@hongmaple0820/med-scale-research-os 0.43.0

package/target-research-platform/src/output/manuscript_generator.py

@@ -0,0 +1,1189 @@
+"""
+期刊论文草稿生成器（IMRaD）
+
+基于四阶段 pipeline 已验证的 JSON 结果合成**中文**研究论文草稿。
+仅引用溯源数据与 PubMed 元数据，不编造实验结果。
+"""
+from __future__ import annotations
+
+import json
+import statistics
+from datetime import datetime
+from pathlib import Path
+from typing import Any, Optional
+
+from .chart_generator import ChartGenerator
+from .delivery_packager import create_delivery_package, write_delivery_manifest
+from .journal_templates import render_zhonghua_oncology
+from .literature_enricher import enrich_papers_for_manuscript, write_literature_evidence_pack
+from .manuscript_fact_checker import validate_manuscript_facts
+from .manuscript_quality import validate_manuscript_delivery
+from .literature_reviewer import review_literature_for_manuscript
+from .medical_skills_bridge import drug_target_evidence_framing
+
+
+_DISEASE_ZH: dict[str, str] = {
+    "breast cancer": "乳腺癌",
+    "lung cancer": "肺癌",
+    "colorectal cancer": "结直肠癌",
+    "prostate cancer": "前列腺癌",
+    "melanoma": "黑色素瘤",
+}
+
+_TISSUE_ZH: dict[str, str] = {
+    "Cells_Cultured_fibroblasts": "培养成纤维细胞",
+    "Cells_EBV-transformed_lymphocytes": "EBV 转化淋巴细胞",
+    "Ovary": "卵巢",
+    "Uterus": "子宫",
+    "Cervix_Endocervix": "宫颈内口",
+    "Cervix_Ectocervix": "宫颈外口",
+    "Fallopian_Tube": "输卵管",
+    "Adrenal_Gland": "肾上腺",
+    "Nerve_Tibial": "胫神经",
+    "Bladder": "膀胱",
+    "Lung": "肺",
+    "Breast": "乳腺",
+    "Liver": "肝脏",
+    "Kidney": "肾脏",
+    "Heart": "心脏",
+    "Brain": "脑",
+    "Thyroid": "甲状腺",
+    "Pancreas": "胰腺",
+    "Spleen": "脾脏",
+    "Colon": "结肠",
+    "Skin": "皮肤",
+}
+
+_PHASE_ZH: dict[str, str] = {
+    "Phase 1": "I 期",
+    "Phase 2": "II 期",
+    "Phase 3": "III 期",
+    "Phase 4": "IV 期",
+    "Not Applicable": "不适用/观察性",
+}
+
+
+class ManuscriptGenerator:
+    """从 pipeline 结果生成期刊论文 Markdown / DOCX 草稿（默认中文正文）。"""
+
+    def __init__(self, output_dir: str):
+        self.output_dir = Path(output_dir)
+        self.output_dir.mkdir(parents=True, exist_ok=True)
+
+    def generate(
+        self,
+        pipeline_results: dict,
+        target: str,
+        disease: str,
+        *,
+        language: str = "zh",
+    ) -> dict:
+        stages = pipeline_results.get("stages", pipeline_results)
+        s1 = stages.get("stage1", {}) or {}
+        s2 = stages.get("stage2", {}) or {}
+        s3 = stages.get("stage3", {}) or {}
+        s4 = stages.get("stage4", {}) or {}
+
+        context = self._build_context(target, disease, s1, s2, s3, s4)
+        slug = f"{target}_{disease}".replace(" ", "_")
+
+        enrich_result = enrich_papers_for_manuscript(
+            context.get("papers_raw", []),
+            target=target,
+            disease_zh=context.get("disease_zh", disease),
+        )
+        enriched_papers = enrich_result.get("papers", [])
+        context["papers_enriched"] = enriched_papers
+        context["literature_enrichment"] = enrich_result.get("enrichment", {})
+        evidence_pack_path = self.output_dir / f"{slug}_literature_evidence_pack.json"
+        evidence_pack = write_literature_evidence_pack(
+            enriched_papers,
+            evidence_pack_path,
+            target=target,
+            disease=context.get("disease_zh", disease),
+            enrichment_meta=context["literature_enrichment"],
+        )
+        context["literature_evidence_pack"] = {
+            "path": str(evidence_pack_path.resolve()),
+            "paper_count": evidence_pack.get("paper_count", 0),
+            "source_summary": evidence_pack.get("source_summary", {}),
+        }
+        context["literature_evidence_framing"] = drug_target_evidence_framing(
+            target,
+            context.get("disease_zh", disease),
+            int(context.get("pubmed_total") or len(enriched_papers)),
+        )
+        if enriched_papers:
+            context["references"] = self._format_references(enriched_papers[:25])
+            context["literature_highlights"] = self._build_literature_highlights(enriched_papers[:12])
+
+        lit_result = review_literature_for_manuscript(
+            enriched_papers,
+            target=target,
+            disease_zh=context.get("disease_zh", disease),
+            enrichment_meta=enrich_result.get("enrichment"),
+            output_dir=str(self.output_dir),
+            slug=slug,
+        )
+        context["agent_review_tasks"] = lit_result.get("agent_review_tasks")
+        context["agent_cache_expected"] = lit_result.get("agent_cache_expected")
+        context["literature_llm_source"] = lit_result.get("llm_source")
+        context["literature_reviews"] = lit_result.get("reviews", [])
+        context["usage_envelopes"] = lit_result.get("usage_envelopes", [])
+        context["usage_summary"] = lit_result.get("usage_summary", {})
+        context["literature_review_sources"] = self._summarize_review_sources(
+            context["literature_reviews"]
+        )
+        context["literature_llm_used"] = lit_result.get("llm_used", False)
+        context["literature_llm_model"] = lit_result.get("llm_model")
+        context["literature_llm_available"] = lit_result.get("llm_available", False)
+        context["figures"] = self._generate_figures(context, s1)
+
+        markdown = self._render_markdown_zh(context) if language == "zh" else self._render_markdown_en(context)
+
+        md_path = self.output_dir / f"{slug}_manuscript.md"
+        zhongliu_path = self.output_dir / f"{slug}_manuscript_zhongliu.md"
+        json_path = self.output_dir / f"{slug}_manuscript_meta.json"
+        fact_check_path = self.output_dir / f"{slug}_manuscript_fact_check.json"
+        quality_path = self.output_dir / f"{slug}_manuscript_quality.json"
+        delivery_manifest_path = self.output_dir / f"{slug}_delivery_manifest.json"
+        delivery_package_path = self.output_dir.parent / "deliveries" / f"{slug}_delivery.zip"
+        agent_review_tasks_path = context.get("agent_review_tasks")
+        agent_reviews_path = context.get("agent_cache_expected")
+        if agent_reviews_path and Path(agent_reviews_path).is_file():
+            agent_reviews_path = str(Path(agent_reviews_path).resolve())
+        else:
+            agent_reviews_path = None
+        md_path.write_text(markdown, encoding="utf-8")
+        zhongliu_md = render_zhonghua_oncology(context, markdown, figures=context.get("figures"))
+        zhongliu_path.write_text(zhongliu_md, encoding="utf-8")
+        json_path.write_text(
+            json.dumps(context, ensure_ascii=False, indent=2, default=str),
+            encoding="utf-8",
+        )
+
+        docx_path: str | None = None
+        docx_error: str | None = None
+        try:
+            docx_path = self._render_docx_zh(context, slug)
+        except Exception as exc:
+            docx_error = str(exc)
+
+        char_count = len(markdown)
+        result = {
+            "success": True,
+            "target": target,
+            "disease": disease,
+            "language": language,
+            "disclaimer": (
+                "本文件为基于公开数据库的计算分析研究论文草稿，需研究者补充实验设计、"
+                "伦理审批、作者信息与目标期刊格式后方能投稿。"
+            ),
+            "paths": {
+                "markdown": str(md_path.resolve()),
+                "zhongliu": str(zhongliu_path.resolve()),
+                "metadata": str(json_path.resolve()),
+                "fact_check": str(fact_check_path.resolve()),
+                "docx": docx_path,
+                "figures_dir": str((self.output_dir / "figures").resolve()),
+                "literature_evidence_pack": str(evidence_pack_path.resolve()),
+                "agent_review_tasks": agent_review_tasks_path,
+                "agent_reviews": agent_reviews_path,
+                "delivery_manifest": str(delivery_manifest_path.resolve()),
+                "delivery_package": str(delivery_package_path.resolve()),
+            },
+            "figures": context.get("figures", []),
+            "data_sources": context.get("data_sources", []),
+            "literature_reviews": {
+                "count": len(context.get("literature_reviews", [])),
+                "source_counts": context.get("literature_review_sources", {}),
+                "llm_used": context.get("literature_llm_used", False),
+                "llm_available": context.get("literature_llm_available", False),
+                "llm_model": context.get("literature_llm_model"),
+                "enrichment": context.get("literature_enrichment", {}),
+                "usage_summary": context.get("usage_summary", {}),
+            },
+            "docx_error": docx_error,
+            "reference_count": len(context.get("references", [])),
+            "word_estimate": len(markdown.split()),
+            "char_count": char_count,
+            "zhongliu_char_count": len(zhongliu_md),
+        }
+        fact_check = validate_manuscript_facts(result)
+        fact_check_path.write_text(
+            json.dumps(fact_check, ensure_ascii=False, indent=2, default=str),
+            encoding="utf-8",
+        )
+        result["fact_check"] = fact_check
+        result["delivery_manifest"] = str(delivery_manifest_path.resolve())
+        write_delivery_manifest(result, delivery_manifest_path, base_dir=self.output_dir.parent)
+
+        quality = validate_manuscript_delivery(result)
+        quality_path.write_text(
+            json.dumps(quality, ensure_ascii=False, indent=2, default=str),
+            encoding="utf-8",
+        )
+        result["paths"]["quality"] = str(quality_path.resolve())
+        result["quality_gate"] = quality
+        write_delivery_manifest(result, delivery_manifest_path, base_dir=self.output_dir.parent)
+        result["delivery_package"] = create_delivery_package(
+            result,
+            delivery_package_path,
+            base_dir=self.output_dir.parent,
+        )
+        return result
+
+    @staticmethod
+    def _summarize_review_sources(reviews: list[dict]) -> dict[str, int]:
+        counts: dict[str, int] = {}
+        for review in reviews or []:
+            source = str(review.get("source") or "unknown")
+            counts[source] = counts.get(source, 0) + 1
+        return counts
+
+    def _build_context(
+        self,
+        target: str,
+        disease: str,
+        s1: dict,
+        s2: dict,
+        s3: dict,
+        s4: dict,
+    ) -> dict[str, Any]:
+        s1_summary = s1.get("summary", s1) or {}
+        s2_summary = s2.get("summary", s2) or {}
+        s3_summary = s3.get("summary", s3) or {}
+        s4_summary = s4.get("summary", s4) or {}
+
+        disease_zh = _DISEASE_ZH.get(disease.lower(), disease)
+
+        # --- Stage1: 表达与生存 ---
+        te_stage = (s1.get("stages", {}) or {}).get("tissue_expression", {}) or {}
+        gtex_block = te_stage.get("gtex", {}) or {}
+        gtex_data = gtex_block.get("data", gtex_block) if isinstance(gtex_block, dict) else {}
+        tissues_raw = gtex_data.get("tissues", {}) if isinstance(gtex_data, dict) else {}
+        top_tissues = sorted(
+            ((k, float(v)) for k, v in tissues_raw.items() if v is not None),
+            key=lambda x: x[1],
+            reverse=True,
+        )[:15]
+
+        tu_stage = (s1.get("stages", {}) or {}).get("tumor_expression", {}) or {}
+        tumor_stats: dict[str, Any] = {}
+        tumor_sample_values: list[float] = []
+        for cancer, block in (tu_stage.get("cancer_results") or {}).items():
+            if not isinstance(block, dict) or not block.get("success"):
+                continue
+            samples = (block.get("data") or {}).get("samples") or []
+            values = [float(s.get("expression_value", 0)) for s in samples if s.get("expression_value") is not None]
+            if values:
+                tumor_stats[cancer] = {
+                    "n_samples": len(values),
+                    "mean": round(statistics.mean(values), 2),
+                    "median": round(statistics.median(values), 2),
+                    "max": round(max(values), 2),
+                    "min": round(min(values), 2),
+                }
+                if cancer == "BRCA" or not tumor_sample_values:
+                    tumor_sample_values = values
+
+        survival: dict[str, Any] = {}
+        sa_stage = (s1.get("stages", {}) or {}).get("survival_analysis", {}) or {}
+        for cancer, block in (sa_stage.get("cancer_results") or {}).items():
+            if isinstance(block, dict) and block.get("success"):
+                survival[cancer] = block.get("data", {})
+
+        # --- Stage2: 筛选与成药性 ---
+        de_stage = (s2.get("stages", {}) or {}).get("differential_expression", {}) or {}
+        de_source = de_stage.get("fallback_source") or de_stage.get("accession") or "TCGA/GEO"
+        de_method = de_stage.get("method", "标准差异表达分析")
+
+        top_targets = (
+            s2_summary.get("top_targets")
+            or s2_summary.get("top_candidates")
+            or []
+        )
+        target_rank = next(
+            (i + 1 for i, t in enumerate(top_targets) if t.get("gene") == target),
+            None,
+        )
+
+        drug_stage = (s2.get("stages", {}) or {}).get("druggability", {}) or {}
+        drug_targets = drug_stage.get("targets", {}) if isinstance(drug_stage.get("targets"), dict) else {}
+        target_drug = drug_targets.get(target, {}) if isinstance(drug_targets, dict) else {}
+        target_druggability = (
+            target_drug.get("composite_score")
+            or next((t.get("druggability_score") or t.get("score") for t in top_targets if t.get("gene") == target), None)
+        )
+        drug_class = target_drug.get("druggability_class", "")
+        drug_evidence = target_drug.get("evidence", []) or []
+        drug_interpretation = target_drug.get("interpretation", "")
+
+        # --- Stage3: 文献、临床、竞争 ---
+        developability = s3_summary.get("developability", {}) or {}
+        recommendations = s3_summary.get("recommendations", []) or []
+
+        cp_stage = (s3.get("stages", {}) or {}).get("clinical_pipeline", {}) or {}
+        cp_data = cp_stage.get("data", {}) if isinstance(cp_stage.get("data"), dict) else {}
+
+        comp_stage = (s3.get("stages", {}) or {}).get("competitive_landscape", {}) or {}
+        comp_data = comp_stage.get("data", {}) if isinstance(comp_stage.get("data"), dict) else {}
+        comp_sections = comp_data.get("sections", comp_data) if isinstance(comp_data, dict) else {}
+        competition = (
+            comp_sections.get("competitive_landscape")
+            or comp_sections.get("competitive_analysis")
+            or comp_sections
+            or {}
+        )
+        if not competition.get("competition_intensity"):
+            assess_stage = (s3.get("stages", {}) or {}).get("assessment", {}) or {}
+            assess_data = assess_stage.get("data", {}) if isinstance(assess_stage.get("data"), dict) else {}
+            for pain in assess_data.get("pain_points", []) or []:
+                if pain.get("type") == "competition":
+                    competition.setdefault("competition_intensity", "high" if pain.get("severity") == "high" else pain.get("severity", "—"))
+                    competition.setdefault("competition_note", pain.get("description", ""))
+                    break
+
+        clinical_summary = {
+            "total_trials": cp_data.get("total_trials"),
+            "active_trials": cp_data.get("active_trials"),
+            "phase_distribution": cp_data.get("phase_distribution", {}),
+            "top_players": cp_data.get("top_players") or cp_data.get("top_sponsors"),
+            "status_distribution": comp_sections.get("clinical_pipeline", {}).get("status_distribution")
+            if isinstance(comp_sections.get("clinical_pipeline"), dict)
+            else cp_data.get("status_distribution"),
+        }
+        if not clinical_summary["total_trials"] and isinstance(comp_sections.get("clinical_pipeline"), dict):
+            cp_sec = comp_sections["clinical_pipeline"]
+            clinical_summary.update({
+                "total_trials": cp_sec.get("total_trials"),
+                "active_trials": cp_sec.get("active_trials"),
+                "phase_distribution": cp_sec.get("phase_distribution", {}),
+            })
+
+        tf_stage = (s3.get("stages", {}) or {}).get("target_function", {}) or {}
+        tf_data = tf_stage.get("data", {}) if isinstance(tf_stage.get("data"), dict) else {}
+        tf_sections = tf_data.get("sections", {}) if isinstance(tf_data, dict) else {}
+        basic_info = tf_sections.get("basic_info", {}) or {}
+        function_info = tf_sections.get("function", {}) or {}
+        protein_function = (
+            function_info.get("function_description")
+            or function_info.get("description")
+            or (function_info.get("functions", [""])[0] if function_info.get("functions") else "")
+        )
+
+        lit_stage = (s3.get("stages", {}) or {}).get("literature", {}) or {}
+        lit_data = lit_stage.get("data", {}) if isinstance(lit_stage.get("data"), dict) else {}
+        papers = lit_data.get("papers", []) or []
+        references = self._format_references(papers[:25])
+        literature_highlights = self._build_literature_highlights(papers[:12])
+
+        # --- Stage4: 里程碑与商业 ---
+        milestones_summary = s4_summary.get("milestones", {}) or {}
+        ms_stage = (s4.get("stages", {}) or {}).get("milestones", {}) or {}
+        ms_data = ms_stage.get("data", {}) if isinstance(ms_stage.get("data"), dict) else {}
+        adj = (ms_data.get("stage_adjustments") or {}).get("adjustment_details", {}) or {}
+        competitive_adj = adj.get("competitive_landscape", {}) or {}
+        milestone_rows = ms_data.get("milestones", []) or []
+
+        funding = s4_summary.get("funding", {}) or {}
+
+        return {
+            "generated_at": datetime.now().isoformat(),
+            "target": target,
+            "disease": disease,
+            "disease_zh": disease_zh,
+            "title_zh": (
+                f"基于多组学与临床注册数据的 {target} 靶点在{disease_zh}中的"
+                f"成药性评估与研发路径分析"
+            ),
+            "title_en": (
+                f"Integrative druggability and development-path analysis of {target} "
+                f"in {disease} using multi-omics and clinical registry data"
+            ),
+            "data_sources": sorted(
+                set(
+                    (s1_summary.get("data_sources") or [])
+                    + (s2_summary.get("data_sources") or [])
+                    + (s3_summary.get("data_sources") or [])
+                )
+            ),
+            "cancer_types": s1_summary.get("cancer_types") or s1.get("cancer_types") or [],
+            "confidence": s1_summary.get("confidence_level", "medium"),
+            "key_findings_s1": s1_summary.get("key_findings", []),
+            "key_findings_s2": s2_summary.get("key_findings", []),
+            "key_findings_s3": s3_summary.get("key_findings", []),
+            "top_tissues": top_tissues,
+            "tissue_count": len(tissues_raw),
+            "tumor_stats": tumor_stats,
+            "survival": survival,
+            "de_source": de_source,
+            "de_method": de_method,
+            "target_rank": target_rank,
+            "target_druggability": target_druggability,
+            "drug_class": drug_class,
+            "drug_evidence": drug_evidence,
+            "drug_interpretation": drug_interpretation,
+            "top_targets": top_targets[:15],
+            "developability": developability,
+            "recommendations": recommendations,
+            "clinical_summary": clinical_summary,
+            "competition": competition or competitive_adj,
+            "basic_info": basic_info,
+            "protein_function": protein_function,
+            "milestones": {
+                "total_duration": milestones_summary.get("total_duration") or ms_data.get("total_duration_months"),
+                "total_cost": milestones_summary.get("total_cost") or ms_data.get("total_cost_estimate"),
+                "success_rate": milestones_summary.get("success_rate") or ms_data.get("overall_success_rate"),
+                "estimated_completion": ms_data.get("estimated_completion"),
+                "modality": s4.get("modality", "small_molecule"),
+                "risk_level": s4.get("risk_level", "medium"),
+            },
+            "milestone_rows": milestone_rows,
+            "funding": funding,
+            "stage4_druggability": adj.get("druggability_score"),
+            "references": references,
+            "literature_highlights": literature_highlights,
+            "papers_raw": papers,
+            "tissues_full": {k: float(v) for k, v in tissues_raw.items() if v is not None},
+            "tumor_sample_values": tumor_sample_values,
+            "pubmed_total": (lit_data.get("search_results") or {}).get("total_count"),
+        }
+
+    @staticmethod
+    def _build_literature_highlights(papers: list[dict]) -> list[dict]:
+        highlights = []
+        for paper in papers:
+            title = paper.get("title", "")
+            highlights.append({
+                "title": title,
+                "title_zh_note": title,
+                "year": paper.get("year", ""),
+                "journal": paper.get("journal", ""),
+                "pmid": paper.get("pmid", ""),
+                "theme": ManuscriptGenerator._infer_paper_theme(title),
+                "study_design": paper.get("study_design", ""),
+                "mesh_terms": (paper.get("mesh_terms") or [])[:4],
+            })
+        return highlights
+
+    @staticmethod
+    def _infer_paper_theme(title: str) -> str:
+        t = title.lower()
+        if "resistance" in t or "耐药" in t:
+            return "耐药机制"
+        if "cdk4/6" in t or "cdk4/6" in t:
+            return "CDK4/6 抑制剂"
+        if "selective" in t and "cdk4" in t:
+            return "CDK4 选择性抑制"
+        if "biomarker" in t or "predict" in t:
+            return "生物标志物预测"
+        if "adjuvant" in t or "metastatic" in t:
+            return "临床疗效与试验设计"
+        if "review" in t or "systematic" in t:
+            return "循证综述"
+        return "机制与转化研究"
+
+    @staticmethod
+    def _format_references(papers: list[dict]) -> list[dict]:
+        refs = []
+        for idx, paper in enumerate(papers, start=1):
+            authors = paper.get("authors") or []
+            author_str = ", ".join(authors[:3])
+            if len(authors) > 3:
+                author_str += ", 等"
+            refs.append({
+                "index": idx,
+                "pmid": paper.get("pmid", ""),
+                "citation_zh": (
+                    f"{author_str}. {paper.get('title', '')}[J]. "
+                    f"{paper.get('journal', '')}, {paper.get('year', 'n.d.')}."
+                ),
+                "citation_en": (
+                    f"{author_str} ({paper.get('year', 'n.d.')}). "
+                    f"{paper.get('title', 'Untitled')}. {paper.get('journal', '')}."
+                ),
+                "doi": paper.get("doi", ""),
+                "url": paper.get("url") or f"https://pubmed.ncbi.nlm.nih.gov/{paper.get('pmid', '')}/",
+            })
+        return refs
+
+    def _tissue_zh(self, name: str) -> str:
+        return _TISSUE_ZH.get(name, name.replace("_", " "))
+
+    @staticmethod
+    def _level_zh(value: Any) -> str:
+        mapping = {"high": "高", "medium": "中", "low": "低", "emerging": "新兴"}
+        if value is None:
+            return "—"
+        return mapping.get(str(value).lower(), str(value))
+
+    def _generate_figures(self, ctx: dict[str, Any], s1: dict) -> list[dict[str, str]]:
+        """生成论文嵌入图表，返回相对路径元数据。"""
+        figures_dir = self.output_dir / "figures"
+        charts = ChartGenerator(str(figures_dir))
+        target = ctx["target"]
+        disease_zh = ctx.get("disease_zh", ctx["disease"])
+        out: list[dict[str, str]] = []
+
+        tissues = ctx.get("tissues_full") or dict(ctx.get("top_tissues") or [])
+        if tissues:
+            paths = charts.tissue_expression_heatmap(
+                tissues, target, top_n=20, tissue_labels_zh=_TISSUE_ZH
+            )
+            if paths.get("png"):
+                out.append({
+                    "id": "fig1",
+                    "path": paths["png"],
+                    "relative": "figures/" + Path(paths["png"]).name,
+                    "title_zh": f"{target} 正常组织表达热图（GTEx）",
+                    "title_en": f"Heatmap of {target} expression across GTEx tissues",
+                })
+
+        samples = ctx.get("tumor_sample_values") or []
+        if samples:
+            paths = charts.tumor_expression_distribution(samples, target, "BRCA")
+            if paths.get("png"):
+                out.append({
+                    "id": "fig2",
+                    "path": paths["png"],
+                    "relative": "figures/" + Path(paths["png"]).name,
+                    "title_zh": f"TCGA-BRCA {target} 肿瘤表达分布",
+                    "title_en": f"Distribution of {target} expression in TCGA-BRCA",
+                })
+
+        clin = ctx.get("clinical_summary") or {}
+        phase_dist = clin.get("phase_distribution") or {}
+        if phase_dist:
+            paths = charts.clinical_trial_sankey(
+                phase_dist,
+                clin.get("status_distribution"),
+                target,
+                disease_zh,
+                total_trials=clin.get("total_trials"),
+            )
+            if paths.get("png"):
+                out.append({
+                    "id": "fig3",
+                    "path": paths["png"],
+                    "relative": "figures/" + Path(paths["png"]).name,
+                    "title_zh": f"{target}/{disease_zh} 临床试验阶段与状态流程图",
+                    "title_en": f"Clinical trial phase-status flow for {target} in {disease_zh}",
+                })
+
+        top_targets = ctx.get("top_targets") or []
+        if top_targets:
+            paths = charts.composite_ranking_bar(
+                top_targets,
+                title=f"{disease_zh} 候选靶点综合排名",
+            )
+            if paths.get("png"):
+                out.append({
+                    "id": "fig4",
+                    "path": paths["png"],
+                    "relative": "figures/" + Path(paths["png"]).name,
+                    "title_zh": "候选靶点成药性综合排名",
+                    "title_en": "Composite ranking of candidate targets",
+                })
+
+        return out
+
+    @staticmethod
+    def _embed_figure(fig: dict[str, str] | None) -> str:
+        if not fig or not fig.get("relative"):
+            return ""
+        num = fig.get("id", "fig").replace("fig", "")
+        title = fig.get("title_zh", "")
+        return f"\n\n![{title}]({fig['relative']})\n\n**图{num}　{title}**\n"
+
+    def _fig(self, ctx: dict[str, Any], fig_id: str) -> dict[str, str] | None:
+        for fig in ctx.get("figures") or []:
+            if fig.get("id") == fig_id:
+                return fig
+        return None
+
+    def _render_markdown_zh(self, ctx: dict[str, Any]) -> str:
+        clin = ctx.get("clinical_summary") or {}
+        dev = ctx.get("developability") or {}
+        ms = ctx.get("milestones") or {}
+        comp = ctx.get("competition") or {}
+        basic = ctx.get("basic_info") or {}
+
+        tissue_rows = "\n".join(
+            f"| {i} | {self._tissue_zh(name)} | {name} | {val:.2f} |"
+            for i, (name, val) in enumerate(ctx.get("top_tissues", []), 1)
+        ) or "| — | 暂无 GTEx 组织数据 | — | — |"
+
+        tumor_rows = "\n".join(
+            f"| {cancer} | {st['n_samples']} | {st['mean']} | {st['median']} | {st['min']}–{st['max']} |"
+            for cancer, st in (ctx.get("tumor_stats") or {}).items()
+        ) or "| — | — | — | — | — |"
+
+        target_rows = "\n".join(
+            f"| {i} | {t.get('gene', '—')} | {float(t.get('composite_score') or t.get('score') or 0):.3f} | "
+            f"{float(t.get('druggability_score') or 0):.3f} | {t.get('class', t.get('druggability_class', '—'))} |"
+            for i, t in enumerate(ctx.get("top_targets", []), 1)
+        ) or f"| 1 | {ctx['target']} | — | — | — |"
+
+        phase_dist = clin.get("phase_distribution") or {}
+        phase_rows = "\n".join(
+            f"| {_PHASE_ZH.get(phase, phase)} | {count} |"
+            for phase, count in sorted(phase_dist.items(), key=lambda x: -x[1])
+        ) or "| — | — |"
+
+        status_dist = clin.get("status_distribution") or {}
+        status_rows = "\n".join(
+            f"| {status} | {count} |"
+            for status, count in (status_dist.items() if isinstance(status_dist, dict) else [])
+        ) or "| 详见 ClinicalTrials.gov | — |"
+
+        players = clin.get("top_players") or {}
+        if isinstance(players, list):
+            player_rows = "\n".join(
+                f"| {i} | {p.get('name', p.get('sponsor', '—'))} | {p.get('trials', p.get('count', '—'))} |"
+                for i, p in enumerate(players[:8], 1)
+            )
+        elif isinstance(players, dict):
+            player_rows = "\n".join(
+                f"| {i} | {name} | {count} |"
+                for i, (name, count) in enumerate(
+                    sorted(players.items(), key=lambda x: -x[1] if isinstance(x[1], (int, float)) else 0)[:8], 1
+                )
+            )
+        else:
+            player_rows = "| — | — | — |"
+
+        milestone_rows = "\n".join(
+            f"| {m.get('name', m.get('stage', '—'))} | {m.get('duration_months', m.get('duration', '—'))} | "
+            f"{m.get('success_rate', '—')} | {m.get('cost_estimate', m.get('cost', '—'))} |"
+            for m in (ctx.get("milestone_rows") or [])[:12]
+        ) or "| 见商业计划书附录 | — | — | — |"
+
+        lit_reviews = ctx.get("literature_reviews") or []
+        enrich = ctx.get("literature_enrichment") or {}
+        enrich_note = []
+        if enrich.get("pubmed_efetch"):
+            enrich_note.append(f"PubMed MeSH 富化 {enrich.get('mesh_count', 0)} 篇")
+        if enrich.get("openalex_used"):
+            enrich_note.append(f"OpenAlex 被引 {enrich.get('citation_count_available', 0)} 篇")
+        if enrich.get("aminer_used"):
+            enrich_note.append("AMiner 引用链")
+        llm_note = (
+            f"（已启用 {ctx.get('literature_llm_source') or 'LLM'}：{ctx.get('literature_llm_model')}）"
+            if ctx.get("literature_llm_used")
+            else "（规则化评述；可用 Agent 缓存 / SCALE 本地模型 / OPENAI_API_KEY）"
+        )
+        source_counts = ctx.get("literature_review_sources") or {}
+        review_source_note = []
+        if source_counts.get("agent_cache"):
+            review_source_note.append(f"Agent 缓存评述 {source_counts['agent_cache']} 篇")
+        if source_counts.get("scale_local"):
+            review_source_note.append(f"SCALE 本地模型评述 {source_counts['scale_local']} 篇")
+        if source_counts.get("openai"):
+            review_source_note.append(f"外部 LLM 评述 {source_counts['openai']} 篇")
+        if source_counts.get("rule_based"):
+            review_source_note.append(f"规则回退评述 {source_counts['rule_based']} 篇")
+        if review_source_note:
+            llm_note = "（" + "；".join(review_source_note) + "）"
+        if ctx.get("agent_review_tasks") and not ctx.get("literature_llm_used"):
+            llm_note += f" | Agent 待办：{ctx.get('agent_review_tasks')}"
+        skills_note = "；".join(enrich_note) if enrich_note else "仅使用 Stage3 缓存元数据"
+        lit_deep_section = "\n\n".join(
+            f"#### 文献 {r['index']}：{r.get('title', '')}\n\n"
+            f"- **主题：** {r.get('theme', '—')}　|　**研究设计：** {r.get('study_design', '—')}　|　"
+            f"**证据等级：** {r.get('evidence_level', '—')}\n"
+            f"- **期刊：** {r.get('journal', '')} ({r.get('year', '')})　|　"
+            f"**PMID：** [{r.get('pmid', '')}](https://pubmed.ncbi.nlm.nih.gov/{r.get('pmid', '')}/)"
+            + (
+                f"　|　**被引：** {r.get('aminer_citation_count')}"
+                if r.get("aminer_citation_count") is not None
+                else ""
+            )
+            + "\n"
+            + (
+                f"- **MeSH：** {'；'.join(r.get('mesh_terms') or [])[:6]}\n"
+                if r.get("mesh_terms")
+                else ""
+            )
+            + f"\n{r.get('commentary_zh', '')}"
+            for r in lit_reviews
+        ) or "暂无逐篇评述。"
+        lit_skills_header = (
+            f"> **医学 Skills 富化：** {skills_note}。{llm_note}\n\n"
+            f"> {ctx.get('literature_evidence_framing', '')}\n\n"
+        )
+
+        lit_section = "\n".join(
+            f"{i}. **【{h['theme']}】** {h['title']}（{h['journal']}, {h['year']}）"
+            for i, h in enumerate(ctx.get("literature_highlights", []), 1)
+        ) or "暂无文献摘要。"
+
+        fig1 = self._embed_figure(self._fig(ctx, "fig1"))
+        fig2 = self._embed_figure(self._fig(ctx, "fig2"))
+        fig3 = self._embed_figure(self._fig(ctx, "fig3"))
+        fig4 = self._embed_figure(self._fig(ctx, "fig4"))
+
+        refs_block = "\n".join(
+            f"[{r['index']}] {r['citation_zh']} doi:{r.get('doi', 'N/A')} {r['url']}"
+            for r in ctx.get("references", [])
+        )
+
+        findings_s1 = "\n".join(f"- {f}" for f in ctx.get("key_findings_s1", [])) or "- 见表达分析结果。"
+        findings_s3 = "\n".join(f"- {f}" for f in ctx.get("key_findings_s3", [])) or "- 见深度调研结果。"
+        recs = "\n".join(f"- {r}" for r in ctx.get("recommendations", [])) or "- 需结合差异化策略推进。"
+        evidence_lines = "\n".join(f"  - {e}" for e in (ctx.get("drug_evidence") or [])[:6])
+
+        survival_text = ""
+        for cancer, data in (ctx.get("survival") or {}).items():
+            hr = data.get("hr") or data.get("hazard_ratio")
+            p = data.get("p_value")
+            if hr is not None:
+                survival_text += f"在 TCGA-{cancer} 队列中，{ctx['target']} 表达与生存结局关联 HR={hr:.3f}（P={p}）。"
+        if not survival_text:
+            survival_text = (
+                "当前运行未获得稳定的预后 Cox 模型输出；"
+                "后续可结合 TCGA 完整临床随访数据进行补充验证。"
+            )
+
+        top_tissues = ctx.get("top_tissues") or []
+        if top_tissues:
+            tissue_result = (
+                f"GTEx 显示 {ctx['target']} 在 {self._tissue_zh(top_tissues[0][0])} 呈相对高表达"
+                f"（最高 {top_tissues[0][1]:.1f} TPM，共 {ctx.get('tissue_count', 0)} 个组织）。"
+            )
+        else:
+            tissue_result = f"GTEx 组织表达数据在当前运行中未完整提取（共 {ctx.get('tissue_count', 0)} 个组织）。"
+
+        abstract = f"""**目的**  整合公开转录组、成药性知识库与临床试验注册信息，系统评估 {ctx['target']} 作为{ctx['disease_zh']}治疗靶点的生物学合理性、成药性、竞争格局与研发可行性。
+
+**方法**  采用四阶段可复现计算管线：①多源表达与预后挖掘（GTEx v8、TCGA、HPA）；②差异表达驱动的靶点筛选与成药性评分（OpenTargets、Pharos、ChEMBL）；③PubMed 文献计量与 ClinicalTrials.gov 临床管线分析；④基于风险调整的里程碑与成本模型。全部指标保留 API 溯源记录。
+
+**结果**  {tissue_result}"""
+        if ctx.get("tumor_stats"):
+            brca = ctx["tumor_stats"].get("BRCA", {})
+            if brca:
+                abstract += f"TCGA-BRCA 肿瘤样本（n={brca.get('n_samples', '—')}）中表达中位数为 {brca.get('median', '—')}。"
+        abstract += f"""成药性综合评分 {self._fmt_num(ctx.get('target_druggability'))}（{ctx.get('drug_class', '—')}），可开发性指数 {dev.get('score', '—')}/100。
+ClinicalTrials.gov 检索到 {clin.get('total_trials', '—')} 项相关试验（活跃 {clin.get('active_trials', '—')} 项），竞争强度 {self._level_zh(comp.get('competition_intensity'))}。
+研发模型估算总周期 {self._fmt_duration(ms.get('total_duration'))}、累计成本约 {self._fmt_money(ms.get('total_cost'))}、综合成功率 {self._fmt_percent(ms.get('success_rate'))}。
+
+**结论**  公开证据支持 {ctx['target']} 在{ctx['disease_zh']}中作为**中等优先级**研发靶点继续投入，但须在 CDK4/6 抑制剂红海格局下明确选择性抑制或联合用药的差异化定位；建议优先开展生物标志物分层与治疗窗口评估。"""
+
+        return f"""# {ctx['title_zh']}
+
+> **稿件类型：** 基于公共数据的计算药理学研究论文（中文草稿）  
+> **生成时间：** {ctx['generated_at']}  
+> **数据置信度：** {ctx.get('confidence', 'medium')}  
+> **说明：** 本文由靶点研究平台自动合成，正文为中文。投稿前须补充作者信息、伦理声明、原创实验与期刊格式。
+
+---
+
+## 摘要
+
+{abstract}
+
+**关键词：** {ctx['target']}；{ctx['disease_zh']}；治疗靶点；成药性；CDK4/6 抑制剂；临床试验；生物信息学
+
+---
+
+## 1 引言
+
+细胞周期蛋白依赖性激酶 4（CDK4）与 Cyclin D 形成复合物，磷酸化 RB1 并释放 E2F 转录因子，驱动 G1/S 期转换，是肿瘤增殖的关键节点。在激素受体阳性（HR+）{ctx['disease_zh']}中，CDK4/6 抑制剂（如 palbociclib、ribociclib、abemaciclib）已成为一线标准治疗，显著改善无进展生存期。然而，**原发性与获得性耐药**、**CDK4 选择性抑制 vs CDK4/6 联合抑制**的疗效-毒性平衡，以及**新一代 CDK4 选择性抑制剂**的临床定位，仍是活跃研究方向。
+
+PubMed 检索显示，与「{ctx['target']} AND {ctx['disease']}」相关的文献约 **{ctx.get('pubmed_total', '—')}** 篇，提示该领域证据密集、迭代迅速。对于药物研发决策而言，仅依赖单一数据源难以形成可执行的靶点优先级判断，需要同时回答：
+
+1. **生物学合理性**：肿瘤 vs 正常组织的表达差异是否支持治疗窗口？
+2. **成药性**：是否具备可药化的化学空间与已有工具化合物？
+3. **临床可行性**：注册试验格局是否过饱和？是否存在差异化切入点？
+4. **研发经济性**：在竞争强度与风险调整下，预期周期与成功率是否可接受？
+
+本研究利用可复现的四阶段计算管线，对 {ctx['target']} 在{ctx['disease_zh']}中进行整合评估，旨在为靶点立项、竞品跟踪与实验设计提供**结构化证据链**。
+
+---
+
+## 2 材料与方法
+
+### 2.1 研究设计
+
+本研究为**基于公开数据库的回顾性计算分析**，未纳入新收集的患者样本。分析流程分为四个阶段，各阶段输出均附带 JSON 溯源文件与 Markdown 分析报告。
+
+### 2.2 数据来源
+
+| 数据库 | 版本/来源 | 用途 |
+|--------|-----------|------|
+| GTEx | v8 | 正常组织 RNA-seq 表达（TPM） |
+| TCGA | BRCA 等 | 肿瘤表达谱与样本级 RSEM |
+| HPA | 最新 API | 蛋白水平定位与组织染色 |
+| OpenTargets / Pharos / ChEMBL | 最新 API | 成药性、已知药物与活性化合物 |
+| {ctx.get('de_source', 'TCGA/GEO')} | 管线回退 | 差异表达（方法：{ctx.get('de_method', '—')}） |
+| PubMed | NCBI E-utilities | 文献检索与引用网络 |
+| ClinicalTrials.gov | API v2 | 试验注册、阶段与申办方 |
+
+**本分析纳入的癌种：** {', '.join(ctx.get('cancer_types') or ['BRCA'])}
+
+### 2.3 分析流程
+
+**阶段一（表达与预后）**：提取 {ctx['target']} 跨组织表达谱，计算 TCGA 肿瘤样本表达分布，并在可用时进行生存关联分析。
+
+**阶段二（靶点筛选）**：以 {ctx['disease_zh']} 相关差异表达基因为基础，综合 DE 强度、成药性分类与文献证据进行排名；对 {ctx['target']} 单独输出成药性组分评分。
+
+**阶段三（深度调研）**：整合靶点功能注释、临床试验阶段分布、竞争格局与可开发性指数（0–100）。
+
+**阶段四（研发建模）**：在成药性、竞争强度与风险等级（{ms.get('risk_level', 'medium')}）约束下，生成里程碑时间表与成本-成功率估计。
+
+### 2.4 统计学与质量控制
+
+差异表达分析采用标准组间检验并控制多重比较；肿瘤表达报告均值、中位数与极差。所有外部 API 响应写入 `provenance/` 目录以便审计。交叉验证置信度：**{ctx.get('confidence', 'medium')}**。
+
+---
+
+## 3 结果
+
+### 3.1 靶点生物学特征
+
+| 项目 | 结果 |
+|------|------|
+| 基因符号 | {ctx['target']} |
+| UniProt | {basic.get('uniprot_id', 'P11802')} |
+| 蛋白名称 | {basic.get('protein_name', ctx['target'])} |
+| 亚细胞定位 | {', '.join(basic.get('subcellular_location', [])[:5]) or '细胞质/细胞核'} |
+
+**功能概述：** {(ctx.get('protein_function') or '—')[:500]}
+
+### 3.2 正常组织表达谱（GTEx）
+
+共分析 **{ctx.get('tissue_count', 0)}** 个正常组织。表达最高的组织如下：
+
+| 排名 | 组织（中文） | GTEx 组织名 | TPM |
+|------|-------------|-------------|-----|
+{tissue_rows}
+
+**阶段一关键发现：**
+
+{findings_s1}
+
+**解读：** {ctx['target']} 在生殖道相关组织及培养成纤维细胞中表达较高，提示药物开发需重点关注**治疗窗口**与正常组织毒性；乳腺本身并非最高表达组织，有利于在{ctx['disease_zh']}中探索肿瘤选择性干预。
+{fig1}
+
+### 3.3 肿瘤组织表达（TCGA）
+
+| 癌种 | 样本数 | 均值 | 中位数 | 范围 |
+|------|--------|------|--------|------|
+{tumor_rows}
+
+肿瘤表达呈右偏分布，部分样本显著高表达，支持在{ctx['disease_zh']}亚型中开展分层研究。
+{fig2}
+
+### 3.4 预后关联
+
+{survival_text}
+
+### 3.5 靶点筛选与成药性
+
+在 {ctx['disease_zh']} 候选靶点排名中，**{ctx['target']}** 位列第 **{ctx.get('target_rank') or '—'}** 名（若未进入 Top 列表，说明同疾病筛选中其他靶点 DE 信号更强，但不排除 {ctx['target']} 的生物学相关性）。
+
+**{ctx['target']} 成药性详情：**
+
+| 指标 | 数值 |
+|------|------|
+| 综合成药性评分 | {self._fmt_num(ctx.get('target_druggability'))} |
+| 成药性分类 | {ctx.get('drug_class', '—')} |
+| 阶段四校正评分 | {self._fmt_num(ctx.get('stage4_druggability'))} |
+| 可开发性指数 | {dev.get('score', '—')}/100（{dev.get('level', '—')}） |
+
+**证据链：**
+
+{evidence_lines or '  - 见 OpenTargets / ChEMBL 原始记录'}
+
+{ctx.get('drug_interpretation', '')}
+
+**候选靶点 Top 排名（节选）：**
+
+| 排名 | 基因 | 综合评分 | 成药性 | 分类 |
+|------|------|----------|--------|------|
+{target_rows}
+{fig4}
+
+### 3.6 文献研究热点
+
+PubMed 检索获得 **{len(ctx.get('references', []))}** 篇核心文献（全文元数据），主要主题分布如下：
+
+{lit_section}
+
+上述文献表明，当前研究焦点集中于 **CDK4/6 抑制剂疗效与耐药机制**、**联合治疗策略**（如 PI3K/mTOR、GPX4 等通路）以及 **CDK4 选择性抑制** 的临床前证据。
+
+### 3.6.1 逐篇文献深度解读
+
+{lit_skills_header}
+
+{lit_deep_section}
+
+### 3.7 临床试验格局
+
+| 指标 | 数值 |
+|------|------|
+| 注册试验总数 | {clin.get('total_trials', '—')} |
+| 活跃试验 | {clin.get('active_trials', '—')} |
+| 竞争强度 | {self._level_zh(comp.get('competition_intensity'))} |
+| 市场集中度 | {self._level_zh(comp.get('market_concentration'))} |
+
+**试验阶段分布：**
+
+| 阶段 | 试验数 |
+|------|--------|
+{phase_rows}
+
+**试验状态分布（节选）：**
+
+| 状态 | 数量 |
+|------|------|
+{status_rows}
+
+**主要申办方/参与方（节选）：**
+
+| 排名 | 机构 | 试验数 |
+|------|------|--------|
+{player_rows}
+{fig3}
+
+**阶段三关键发现：**
+
+{findings_s3}
+
+### 3.8 研发里程碑与经济性估计
+
+在 **{ms.get('modality', '小分子')}** 模态与 **{ms.get('risk_level', 'medium')}** 风险假设下：
+
+| 指标 | 估计值 |
+|------|--------|
+| 总研发周期 | {self._fmt_duration(ms.get('total_duration'))} |
+| 预计完成时间 | {ms.get('estimated_completion', '—')} |
+| 累计成本 | {self._fmt_money(ms.get('total_cost'))} |
+| 综合成功率 | {self._fmt_percent(ms.get('success_rate'))} |
+| 融资轮次 | {ctx.get('funding', {}).get('rounds', '—')} |
+
+**分阶段里程碑（节选）：**
+
+| 阶段 | 周期（月） | 成功率 | 成本 |
+|------|------------|--------|------|
+{milestone_rows}
+
+---
+
+## 4 讨论
+
+### 4.1 生物学与表达层面的启示
+
+本分析显示 {ctx['target']} 在 TCGA-BRCA 肿瘤样本中表达变异较大，而在正常组织中于多种增殖活跃组织呈中高表达。这意味着：
+
+- **支持点**：肿瘤中确实存在高表达亚群，可用于患者分层或伴随诊断开发；
+- **风险点**：正常组织表达提示需警惕骨髓抑制、胃肠道毒性等与 CDK4/6 抑制剂类似的不良反应谱。
+
+与已上市 CDK4/6 抑制剂相比，**CDK4 选择性抑制**（见参考文献中 2025 年 Cancer Cell 报道）试图在保留疗效的同时降低 CDK6 相关毒性，是本靶点差异化的重要科学假设。
+
+### 4.2 成药性与可开发性
+
+成药性评分 **{self._fmt_num(ctx.get('target_druggability'))}**，ChEMBL 显示已有活性化合物与上市药物关联，Pharos 分类为 Tchem，表明**化学工具链成熟**。可开发性指数 **{dev.get('score', '—')}/100** 处于中等水平，主要受**竞争强度**与**临床开发不确定性**拖累，而非生物学证据不足。
+
+### 4.3 临床竞争格局
+
+{clin.get('total_trials', '—')} 项注册试验中，II/III 期占比高，说明该领域已越过早期概念验证阶段，进入**红海竞争**。申办方分散（市场集中度 {comp.get('market_concentration', 'low')}），既有大型药企也有学术主导试验，意味着：
+
+- 常规 CDK4/6 抑制路径难以获得同类最佳（Best-in-class）定位；
+- **细分人群**（如 CDK4 扩增、CDK6 低表达、内分泌耐药后线）可能是突破口。
+
+### 4.4 研发经济性与决策建议
+
+模型估算总周期 **{self._fmt_duration(ms.get('total_duration'))}**、成本 **{self._fmt_money(ms.get('total_cost'))}**，综合成功率仅 **{self._fmt_percent(ms.get('success_rate'))}**，符合肿瘤靶向药行业基线。对于立项决策：
+
+{recs}
+
+### 4.5 局限性
+
+1. **计算回顾性设计**：未包含前瞻性验证实验，表达-疗效因果链需体外/体内实验确认；
+2. **公共数据滞后**：ClinicalTrials.gov 与 PubMed 更新存在延迟，近期试验可能未完全纳入；
+3. **GEO 数据缺失**：部分差异表达队列因 GEO 矩阵解析失败而回退至 TCGA，可能低估特定亚型信号；
+4. **经济性模型为情景假设**：成功率与成本受模态、适应症扩展策略影响较大。
+
+---
+
+## 5 结论
+
+综合多组学表达、成药性证据、文献趋势与临床注册数据，**{ctx['target']} 在{ctx['disease_zh']}中具备继续研发的生物学与化学基础**，但临床竞争已高度激烈。建议后续工作优先围绕：
+
+1. **差异化科学假说**：CDK4 选择性抑制 vs 传统 CDK4/6 抑制的疗效/毒性优势；
+2. **生物标志物分层**：结合肿瘤表达异质性与耐药突变谱；
+3. **联合用药策略**：参考近期文献中的 PI3K/mTOR、铁死亡等联合路径；
+4. **监管与临床路径**：在 II 期单臂或随机对照设计中明确终点与伴随诊断。
+
+---
+
+## 6 临床转化与后续实验建议
+
+| 优先级 | 建议实验 | 预期产出 |
+|--------|----------|----------|
+| P0 | HR+/{ctx['disease_zh']} 细胞系 CDK4 磷酸化/RB 磷酸化验证 | 靶点 occupancy 与通路抑制证据 |
+| P0 | 患者来源异种移植（PDX）分层疗效 | 高表达亚群响应数据 |
+| P1 | 与内分泌治疗及 CDK4/6 抑制剂的序贯/联合方案 | 耐药逆转或延迟耐药 |
+| P1 | 正常组织毒性面板（骨髓、肠道上皮） | 治疗窗口评估 |
+| P2 | 真实世界注册研究方案设计 | 上市后差异化定位 |
+
+---
+
+## 数据可用性声明
+
+全部结构化结果、HTML 证据包与 API 溯源文件存放于管线 `output/` 目录，包括：
+
+- `stage1/{ctx['target']}_stage1_results.json`
+- `stage2/{ctx['disease']}_stage2_results.json`
+- `stage3/{ctx['target']}_{ctx['disease']}_stage3_results.json`
+- `stage4/{ctx['target']}_{ctx['disease']}_stage4_results.json`
+
+---
+
+## 参考文献
+
+{refs_block}
+
+---
+
+## 图表目录（建议）
+
+- **图 1** {ctx['target']} 在 GTEx 54 个正常组织中的 TPM 表达热图
+- **图 2** TCGA-BRCA 肿瘤样本 {ctx['target']} 表达分布（小提琴图/直方图）
+- **图 3** 候选靶点成药性评分排名森林图
+- **图 4** ClinicalTrials.gov 试验阶段与状态桑基图
+- **图 5** 研发里程碑甘特图与累计成本曲线
+- **表 1** 数据来源与检索式汇总（见方法部分）
+- **表 2** 核心文献主题分类与代表性研究
+
+---
+
+## 作者贡献、利益冲突与伦理
+
+- **作者贡献：** [待填写，建议采用 CRediT 分类]
+- **利益冲突：** 待全体作者确认后填写；本自动草稿不代替作者声明。
+- **伦理声明：** 本稿基于公开数据库与去标识化数据；是否需要伦理审查或豁免应由作者所在机构确认。
+
+---
+
+*靶点研究平台 · 期刊论文生成器 v2.0（中文版）*
+"""
+
+    def _render_markdown_en(self, ctx: dict[str, Any]) -> str:
+        """英文简版（保留兼容）。"""
+        return self._render_markdown_zh(ctx).replace("（中文版）", "(English placeholder — use language=zh)")
+
+    @staticmethod
+    def _fmt_percent(value: Any) -> str:
+        if value is None:
+            return "—"
+        try:
+            v = float(value)
+            if v <= 1:
+                return f"{v * 100:.2f}%"
+            return f"{v:.2f}%"
+        except (TypeError, ValueError):
+            return str(value)
+
+    @staticmethod
+    def _fmt_num(value: Any) -> str:
+        if value is None:
+            return "—"
+        try:
+            return f"{float(value):.3f}"
+        except (TypeError, ValueError):
+            return str(value)
+
+    @staticmethod
+    def _fmt_money(value: Any) -> str:
+        if value is None:
+            return "—"
+        try:
+            return f"${float(value):.1f}M"
+        except (TypeError, ValueError):
+            return str(value)
+
+    @staticmethod
+    def _fmt_duration(value: Any) -> str:
+        if value is None:
+            return "—"
+        s = str(value)
+        if "个月" in s or "月" in s:
+            return s
+        try:
+            months = int(float(value))
+            years = months / 12
+            return f"{months} 个月（约 {years:.1f} 年）"
+        except (TypeError, ValueError):
+            return s
+
+    def _render_docx_zh(self, ctx: dict[str, Any], slug: str) -> str:
+        from docx import Document
+        from docx.enum.text import WD_ALIGN_PARAGRAPH
+        from docx.shared import Pt, RGBColor
+
+        doc = Document()
+        title = doc.add_heading(ctx["title_zh"], level=0)
+        title.alignment = WD_ALIGN_PARAGRAPH.CENTER
+
+        meta = doc.add_paragraph("计算药理学研究论文草稿（中文）— 投稿前须人工修订")
+        meta.alignment = WD_ALIGN_PARAGRAPH.CENTER
+        meta.runs[0].font.size = Pt(9)
+        meta.runs[0].font.color.rgb = RGBColor(0x88, 0x88, 0x88)
+
+        doc.add_heading("摘要", level=1)
+        clin = ctx.get("clinical_summary") or {}
+        dev = ctx.get("developability") or {}
+        doc.add_paragraph(
+            f"目的：评估 {ctx['target']} 在{ctx['disease_zh']}中的治疗靶点价值。"
+            f"结果：临床试验 {clin.get('total_trials', '—')} 项，可开发性 {dev.get('score', '—')}/100，"
+            f"成药性 {self._fmt_num(ctx.get('target_druggability'))}。"
+            f"结论：具备研发基础，需明确差异化定位。"
+        )
+
+        for heading, body in [
+            ("1 引言", f"{ctx['target']} 是细胞周期关键激酶，在{ctx['disease_zh']}治疗中具有重要意义。"),
+            ("2 方法", f"数据来源：{', '.join(ctx.get('data_sources') or [])}。四阶段可复现管线。"),
+            ("3 结果", ctx.get("drug_interpretation") or "见 Markdown 完整版。"),
+            ("4 讨论", f"竞争强度 {ctx.get('competition', {}).get('competition_intensity', '—')}，需关注治疗窗口。"),
+            ("5 结论", f"建议围绕生物标志物分层与 CDK4 选择性抑制继续推进。"),
+        ]:
+            doc.add_heading(heading, level=1)
+            doc.add_paragraph(body)
+
+        doc.add_heading("参考文献", level=1)
+        for ref in ctx.get("references", [])[:20]:
+            doc.add_paragraph(f"[{ref['index']}] {ref['citation_zh']}", style="List Number")
+
+        path = self.output_dir / f"{slug}_manuscript.docx"
+        doc.save(str(path))
+        return str(path.resolve())
+
+
+def generate_manuscript_from_stages(
+    output_root: str | Path,
+    target: str,
+    disease: str,
+    stage1: dict,
+    stage2: dict,
+    stage3: dict,
+    stage4: dict,
+    *,
+    language: str = "zh",
+) -> dict:
+    """从各阶段结果 dict 生成稿件（无需完整 pipeline JSON）。"""
+    generator = ManuscriptGenerator(str(Path(output_root) / "manuscripts"))
+    return generator.generate(
+        {
+            "stages": {
+                "stage1": stage1,
+                "stage2": stage2,
+                "stage3": stage3,
+                "stage4": stage4,
+            }
+        },
+        target,
+        disease,
+        language=language,
+    )