputed_vcfs(data_dir: str, genome: str = "all", variant_type: str = "all"):
+    """Auto-download pre-computed VCF files + tabix indices from merge.fanglab.cn."""
+    import shutil
+
+    # Files hosted at: https://merge.fanglab.cn/static/precomputed/<filename>
+    # Adjust BASE_STATIC_URL if you move the files elsewhere (e.g. OSS / GitHub Releases)
+    BASE_STATIC_URL = f"{FIXED_API_URL}/static/precomputed"
+
+    ALL_FILES = {
+        ("hg38", "coding"):   "coding_merged.vcf.gz",
+        ("hg38", "splicing"): "splicing_merged.vcf.gz",
+        ("hg19", "coding"):   "coding_merged_hg19.vcf.gz",
+        ("hg19", "splicing"): "splicing_merged_hg19.vcf.gz",
+    }
+
+    # Filter by --genome and --variant-type
+    targets = {
+        (g, vt): fname for (g, vt), fname in ALL_FILES.items()
+        if (genome == "all" or g == genome)
+        and (variant_type == "all" or vt == variant_type)
+    }
+
+    if not targets:
+        console.print("[red]No files match the specified genome/variant-type combination.[/red]")
+        return
+
+    has_tabix = bool(shutil.which("tabix"))
+    if not has_tabix:
+        console.print(
+            "[yellow]⚠ tabix not found in PATH. VCF files will be downloaded but NOT indexed.\n"
+            "  Install htslib (conda install -c bioconda htslib) then run this command again\n"
+            "  or index manually: tabix -p vcf <file.vcf.gz>[/yellow]\n"
+        )
+
+    ok_count = 0
+    for (g, vt), fname in targets.items():
+        dest = os.path.join(data_dir, fname)
+        dest_tbi = dest + ".tbi"
+        url     = f"{BASE_STATIC_URL}/{fname}"
+        url_tbi = f"{BASE_STATIC_URL}/{fname}.tbi"
+
+        console.print(f"\n[bold cyan]── {g} {vt}: {fname} ──[/bold cyan]")
+
+        # ── Download .vcf.gz ──────────────────────────────────────
+        if os.path.exists(dest):
+            size_mb = os.path.getsize(dest) / 1e6
+            console.print(f"  [green]✓ Already exists ({size_mb:.0f} MB), skipping.[/green]")
+        else:
+            console.print(f"  Downloading {url} …")
+            success = _stream_download(url, dest)
+            if not success:
+                console.print(f"  [red]✗ Download failed. Check URL or network.[/red]")
+                continue
+
+        # ── Download .tbi (pre-built index) ──────────────────────
+        if os.path.exists(dest_tbi):
+            console.print(f"  [green]✓ Index already exists, skipping.[/green]")
+        else:
+            console.print(f"  Downloading index {url_tbi} …")
+            idx_ok = _stream_download(url_tbi, dest_tbi)
+            if not idx_ok:
+                # Fall back to local tabix
+                if has_tabix:
+                    console.print("  Pre-built index not found, building with local tabix…")
+                    result = subprocess.run(
+                        ["tabix", "-p", "vcf", dest],
+                        capture_output=True, text=True, timeout=300,
+                    )
+                    if result.returncode == 0:
+                        console.print("  [green]✓ Index built.[/green]")
+                    else:
+                        console.print(f"  [red]✗ tabix failed: {result.stderr.strip()}[/red]")
+                        continue
+                else:
+                    console.print(
+                        "  [red]✗ No pre-built index and tabix not found.\n"
+                        "     Install htslib and run: tabix -p vcf " + dest + "[/red]"
+                    )
+                    continue
+
+        console.print(f"  [green]✓ Ready: {dest}[/green]")
+        ok_count += 1
+
+    console.print(f"\n[bold green]✓ Done: {ok_count}/{len(targets)} files ready in {data_dir}[/bold green]")
+    if ok_count > 0:
+        console.print(
+            "  Pre-computed scores will now be used automatically for common variants.\n"
+            "  Run [bold]merge local status[/bold] to verify."
+        )
+
+
+def _stream_download(url: str, dest: str, chunk_size: int = 1 << 20) -> bool:
+    """Download url → dest with a progress bar. Returns True on success."""
+    import math
+    tmp = dest + ".part"
+    try:
+        r = requests.get(url, stream=True, timeout=30)
+        if r.status_code == 404:
+            return False
+        r.raise_for_status()
+        total = int(r.headers.get("content-length", 0))
+        downloaded = 0
+        with open(tmp, "wb") as f:
+            for chunk in r.iter_content(chunk_size=chunk_size):
+                if chunk:
+                    f.write(chunk)
+                    downloaded += len(chunk)
+                    if total:
+                        pct  = downloaded / total * 100
+                        done = int(pct / 5)
+                        bar  = "█" * done + "░" * (20 - done)
+                        mb   = downloaded / 1e6
+                        print(f"\r    [{bar}] {pct:5.1f}%  {mb:.0f} MB", end="", flush=True)
+        print()  # newline after progress bar
+        os.replace(tmp, dest)
+        return True
+    except Exception as exc:
+        if os.path.exists(tmp):
+            os.unlink(tmp)
+        console.print(f"\n  [red]Download error: {exc}[/red]")
+        return False
+
+
 # ── merge local predict ───────────────────────────────────────
 @local.command("predict")
 @click.option("--chrom",   required=True)

{merge_cli-3.5.0 → merge_cli-3.6}/merge_cli/data/models/ensemble_predict.py

@@ -122,9 +122,9 @@ sys.modules['__main__'].FeatureEngineer = FeatureEngineer
 _THIS_DIR = os.path.dirname(os.path.abspath(__file__))
 
 MODEL_PATHS = {
-    'ClinVar':               os.path.join(_THIS_DIR, 'BestModel_Clinvar.pkl'),
-    'Splice_ClinVar_GnomAD': os.path.join(_THIS_DIR, 'BestModel_Splice_Unsupervised Only.pkl'),
-    'NonCoding_ClinVar':     os.path.join(_THIS_DIR, 'BestModel_Clinvar-noncoding.pkl'),
+    'ClinVar':               os.path.join(_THIS_DIR, 'BestModel_coding.pkl'),
+    'Splice_ClinVar_GnomAD': os.path.join(_THIS_DIR, 'BestModel_splice.pkl'),
+    'NonCoding_ClinVar':     os.path.join(_THIS_DIR, 'BestModel_noncoding.pkl'),
 }
 
 for _name, _path in MODEL_PATHS.items():

{merge_cli-3.5.0 → merge_cli-3.6}/merge_cli/ensemble_predict.py

@@ -122,9 +122,9 @@ sys.modules['__main__'].FeatureEngineer = FeatureEngineer
 _THIS_DIR = os.path.dirname(os.path.abspath(__file__))
 
 MODEL_PATHS = {
-    'ClinVar':               os.path.join(_THIS_DIR, 'BestModel_Clinvar.pkl'),
-    'Splice_ClinVar_GnomAD': os.path.join(_THIS_DIR, 'BestModel_Splice_Unsupervised Only.pkl'),
-    'NonCoding_ClinVar':     os.path.join(_THIS_DIR, 'BestModel_Clinvar-noncoding.pkl'),
+    'ClinVar':               os.path.join(_THIS_DIR, 'BestModel_coding.pkl'),
+    'Splice_ClinVar_GnomAD': os.path.join(_THIS_DIR, 'BestModel_splice.pkl'),
+    'NonCoding_ClinVar':     os.path.join(_THIS_DIR, 'BestModel_noncoding.pkl'),
 }
 
 for _name, _path in MODEL_PATHS.items():

{merge_cli-3.5.0 → merge_cli-3.6}/merge_cli/local_engine.py

@@ -1251,11 +1251,190 @@ def run_local_ensemble(pred_data, all_transcripts, ensemble_type=None, skip_shap
     return {chosen_key: result}, None
 
 
+# ─── Pre-computed VCF Cache ───────────────────────────────────────
+
+# VCF filenames mirror the server-side PRECOMPUTED_VCFS config in views.py
+_PRECOMPUTED_VCF_NAMES = {
+    "hg38": {
+        "coding":    "coding_merged.vcf.gz",
+        "splicing":  "splicing_merged.vcf.gz",
+    },
+    "hg19": {
+        "coding":    "coding_merged_hg19.vcf.gz",
+        "splicing":  "splicing_merged_hg19.vcf.gz",
+    },
+}
+
+
+def query_precomputed_local(chrom: str, pos, ref: str, alt: str,
+                            genome: str = "hg38") -> Optional[dict]:
+    """
+    Query the local pre-computed VCF cache files (same format as the server).
+    Splicing is checked first to prevent coding-model mis-classification.
+
+    Returns the same dict shape as views.query_precomputed_vcf on hit, or
+    None on miss / unavailable.
+
+    INFO fields expected (set by the pre-computation pipeline):
+      ENSEMBLE  – MERGE score (0-1)
+      INTERP    – interpretation label (spaces encoded as underscores)
+      HYENA     – HyenaDNA score
+      NT        – Nucleotide Transformer score
+      GPN       – GPN-MSA score
+      POPEVE    – popEVE score
+      EVO2      – Evo2 LLR score
+      AM        – AlphaMissense score
+      ESM1B     – ESM1b score
+      AG_SPLICE           – AlphaGenome splicing composite (splice variants)
+      AG_RAW_MEAN / MAX / MIN         – AlphaGenome raw scores (coding)
+      AG_Q_MEAN / MAX / MIN           – AlphaGenome quantile scores (coding)
+    """
+    try:
+        import pysam
+    except ImportError:
+        return None  # pysam unavailable; remote fallback will be used
+
+    data_dir = _cfg()["data_dir"]
+    names = _PRECOMPUTED_VCF_NAMES.get(genome, {})
+
+    for vtype in ("splicing", "coding"):   # splicing first — mirrors server logic
+        fname = names.get(vtype)
+        if not fname:
+            continue
+        fpath = os.path.join(data_dir, fname)
+        if not os.path.exists(fpath):
+            continue
+
+        try:
+            tbx = pysam.TabixFile(fpath)
+            pos_int = int(pos)
+            chrom_queries = [chrom,
+                             ("chr" + chrom) if not chrom.startswith("chr") else chrom.lstrip("chr")]
+            records = []
+            for cq in chrom_queries:
+                try:
+                    records = list(tbx.fetch(cq, pos_int - 1, pos_int))
+                    if records:
+                        break
+                except Exception:
+                    continue
+            tbx.close()
+
+            for record in records:
+                parts = record.split("\t")
+                if len(parts) < 8 or parts[3] != ref or parts[4] != alt:
+                    continue
+
+                info: dict = {}
+                for item in parts[7].split(";"):
+                    if "=" in item:
+                        k, v = item.split("=", 1)
+                        info[k] = v
+
+                if "ENSEMBLE" not in info:
+                    continue
+
+                score_val  = _safe_float(info["ENSEMBLE"]) or 0.0
+                interp_val = info.get("INTERP", "Precomputed").replace("_", " ")
+
+                sub_models = {
+                    "alphagenome": {"statistics": {}},
+                    "hyenadna":    {"score": _safe_float(info.get("HYENA"))},
+                    "nt":          {"score": _safe_float(info.get("NT"))},
+                    "gpn_msa":     {"score": _safe_float(info.get("GPN"))},
+                    "popeve":      {"score": _safe_float(info.get("POPEVE"))},
+                    "evo2": {
+                        "llr_score":      _safe_float(info.get("EVO2")),
+                        "ref_score":      None,
+                        "var_score":      None,
+                        "context_length": 8192,
+                        "interpretation": "Precomputed cache",
+                        "score_class": (
+                            "benign"     if (_safe_float(info.get("EVO2")) or 0) > 0
+                            else "deleterious" if (_safe_float(info.get("EVO2")) or 0) < 0
+                            else "unknown"
+                        ),
+                    },
+                    "AlphaMissense": {"score": _safe_float(info.get("AM"))},
+                    "ESM1b":         {"score": _safe_float(info.get("ESM1B"))},
+                }
+
+                if vtype == "splicing":
+                    sub_models["alphagenome"]["statistics"]["alphagenome_splicing"] = \
+                        _safe_float(info.get("AG_SPLICE"))
+                else:
+                    sub_models["alphagenome"]["statistics"].update({
+                        "raw_score_mean":      _safe_float(info.get("AG_RAW_MEAN")),
+                        "raw_score_max":       _safe_float(info.get("AG_RAW_MAX")),
+                        "raw_score_min":       _safe_float(info.get("AG_RAW_MIN")),
+                        "quantile_score_mean": _safe_float(info.get("AG_Q_MEAN")),
+                        "quantile_score_max":  _safe_float(info.get("AG_Q_MAX")),
+                        "quantile_score_min":  _safe_float(info.get("AG_Q_MIN")),
+                    })
+
+                return {
+                    "ensemble": {
+                        "score":        score_val,
+                        "interpretation": {
+                            "label":  interp_val,
+                            "badge":  "⚡ Cache",
+                            "color":  "#ff6b6b" if score_val > 0.5 else "#6bcf7f",
+                        },
+                        "variant_type": "splice" if vtype == "splicing" else "coding",
+                        "model_used":   ("Splice_ClinVar_GnomAD"
+                                         if vtype == "splicing" else "ClinVar"),
+                        "from_cache":        True,
+                        "features_raw_aligned": {},
+                        "imputed_features":     {},
+                        "shap_plot":            None,
+                    },
+                    "sub_models": sub_models,
+                }
+
+        except Exception as exc:
+            logger.debug(f"[precomputed_local] {vtype}/{genome}: {exc}")
+            continue
+
+    return None   # cache miss
+
+
 # ─── Top-level Prediction Entry ───────────────────────────────────
 
 def predict_local(chrom, pos, ref, alt, genome="hg38", ensemble_type=None, **flags) -> dict:
     vtype = ensemble_type or get_variant_type(chrom, pos, ref, alt, genome)
 
+    # ── Pre-computed cache fast path ──────────────────────────────
+    cached = query_precomputed_local(chrom, pos, ref, alt, genome)
+    if cached:
+        logger.debug(f"[precomputed_local] cache hit: {chrom}:{pos} {ref}>{alt}")
+        ens    = cached["ensemble"]
+        subs   = cached["sub_models"]
+        return {
+            "success":   True,
+            "from_cache": True,
+            "prediction": {
+                "input":        {"chrom": chrom, "pos": pos, "ref": ref, "alt": alt},
+                "genome_version": genome,
+                "_variant_type": ens.get("variant_type", vtype),
+                "precomputed":   ens,
+                "dbnsfp": {
+                    "annotations": {},
+                    "dl_models":   {
+                        "AlphaMissense": subs.get("AlphaMissense", {}),
+                        "ESM1b":         subs.get("ESM1b", {}),
+                    },
+                },
+                "alphagenome": subs.get("alphagenome"),
+                "hyenadna":    subs.get("hyenadna"),
+                "nt":          subs.get("nt"),
+                "gpn_msa":     subs.get("gpn_msa"),
+                "popeve":      subs.get("popeve"),
+                "evo2":        subs.get("evo2"),
+                "errors":      {},
+            },
+        }
+    # ─────────────────────────────────────────────────────────────
+
     tasks = {
         "dbnsfp":   lambda: query_dbnsfp_local(chrom, pos, ref, alt, genome),
         "hyenadna": lambda: call_local_service("hyenadna", chrom, pos, ref, alt, genome),
@@ -1317,6 +1496,42 @@ def predict_local_batch(vcf_path: str, genome: str = "hg38",
         _log(f"  [{i}/{len(variants)}] {chrom}:{pos} {ref}>{alt}…")
         try:
             vtype = get_variant_type(chrom, pos, ref, alt, genome, ensemble_type)
+
+            # ── Pre-computed cache fast path ──────────────────────
+            cached = query_precomputed_local(chrom, pos, ref, alt, genome)
+            if cached:
+                _log(f"    ⚡ Pre-computed cache hit")
+                ens  = cached["ensemble"]
+                subs = cached["sub_models"]
+                ag_stats = (subs.get("alphagenome") or {}).get("statistics", {})
+                row = {
+                    "chrom": chrom, "pos": pos, "ref": ref, "alt": alt,
+                    "genome":         genome,
+                    "variant_type":   ens.get("variant_type", vtype),
+                    "gene":           "-",
+                    "transcript":     "-",
+                    "merge_ensemble_score": ens.get("score"),
+                    "merge_label":    (ens.get("interpretation") or {}).get("label"),
+                    "from_cache":     True,
+                    "alphamissense_score": (subs.get("AlphaMissense") or {}).get("score"),
+                    "esm1b_score":         (subs.get("ESM1b") or {}).get("score"),
+                    "gpn_msa_score":  (subs.get("gpn_msa") or {}).get("score"),
+                    "popeve_score":   (subs.get("popeve") or {}).get("score"),
+                    "hyenadna_score": (subs.get("hyenadna") or {}).get("score"),
+                    "nt_score":       (subs.get("nt") or {}).get("score"),
+                    "evo2_score":     (subs.get("evo2") or {}).get("llr_score"),
+                    "alphagenome_raw_max":       ag_stats.get("raw_score_max"),
+                    "alphagenome_raw_min":       ag_stats.get("raw_score_min"),
+                    "alphagenome_raw_mean":      ag_stats.get("raw_score_mean"),
+                    "alphagenome_quantile_max":  ag_stats.get("quantile_score_max"),
+                    "alphagenome_quantile_min":  ag_stats.get("quantile_score_min"),
+                    "alphagenome_quantile_mean": ag_stats.get("quantile_score_mean"),
+                    "alphagenome_splicing":      ag_stats.get("alphagenome_splicing"),
+                    "errors": "",
+                }
+                results_rows.append(row)
+                continue
+            # ─────────────────────────────────────────────────────
             resp  = predict_local(chrom, pos, ref, alt, genome,
                                   local_genome_path=None, ensemble_type=vtype, **model_flags)
             pred  = resp.get("prediction", {})
@@ -1399,6 +1614,12 @@ def check_local_files() -> dict:
         "dbNSFP (hg19)": os.path.join(data_dir, "dbNSFP5.3a_grch37.gz"),
         "GPN-MSA":       os.path.join(data_dir, "scores.tsv.bgz"),
         "popEVE":        os.path.join(data_dir, "grch38_popEVE_ukbb_20250715.vcf.gz"),
+        # ── Pre-computed VCF cache ──────────────────────────────────
+        "Pre-computed coding (hg38)":    os.path.join(data_dir, "coding_merged.vcf.gz"),
+        "Pre-computed splicing (hg38)":  os.path.join(data_dir, "splicing_merged.vcf.gz"),
+        "Pre-computed coding (hg19)":    os.path.join(data_dir, "coding_merged_hg19.vcf.gz"),
+        "Pre-computed splicing (hg19)":  os.path.join(data_dir, "splicing_merged_hg19.vcf.gz"),
+        # ───────────────────────────────────────────────────────────
         f"Evo2 local weights ({evo2_model_name}, optional)": evo2_weight_path,
     }
     result = {}

{merge_cli-3.5.0 → merge_cli-3.6}/merge_cli/output.py

@@ -62,11 +62,13 @@ def _extract_flat(prediction: dict, ensemble=None) -> dict:
 
     ens_score = ens_label = ens_model = ens_variant_type = None
     imputed_feats = {}
+    _from_cache = False
 
     if ensemble:
         ens_results = ensemble.get("ensemble_results") or {
             k: v for k, v in ensemble.items() if k not in ("success", "error")
         }
+        _from_cache = bool(ensemble.get("from_cache"))
         for _key, _val in ens_results.items():
             if isinstance(_val, dict):
                 ens_score        = _val.get("score")
@@ -74,6 +76,8 @@ def _extract_flat(prediction: dict, ensemble=None) -> dict:
                 ens_model        = _key
                 ens_variant_type = _val.get("variant_type")
                 imputed_feats    = _val.get("imputed_features", {})
+                if _val.get("from_cache"):
+                    _from_cache = True
                 break
 
     def _get_val(raw_val, imputed_key):
@@ -115,6 +119,7 @@ def _extract_flat(prediction: dict, ensemble=None) -> dict:
         "AG_raw_mean":     _ag("alphagenome_raw_score_mean",     "raw_score_mean"),
         "AG_quantile_mean":_ag("alphagenome_quantile_score_mean","quantile_score_mean"),
         "AG_Splicing":     _ag("alphagenome_splicing",            "alphagenome_splicing"),
+        "from_cache":      _from_cache,
     }
 
 
@@ -172,6 +177,17 @@ def render_single(prediction: dict, ensemble=None, fmt="table", errors=None) ->
         vt_style = {"coding": "bold magenta", "noncoding": "bold blue", "splice": "bold yellow"}.get(str(vt).lower(), "white")
         t.add_row("Variant Type", Text(str(vt), style=vt_style))
 
+    # Cache indicator
+    _is_cached = (
+        (ensemble or {}).get("from_cache") or
+        any(
+            isinstance(v, dict) and v.get("from_cache")
+            for v in ((ensemble or {}).get("ensemble_results") or {}).values()
+        )
+    )
+    if _is_cached:
+        t.add_row("Score Source", Text("⚡ Pre-computed cache", style="bold yellow"))
+
     # MERGE ensemble
     t.add_section()
     t.add_row("MERGE Pathogenicity", _score_text(flat["MERGE_Score"], flat["MERGE_Label"]))

{merge_cli-3.5.0 → merge_cli-3.6}/merge_cli.egg-info/PKG-INFO

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: merge-cli
-Version: 3.5.0
-Summary: MERGE 变异致病性预测 CLI（服务器固定，集成模型内嵌，无需手动配置）
+Version: 3.6
+Summary: MERGE variant pathogenicity prediction CLI (fixed server, integrated model embedded, no manual configuration required)
 Project-URL: Homepage, https://merge.fanglab.cn
 Requires-Python: >=3.11
 Requires-Dist: click>=8.1

{merge_cli-3.5.0 → merge_cli-3.6}/pyproject.toml

@@ -4,8 +4,8 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "merge-cli"
-version = "3.5.0"
-description = "MERGE 变异致病性预测 CLI（服务器固定，集成模型内嵌，无需手动配置）"
+version = "3.6"
+description = "MERGE variant pathogenicity prediction CLI (fixed server, integrated model embedded, no manual configuration required)"
 requires-python = ">=3.11"
 dependencies = [
     # 核心 CLI 框架