admin/PlantAI-rag

Fork 0

Files

T

potaser 5bb98a937a 初始提交

2026-07-10 17:59:59 +08:00

15 KiB

Executable File

Raw Blame History

增强版实体-关系提取规范

基于70+维结构化提取的实体关系扩展方案创建时间: 2026-03-18

当前70+维中的实体关系定义

现有 entities 字段

"entities": [
  {"name": "基因/蛋白名称", "type": "gene/domain/database", "species": "物种", "description": "描述"}
]

问题:

type 只有简单的3种类型
缺少表型、代谢物、处理/胁迫等实体类型
没有标准化的实体分类体系

现有 relations 字段

"relations": [
  {"source": "源实体", "target": "目标实体", "type": "关系类型", "description": "描述"}
]

问题:

type 没有预定义的关系类型列表
缺少调控关系、表达关系等生物学关系

增强版实体类型定义

1. 核心实体类型 (Core Entities)

实体类型	类型标识	属性	示例
基因	`gene`	name, symbol, id, family, chromosome, position	FtSPL1, AtSPL15
蛋白质	`protein`	name, accession, domain, molecular_weight, localization	FtSPL1蛋白
基因家族	`gene_family`	name, symbol, member_count, conserved_domain	SPL, NLR, MYB

2. 物种相关实体 (Organism Entities)

实体类型	类型标识	属性	示例
物种	`organism`	scientific_name, common_name, taxonomy_id, clade	Fagopyrum tataricum (苦荞)
品种/品系	`cultivar`	name, parent_organism, traits	特定品种
比较物种	`reference_organism`	name, evolutionary_distance, genome_version	Arabidopsis thaliana

3. 表型与性状 (Phenotype Entities)

实体类型	类型标识	属性	示例
表型	`phenotype`	name, trait_category, observable, measurement	花发育、果实大小
性状	`trait`	name, quantitative/qualitative, heritability	株高、产量
发育阶段	`developmental_stage`	name, tissue_type, age/days	开花期、果实成熟期

4. 实验处理 (Treatment Entities)

实体类型	类型标识	属性	示例
处理/胁迫	`treatment`	name, type, duration, intensity, conditions	干旱处理、盐胁迫、ABA处理
激素	`hormone`	name, concentration, application_method	ABA, GA3, IAA
环境因子	`environmental_factor`	name, value, unit	温度、光照强度

5. 分子实体 (Molecular Entities)

实体类型	类型标识	属性	示例
代谢物	`metabolite`	name, class, pathway, detection_method	花青素、黄酮类
miRNA	`mirna`	name, sequence, target_genes	miR156, miR529
转录因子	`transcription_factor`	name, family, binding_motif	FtSPL1 (SPL家族)
启动子元件	`promoter_element`	name, sequence, position, function	GTAC motif (SPL结合位点)
结构域	`protein_domain`	name, pfam_id, function, conservation	SBP domain (PF03110)

6. 生物过程 (Process Entities)

实体类型	类型标识	属性	示例
生物过程	`biological_process`	name, go_terms, pathway	花发育、果实成熟、胁迫响应
信号通路	`signaling_pathway`	name, components, trigger	SPL-miR156通路
调控网络	`regulatory_network`	name, nodes, edges_count	SPL调控网络

7. 数据与资源 (Data Entities)

实体类型	类型标识	属性	示例
数据集	`dataset`	accession, type, source, tissue, treatment	GSE111937
基因组区域	`genomic_region`	chromosome, start, end, strand, type	基因座位置
共线性区块	`syntenic_block`	chromosome, genes, conservation_score	共线性区域

增强版关系类型定义

1. 调控关系 (Regulatory Relations)

关系类型	关系标识	方向	描述	示例
正向调控	`activates`	A → B	A激活B的表达或活性	SPL9 → miR156
负向调控	`inhibits`	A → B	A抑制B的表达或活性	miR156 → SPL
调控	`regulates`	A → B	A调控B（方向不明确）	SPL regulates flowering
转录激活	`transcriptionally_activates`	A → B	A转录因子激活B基因	SPL → FT
转录抑制	`transcriptionally_represses`	A → B	A转录因子抑制B基因	SPL → TFL
翻译调控	`translationally_regulates`	A → B	A调控B的翻译	miRNA → mRNA

2. 包含关系 (Hierarchical Relations)

关系类型	关系标识	方向	描述	示例
属于	`belongs_to`	A → B	A属于B类别	FtSPL1 belongs_to SPL family
包含	`contains`	A → B	A包含B	SPL family contains FtSPL1
部分	`part_of`	A → B	A是B的一部分	SBP domain part_of SPL protein
成员	`member_of`	A → B	A是B的成员	FtSPL1 member_of SPL family
子类	`subclass_of`	A → B	A是B的子类	SPL15 subclass_of SPL family

3. 表达关系 (Expression Relations)

关系类型	关系标识	方向	描述	示例
表达于	`expressed_in`	gene → tissue	基因在组织中表达	FtSPL1 expressed_in flower
上调	`upregulated_in`	gene → condition	基因在条件下上调	FtSPL9 upregulated_in drought
下调	`downregulated_in`	gene → condition	基因在条件下下调	FtSPL2 downregulated_in salt
诱导表达	`induced_by`	gene → treatment	基因被处理诱导	FtSPL1 induced_by ABA
抑制表达	`repressed_by`	gene → treatment	基因被处理抑制	FtSPL3 repressed_by GA
时序表达	`temporally_expressed`	gene → stage	基因在发育阶段表达	FtSPL expressed_at flowering

4. 功能关系 (Functional Relations)

关系类型	关系标识	方向	描述	示例
参与	`involved_in`	gene → process	基因参与生物过程	FtSPL involved_in flowering
控制	`controls`	gene → phenotype	基因控制表型	FtSPL16 controls fruit_size
影响	`affects`	gene → trait	基因影响性状	FtSPL affects plant_height
介导	`mediates`	gene → response	基因介导响应	FtSPL mediates stress_response
促进	`promotes`	gene → process	基因促进过程	FtSPL promotes flowering
抑制	`represses`	gene → process	基因抑制过程	SPL represses juvenile_phase

5. 进化关系 (Evolutionary Relations)

关系类型	关系标识	方向	描述	示例
同源	`homologous_to`	gene ↔ gene	基因同源	FtSPL1 homologous_to AtSPL1
直系同源	`orthologous_to`	gene ↔ gene	直系同源	FtSPL1 orthologous_to AtSPL1
旁系同源	`paralogous_to`	gene ↔ gene	旁系同源	FtSPL1 paralogous_to FtSPL2
来源于	`derived_from`	gene → gene	基因来源于	FtSPL derived_from duplication
复制	`duplicated_to`	gene → gene	基因复制为	SPL1 duplicated_to SPL2
共线性	`syntenic_with`	gene ↔ gene	基因共线性	FtSPL1 syntenic_with GmSPL1

6. 互作关系 (Interaction Relations)

关系类型	关系标识	方向	描述	示例
互作	`interacts_with`	protein ↔ protein	蛋白互作	FtSPL1 interacts_with FtSPL2
结合	`binds_to`	protein → DNA/metabolite	蛋白结合	SPL binds_to GTAC_motif
磷酸化	`phosphorylates`	kinase → substrate	激酶磷酸化底物	kinase phosphorylates SPL
去磷酸化	`dephosphorylates`	phosphatase → substrate	磷酸酶去磷酸化	phosphatase dephosphorylates SPL
切割	`cleaves`	enzyme → substrate	酶切割底物	Dicer cleaves pre-miRNA

7. 数据关系 (Data Relations)

关系类型	关系标识	方向	描述	示例
来源于	`originated_from`	data → organism	数据来源于物种	Dataset originated_from 苦荞
测序	`sequenced_from`	data → tissue	数据测序自组织	RNA-seq sequenced_from flower
注释	`annotated_in`	gene → database	基因注释于数据库	FtSPL1 annotated_in TAIR
存储于	`stored_in`	data → database	数据存储于数据库	GSE111937 stored_in GEO

增强版JSON Schema示例

完整实体示例

{
  "entities": [
    {
      "id": "FtSPL1",
      "name": "FtSPL1",
      "type": "gene",
      "symbol": "FtSPL1",
      "species": "Fagopyrum tataricum",
      "family": "SPL",
      "chromosome": "Chr1",
      "position": "123456-123890",
      "strand": "+",
      "description": "SQUAMOSA PROMOTER BINDING PROTEIN-LIKE 1 in tartary buckwheat"
    },
    {
      "id": "SPL_family",
      "name": "SPL transcription factor family",
      "type": "gene_family",
      "symbol": "SPL",
      "member_count": 24,
      "conserved_domain": "SBP domain (PF03110)",
      "description": "SQUAMOSA PROMOTER BINDING PROTEIN-LIKE family"
    },
    {
      "id": "Fagopyrum_tataricum",
      "name": "Fagopyrum tataricum",
      "type": "organism",
      "common_name": "Tartary buckwheat",
      "common_name_cn": "苦荞",
      "taxonomy_id": "NCBI:txid321",
      "clade": "Eudicots",
      "description": "Tartary buckwheat, an important pseudocereal crop"
    },
    {
      "id": "flower_development",
      "name": "flower development",
      "type": "phenotype",
      "trait_category": "developmental",
      "observable": true,
      "description": "Process of flower formation and development"
    },
    {
      "id": "drought_stress",
      "name": "drought stress",
      "type": "treatment",
      "treatment_type": "abiotic_stress",
      "duration": "7 days",
      "intensity": "severe",
      "description": "Water deficit treatment"
    },
    {
      "id": "PF03110",
      "name": "SBP domain",
      "type": "protein_domain",
      "pfam_id": "PF03110",
      "function": "DNA binding",
      "conservation": "highly conserved in plants",
      "description": "SQUAMOSA PROMOTER BINDING PROTEIN domain"
    }
  ],
  
  "relations": [
    {
      "id": "rel_1",
      "source": "FtSPL1",
      "target": "SPL_family",
      "type": "member_of",
      "description": "FtSPL1 is a member of SPL family",
      "confidence": "high",
      "evidence": "phylogenetic analysis"
    },
    {
      "id": "rel_2",
      "source": "FtSPL1",
      "target": "Fagopyrum_tataricum",
      "type": "belongs_to",
      "description": "FtSPL1 belongs to tartary buckwheat",
      "confidence": "high",
      "evidence": "genome annotation"
    },
    {
      "id": "rel_3",
      "source": "FtSPL1",
      "target": "flower_development",
      "type": "involved_in",
      "description": "FtSPL1 is involved in flower development",
      "confidence": "medium",
      "evidence": "expression analysis"
    },
    {
      "id": "rel_4",
      "source": "FtSPL1",
      "target": "PF03110",
      "type": "contains",
      "description": "FtSPL1 protein contains SBP domain",
      "confidence": "high",
      "evidence": "Pfam annotation"
    },
    {
      "id": "rel_5",
      "source": "drought_stress",
      "target": "FtSPL1",
      "type": "upregulates",
      "description": "Drought stress upregulates FtSPL1 expression",
      "confidence": "medium",
      "evidence": "RNA-seq data"
    }
  ]
}

与现有70+维的整合建议

方案1: 扩展现有字段

保持现有 entities 和 relations 字段，但增强其内容规范：

{
  "entities": [
    {
      "id": "唯一标识",
      "name": "实体名称",
      "type": "gene|protein|organism|phenotype|treatment|metabolite|process|...",
      "species": "物种",
      "attributes": {},
      "description": "描述"
    }
  ],
  "relations": [
    {
      "id": "关系唯一标识",
      "source": "源实体ID",
      "target": "目标实体ID",
      "type": "activates|inhibits|regulates|expressed_in|involved_in|...",
      "description": "关系描述",
      "confidence": "high|medium|low",
      "evidence": "证据来源"
    }
  ]
}

方案2: 新增独立知识图谱字段

在70+维之外新增 knowledge_graph 字段：

{
  "knowledge_graph": {
    "nodes": [...],  // 实体
    "edges": [...],  // 关系
    "pathways": [...],  // 通路
    "networks": [...]   // 网络
  }
}

方案3: 分层提取

第一层: 现有70+维（结构化信息）
第二层: 实体关系（知识图谱）
第三层: 通路网络（系统生物学）

实施建议

1. 提示词修改

在 extraction_prompt.md 中扩展 entities 和 relations 的定义：

"entities": [
  {
    "id": "唯一标识符（如FtSPL1）",
    "name": "实体名称",
    "type": "实体类型：gene|protein|organism|phenotype|treatment|metabolite|process|gene_family|protein_domain|...",
    "species": "所属物种",
    "attributes": {},
    "description": "描述"
  }
],

"relations": [
  {
    "id": "关系唯一标识",
    "source": "源实体ID",
    "target": "目标实体ID", 
    "type": "关系类型：activates|inhibits|regulates|expressed_in|upregulated_in|downregulated_in|involved_in|controls|homologous_to|orthologous_to|paralogous_to|member_of|contains|part_of|interacts_with|binds_to|...",
    "description": "关系描述",
    "confidence": "high|medium|low",
    "evidence": "证据来源"
  }
]

2. 验证规则

添加实体关系验证：

实体ID必须唯一
关系中的source和target必须存在于entities中
关系类型必须符合预定义列表
实体类型必须符合预定义列表

3. 与知识库整合

提取的实体关系可用于：

构建知识图谱
支持图检索
发现新的基因-表型关联
跨物种比较分析

总结

方面	当前70+维	增强版建议
实体类型	3种 (gene/domain/database)	30+种 (基因、蛋白、物种、表型、处理、代谢物、过程等)
关系类型	未预定义	40+种 (调控、包含、表达、参与、进化、互作等)
属性完整性	基础 (name, type, species)	丰富 (id, symbol, position, confidence, evidence等)
验证机制	无	有 (唯一性、类型检查、引用完整性)
应用场景	信息提取	知识图谱构建、图检索、通路分析

建议: 采用方案1（扩展现有字段），保持向后兼容的同时增强实体关系表达能力。

文档创建: 2026-03-18
基于: 70+维结构化提取规范 v1.0

15 KiB Executable File Raw Blame History Unescape Escape