structflo-ner 0.5.0__tar.gz → 0.7.0__tar.gz

structflo_ner-0.5.0/README.md → structflo_ner-0.7.0/PKG-INFO

@@ -1,3 +1,16 @@
+Metadata-Version: 2.5
+Name: structflo-ner
+Version: 0.7.0
+Summary: Drug discovery NER wrapper around LangExtract — zero-config entity extraction for chemistry and biology.
+License: Apache-2.0
+Requires-Python: >=3.10
+Requires-Dist: langextract>=1.7.0
+Requires-Dist: pyyaml>=6.0
+Requires-Dist: rapidfuzz>=3.0
+Provides-Extra: dataframe
+Requires-Dist: pandas>=1.5; extra == 'dataframe'
+Description-Content-Type: text/markdown
+
 
 <h1 align="center">structflo.ner</h1>
 <p align="center">
@@ -334,6 +347,29 @@ df = result.to_dataframe()
 result.to_dict()
 ```
 
+Each bioactivity carries its measurement in `attributes`:
+
+| Attribute       | Example                     | Meaning                                                    |
+| --------------- | --------------------------- | ---------------------------------------------------------- |
+| `value`         | `>20.0`                     | number as reported, qualifier kept                         |
+| `unit`          | `µM`                        | unit as written                                            |
+| `assay_type`    | `CC50`                      | the endpoint (IC50, MIC90, GI50, ED90, ...)                |
+| `assay`         | `HepG2 MTT`                 | assay, cell line or read-out the value was measured in     |
+| `compound_name` | `8t`                        | compound the value belongs to                              |
+| `target`        | `InhA`                      | protein measured against (`TB` profile only)               |
+| `strain`        | `H37Rv`                     | organism, strain or virus measured against (`TB` only)     |
+| `combination`   | `meropenem`                 | second compound dosed alongside (`TB` profile only)        |
+
+An attribute the text does not state is absent from `attributes`.
+
+A table on one page often names its target only on another. Pass text from
+elsewhere in the document (its title or summary) as `context`; the model uses it
+to fill attributes the page leaves unstated, and extracts nothing from it:
+
+```python
+result = extractor.extract(page_text, context=document_summary)
+```
+
 
 ## Notebooks
 

structflo_ner-0.5.0/PKG-INFO → structflo_ner-0.7.0/README.md

@@ -1,16 +1,3 @@
-Metadata-Version: 2.5
-Name: structflo-ner
-Version: 0.5.0
-Summary: Drug discovery NER wrapper around LangExtract — zero-config entity extraction for chemistry and biology.
-License: Apache-2.0
-Requires-Python: >=3.10
-Requires-Dist: langextract>=1.6.0
-Requires-Dist: pyyaml>=6.0
-Requires-Dist: rapidfuzz>=3.0
-Provides-Extra: dataframe
-Requires-Dist: pandas>=1.5; extra == 'dataframe'
-Description-Content-Type: text/markdown
-
 
 <h1 align="center">structflo.ner</h1>
 <p align="center">
@@ -347,6 +334,29 @@ df = result.to_dataframe()
 result.to_dict()
 ```
 
+Each bioactivity carries its measurement in `attributes`:
+
+| Attribute       | Example                     | Meaning                                                    |
+| --------------- | --------------------------- | ---------------------------------------------------------- |
+| `value`         | `>20.0`                     | number as reported, qualifier kept                         |
+| `unit`          | `µM`                        | unit as written                                            |
+| `assay_type`    | `CC50`                      | the endpoint (IC50, MIC90, GI50, ED90, ...)                |
+| `assay`         | `HepG2 MTT`                 | assay, cell line or read-out the value was measured in     |
+| `compound_name` | `8t`                        | compound the value belongs to                              |
+| `target`        | `InhA`                      | protein measured against (`TB` profile only)               |
+| `strain`        | `H37Rv`                     | organism, strain or virus measured against (`TB` only)     |
+| `combination`   | `meropenem`                 | second compound dosed alongside (`TB` profile only)        |
+
+An attribute the text does not state is absent from `attributes`.
+
+A table on one page often names its target only on another. Pass text from
+elsewhere in the document (its title or summary) as `context`; the model uses it
+to fill attributes the page leaves unstated, and extracts nothing from it:
+
+```python
+result = extractor.extract(page_text, context=document_summary)
+```
+
 
 ## Notebooks
 

{structflo_ner-0.5.0 → structflo_ner-0.7.0}/pyproject.toml

@@ -6,7 +6,7 @@ readme = "README.md"
 requires-python = ">=3.10"
 license = { text = "Apache-2.0" }
 dependencies = [
-    "langextract>=1.6.0",
+    "langextract>=1.7.0",
     "rapidfuzz>=3.0",
     "PyYAML>=6.0",
 ]

{structflo_ner-0.5.0 → structflo_ner-0.7.0}/structflo/ner/__init__.py

@@ -63,7 +63,7 @@ from structflo.ner.profiles import (
     EntityProfile,
 )
 
-__version__ = "0.5.0"
+__version__ = "0.7.0"
 
 __all__ = [
     # Main classes

{structflo_ner-0.5.0 → structflo_ner-0.7.0}/structflo/ner/_examples.py

@@ -158,8 +158,8 @@ BIOLOGY_EXAMPLES: list[lx.data.ExampleData] = [
 
 _BIOACTIVITY_EXAMPLE_1 = lx.data.ExampleData(
     text=(
-        "Compound 7 inhibited EGFR with an IC50 of 2.3 nM in a cell-free enzymatic assay "
-        "and showed an EC50 of 45 nM in A549 (human lung adenocarcinoma) cell proliferation assay. "
+        "Compound 7 (CHEMBL5171042) inhibited EGFR with an IC50 of 2.3 nM in a cell-free "
+        "enzymatic assay and showed an EC50 of 45 nM in A549 (human lung adenocarcinoma) cell proliferation assay. "
         "Selectivity over ERBB2 was >100-fold (Ki = 0.8 nM vs 95 nM)."
     ),
     extractions=[
@@ -170,7 +170,8 @@ _BIOACTIVITY_EXAMPLE_1 = lx.data.ExampleData(
                 "value": "2.3",
                 "unit": "nM",
                 "assay_type": "IC50",
-                "compound_name": "Compound 7",
+                "compound_name": "CHEMBL5171042",
+                "assay": "cell-free enzymatic assay",
             },
         ),
         lx.data.Extraction(
@@ -185,7 +186,8 @@ _BIOACTIVITY_EXAMPLE_1 = lx.data.ExampleData(
                 "value": "45",
                 "unit": "nM",
                 "assay_type": "EC50",
-                "compound_name": "Compound 7",
+                "compound_name": "CHEMBL5171042",
+                "assay": "A549 cell proliferation assay",
             },
         ),
         lx.data.Extraction(
@@ -204,7 +206,7 @@ _BIOACTIVITY_EXAMPLE_1 = lx.data.ExampleData(
                 "value": "0.8",
                 "unit": "nM",
                 "assay_type": "Ki",
-                "compound_name": "Compound 7",
+                "compound_name": "CHEMBL5171042",
             },
         ),
     ],
@@ -337,6 +339,7 @@ _FULL_EXAMPLE_1 = lx.data.ExampleData(
                 "unit": "µM",
                 "assay_type": "IC50",
                 "compound_name": "Gefitinib",
+                "assay": "cell-free biochemical assay",
             },
         ),
         lx.data.Extraction(
@@ -357,6 +360,7 @@ _FULL_EXAMPLE_1 = lx.data.ExampleData(
                 "unit": "µM",
                 "assay_type": "IC50",
                 "compound_name": "Gefitinib",
+                "assay": "A431 cells",
             },
         ),
         lx.data.Extraction(
@@ -449,6 +453,7 @@ _TB_EXAMPLE_1 = lx.data.ExampleData(
                 "assay_type": "MIC",
                 "strain": "H37Rv",
                 "compound_name": "Bedaquiline",
+                "assay": "MABA",
             },
         ),
         lx.data.Extraction(
@@ -619,6 +624,8 @@ _TB_EXAMPLE_3 = lx.data.ExampleData(
                 "unit": "nM",
                 "assay_type": "IC50",
                 "compound_name": "Compound 14a",
+                "assay": "biochemical assay",
+                "target": "InhA",
             },
         ),
         lx.data.Extraction(
@@ -643,6 +650,7 @@ _TB_EXAMPLE_3 = lx.data.ExampleData(
                 "assay_type": "MIC90",
                 "strain": "H37Rv",
                 "compound_name": "Compound 14a",
+                "assay": "REMA",
             },
         ),
         lx.data.Extraction(
@@ -658,6 +666,7 @@ _TB_EXAMPLE_3 = lx.data.ExampleData(
                 "unit": "uM",
                 "assay_type": "EC50",
                 "compound_name": "Compound 14a",
+                "assay": "THP-1 macrophage infection assay",
             },
         ),
         lx.data.Extraction(
@@ -673,6 +682,7 @@ _TB_EXAMPLE_3 = lx.data.ExampleData(
                 "unit": "uM",
                 "assay_type": "MIC",
                 "compound_name": "Compound 14a",
+                "assay": "LORA",
             },
         ),
         lx.data.Extraction(
@@ -688,6 +698,7 @@ _TB_EXAMPLE_3 = lx.data.ExampleData(
                 "unit": "uM",
                 "assay_type": "CC50",
                 "compound_name": "Compound 14a",
+                "assay": "HepG2 cells",
             },
         ),
     ],
@@ -843,6 +854,7 @@ _TB_EXAMPLE_5 = lx.data.ExampleData(
                 "unit": "nM",
                 "assay_type": "IC50",
                 "compound_name": "SACC-3060",
+                "target": "DprE1",
             },
         ),
         lx.data.Extraction(
@@ -860,12 +872,217 @@ _TB_EXAMPLE_5 = lx.data.ExampleData(
     ],
 )
 
+# Table as Docling emits it: the column header names the assay, the endpoint
+# and unit come from the footnote, and every cell is its own bioactivity. The
+# ID column, not the deck-local row label, names the compound.
+_TB_EXAMPLE_6 = lx.data.ExampleData(
+    text=(
+        "Table 2. Whole-cell activity\n\n"
+        "| Cpd | ID | MABA | THP-1 | Vero |\n"
+        "| --- | --- | --- | --- | --- |\n"
+        "| 21a | SACC-4412 | 0.12 | 0.9 | >50 |\n"
+        "| 21b | CHEMBL5290347 | <0.03 | 2.4* | 18.5 |\n\n"
+        "MABA: MIC90 (µM) against M. tuberculosis H37Rv. THP-1: intracellular EC90 (µM). "
+        "Vero: CC50 (µM). *Single determination."
+    ),
+    extractions=[
+        lx.data.Extraction(
+            extraction_class="assay",
+            extraction_text="MABA",
+            attributes={"assay_format": "whole-cell", "strain": "H37Rv"},
+        ),
+        lx.data.Extraction(
+            extraction_class="assay",
+            extraction_text="THP-1",
+            attributes={"cell_line": "THP-1", "assay_format": "intracellular"},
+        ),
+        lx.data.Extraction(
+            extraction_class="assay",
+            extraction_text="Vero",
+            attributes={"cell_line": "Vero", "assay_format": "cytotoxicity"},
+        ),
+        lx.data.Extraction(
+            extraction_class="compound_name",
+            extraction_text="SACC-4412",
+            attributes={"synonyms": "21a"},
+        ),
+        lx.data.Extraction(
+            extraction_class="bioactivity",
+            extraction_text="0.12",
+            attributes={
+                "value": "0.12",
+                "unit": "µM",
+                "assay_type": "MIC90",
+                "strain": "H37Rv",
+                "compound_name": "SACC-4412",
+                "assay": "MABA",
+            },
+        ),
+        lx.data.Extraction(
+            extraction_class="bioactivity",
+            extraction_text="0.9",
+            attributes={
+                "value": "0.9",
+                "unit": "µM",
+                "assay_type": "EC90",
+                "compound_name": "SACC-4412",
+                "assay": "THP-1",
+            },
+        ),
+        lx.data.Extraction(
+            extraction_class="bioactivity",
+            extraction_text=">50",
+            attributes={
+                "value": ">50",
+                "unit": "µM",
+                "assay_type": "CC50",
+                "compound_name": "SACC-4412",
+                "assay": "Vero",
+            },
+        ),
+        lx.data.Extraction(
+            extraction_class="compound_name",
+            extraction_text="CHEMBL5290347",
+            attributes={"synonyms": "21b"},
+        ),
+        lx.data.Extraction(
+            extraction_class="bioactivity",
+            extraction_text="<0.03",
+            attributes={
+                "value": "<0.03",
+                "unit": "µM",
+                "assay_type": "MIC90",
+                "strain": "H37Rv",
+                "compound_name": "CHEMBL5290347",
+                "assay": "MABA",
+            },
+        ),
+        lx.data.Extraction(
+            extraction_class="bioactivity",
+            extraction_text="2.4",
+            attributes={
+                "value": "2.4",
+                "unit": "µM",
+                "assay_type": "EC90",
+                "compound_name": "CHEMBL5290347",
+                "assay": "THP-1",
+            },
+        ),
+        lx.data.Extraction(
+            extraction_class="bioactivity",
+            extraction_text="18.5",
+            attributes={
+                "value": "18.5",
+                "unit": "µM",
+                "assay_type": "CC50",
+                "compound_name": "CHEMBL5290347",
+                "assay": "Vero",
+            },
+        ),
+    ],
+)
+
+# Strain panel: organism column headers are the strain, a protein column the
+# target, a cytotoxicity column the assay; a column dosed with a second drug
+# names the combination partner. The disease stays a disease entity.
+_TB_EXAMPLE_7 = lx.data.ExampleData(
+    text=(
+        "Table 3. Antibacterial activity against hospital-acquired pneumonia isolates\n\n"
+        "| Cpd | ID | E. coli ATCC 25922 | K. pneumoniae BAA-1705 | A. baumannii ATCC 19606 "
+        "| ATCC 19606 + polymyxin B | HepG2 | hERG |\n"
+        "| --- | --- | --- | --- | --- | --- | --- | --- |\n"
+        "| 9c | CHEMBL5311027 | 0.25 | 1 | >64 | 2 | >100 | 12.5 |\n\n"
+        "MIC (µg/mL); polymyxin B at 0.5 µg/mL. HepG2: CC50 (µM). hERG: IC50 (µM)."
+    ),
+    extractions=[
+        lx.data.Extraction(
+            extraction_class="disease",
+            extraction_text="hospital-acquired pneumonia",
+            attributes={"therapeutic_area": "infectious disease"},
+        ),
+        lx.data.Extraction(
+            extraction_class="compound_name",
+            extraction_text="CHEMBL5311027",
+            attributes={"synonyms": "9c"},
+        ),
+        lx.data.Extraction(
+            extraction_class="bioactivity",
+            extraction_text="0.25",
+            attributes={
+                "value": "0.25",
+                "unit": "µg/mL",
+                "assay_type": "MIC",
+                "strain": "E. coli ATCC 25922",
+                "compound_name": "CHEMBL5311027",
+            },
+        ),
+        lx.data.Extraction(
+            extraction_class="bioactivity",
+            extraction_text="1",
+            attributes={
+                "value": "1",
+                "unit": "µg/mL",
+                "assay_type": "MIC",
+                "strain": "K. pneumoniae BAA-1705",
+                "compound_name": "CHEMBL5311027",
+            },
+        ),
+        lx.data.Extraction(
+            extraction_class="bioactivity",
+            extraction_text=">64",
+            attributes={
+                "value": ">64",
+                "unit": "µg/mL",
+                "assay_type": "MIC",
+                "strain": "A. baumannii ATCC 19606",
+                "compound_name": "CHEMBL5311027",
+            },
+        ),
+        lx.data.Extraction(
+            extraction_class="bioactivity",
+            extraction_text="2",
+            attributes={
+                "value": "2",
+                "unit": "µg/mL",
+                "assay_type": "MIC",
+                "strain": "ATCC 19606",
+                "combination": "polymyxin B",
+                "compound_name": "CHEMBL5311027",
+            },
+        ),
+        lx.data.Extraction(
+            extraction_class="bioactivity",
+            extraction_text=">100",
+            attributes={
+                "value": ">100",
+                "unit": "µM",
+                "assay_type": "CC50",
+                "compound_name": "CHEMBL5311027",
+                "assay": "HepG2",
+            },
+        ),
+        lx.data.Extraction(
+            extraction_class="bioactivity",
+            extraction_text="12.5",
+            attributes={
+                "value": "12.5",
+                "unit": "µM",
+                "assay_type": "IC50",
+                "compound_name": "CHEMBL5311027",
+                "target": "hERG",
+            },
+        ),
+    ],
+)
+
 TB_EXAMPLES: list[lx.data.ExampleData] = [
     _TB_EXAMPLE_1,
     _TB_EXAMPLE_2,
     _TB_EXAMPLE_3,
     _TB_EXAMPLE_4,
     _TB_EXAMPLE_5,
+    _TB_EXAMPLE_6,
+    _TB_EXAMPLE_7,
 ]
 
 TB_CHEMISTRY_EXAMPLES: list[lx.data.ExampleData] = [

{structflo_ner-0.5.0 → structflo_ner-0.7.0}/structflo/ner/_mapping.py

@@ -29,6 +29,10 @@ def extraction_to_entity(extraction: lx.data.Extraction) -> NEREntity:
     attributes: dict[str, str] = {}
     if extraction.attributes:
         for key, value in extraction.attributes.items():
+            # Strict-schema providers (OpenAI) must emit every attribute key, so an
+            # unstated one arrives as null; leave it out rather than store "None".
+            if value is None:
+                continue
             if isinstance(value, list):
                 attributes[key] = ", ".join(value)
             else:

structflo_ner-0.7.0/structflo/ner/_prompts.py

@@ -0,0 +1,164 @@
+"""Prompt strings for each built-in EntityProfile."""
+
+# Shared by every prompt that extracts bioactivity values. The attribute names
+# must match the keys used in the few-shot examples: langextract derives the
+# output schema from those keys, not from this text.
+_BIOACTIVITY_ATTRIBUTES = (
+    "For each bioactivity value capture:\n"
+    "- value: the number as reported, keeping qualifiers ('>20.0', '<1.0') and dropping "
+    "footnote markers ('0.3*' → '0.3').\n"
+    "- unit: as written (nM, µM, µg/mL, mg/kg, %, h).\n"
+    "- assay_type: the endpoint only, in plain characters (CC₅₀ → CC50). Common endpoints: "
+    "IC50, IC90, EC50, EC90, pIC50, Ki, Kd, '% inhibition at <concentration>'; "
+    "MIC, MIC50, MIC90, MIC99, MBC, log10 CFU reduction; CC50, TC50, SI (selectivity index); "
+    "GI50, TGI, LC50; DC50, Dmax (degraders); ED50, ED90, %TGI, T/C (in vivo); "
+    "PRR (parasite reduction ratio), parasite clearance half-life.\n"
+    "- compound_name: the compound the value belongs to, by its name or identifier "
+    "(e.g. 'bedaquiline', 'BTZ043', 'CHEMBL1234', 'SACC-3000'); a document-local label "
+    "('7a', 'Compound 9b') only when nothing else names the compound.\n"
+)
+_ASSAY_ATTRIBUTE = (
+    "- assay: the assay, cell line or read-out the value was measured in, e.g. 'MABA', "
+    "'LORA', 'HepG2 MTT', 'THP-1 macrophage infection assay', 'asexual blood-stage', "
+    "'liver-stage', 'gametocyte', 'NCI-60', 'hERG', 'P. berghei 4-day suppressive test'. "
+    "null when no assay is stated.\n"
+)
+# TB gives each kind of context its own slot: what the value was measured in
+# (assay), against (target, strain), and with (combination).
+_TB_SLOT_ATTRIBUTES = (
+    "- assay: the assay system the value was measured in: a cell line, format or read-out "
+    "('MABA', 'LORA', 'HepG2 MTT', 'THP-1 macrophage infection assay', 'asexual blood-stage', "
+    "'AlphaScreen', 'FP', 'NCI-60', 'P. berghei 4-day suppressive test'). null when none is "
+    "stated.\n"
+    "- target: the protein the value was measured against: an enzyme, receptor, channel or "
+    "other molecular target, as written ('InhA', 'DprE1', 'hERG', 'KasA', 'PDE4B', "
+    "'BACE1'). null for whole-cell, organism and cytotoxicity values unless one is named.\n"
+    "- strain: the organism the value was measured against, as written: species, strain, "
+    "isolate or virus (H37Rv, M. tuberculosis, E. coli ATCC 25922, A. baumannii ATCC 19606, "
+    "Pf3D7, Dd2, S. mansoni, HIV-1, norovirus). An organism always goes here, never in assay or target, even "
+    "when it heads a table column or is all that names the assay. null when not stated.\n"
+    "- combination: a second compound dosed together with the one measured (a combination "
+    "or potentiation partner), as written ('avibactam', 'polymyxin B 0.5 µg/mL'). Not a "
+    "reference compound the value is relative to ('% of isoproterenol') or an agonist, substrate "
+    "or tracer the assay uses. null for a single agent.\n"
+)
+_BIOACTIVITY_TABLES = (
+    "In a data table, extract every cell value as its own bioactivity. An ID or registry "
+    "column, when present, supplies the compound_name and the row label is only its synonym; "
+    "otherwise the row label is the compound_name. A column header fills the attribute for "
+    "what it names (an assay or cell line, a protein target, an organism or strain), and the "
+    "endpoint and unit often come from the caption or a footnote ('CC50 in µM')."
+)
+
+CHEMISTRY_PROMPT = (
+    "Extract chemical entities from the text. "
+    "Include: compound names (generic names, IUPAC names, code names like 'Compound 5b', "
+    "brand names, and abbreviations), SMILES strings (the exact SMILES notation as written), "
+    "CAS registry numbers (e.g. '50-78-2'), and molecular formulas (e.g. 'C9H8O4'). "
+    "Do not infer or generate SMILES — only extract them if explicitly present in the text."
+)
+
+BIOLOGY_PROMPT = (
+    "Extract biological target entities from the text. "
+    "Include: protein targets (e.g. 'EGFR', 'CDK4/6', 'PD-L1'), gene names (e.g. 'KRAS', 'TP53', "
+    "'BRCA1'), receptor names, enzyme names, and pathway names. "
+    "For each target, capture the gene symbol if mentioned alongside a protein name. "
+    "Capture the organism context if specified (e.g. 'human', 'mouse')."
+)
+
+BIOACTIVITY_PROMPT = (
+    "Extract bioactivity measurements and assay data from the text: potency, selectivity, "
+    "cytotoxicity and efficacy values.\n"
+    + _BIOACTIVITY_ATTRIBUTES
+    + _ASSAY_ATTRIBUTE
+    + _BIOACTIVITY_TABLES
+    + "\nAlso extract assay descriptions: cell lines used (e.g. 'HeLa', 'A549'), assay formats "
+    "(e.g. 'cell viability', 'binding assay', 'enzymatic assay'), and organisms."
+)
+
+DISEASE_PROMPT = (
+    "Extract disease names and clinical indications from the text. "
+    "Include: cancer types (e.g. 'non-small cell lung cancer', 'AML', 'NSCLC'), "
+    "non-oncology diseases (e.g. 'type 2 diabetes', 'rheumatoid arthritis'), "
+    "and therapeutic areas (e.g. 'oncology', 'CNS'). "
+    "Capture both full names and abbreviations. "
+    "For each disease, note the therapeutic area if discernible from context."
+)
+
+FULL_PROMPT = (
+    "Extract all drug discovery entities from the text. "
+    "This includes:\n"
+    "- Chemical entities: compound names (generic, IUPAC, code names, brand names), "
+    "SMILES strings (only if explicitly written), CAS numbers, molecular formulas.\n"
+    "- Biological targets: protein names, gene names, receptor names, enzyme names, pathways.\n"
+    "- Bioactivity data: potency, selectivity, cytotoxicity and efficacy values "
+    "(attributes below).\n"
+    "- Assay information: cell lines, assay formats, experimental organisms.\n"
+    "- Diseases and indications: cancer types, disease names, therapeutic areas.\n"
+    "- Mechanisms of action: binding modes, inhibition types, selectivity descriptions.\n"
+    + _BIOACTIVITY_ATTRIBUTES
+    + _ASSAY_ATTRIBUTE
+    + _BIOACTIVITY_TABLES
+    + "\nExtract only what is explicitly stated; do not infer or generate values."
+)
+
+# ── Tuberculosis early drug discovery prompts ──────────────────────────
+
+TB_PROMPT = (
+    "Extract drug discovery entities from this tuberculosis research text.\n\n"
+    + _BIOACTIVITY_ATTRIBUTES
+    + _TB_SLOT_ATTRIBUTES
+    + _BIOACTIVITY_TABLES
+    + "\n\n"
+    "DISAMBIGUATION RULES:\n"
+    "- Mycobacterial proteins (e.g. ClpC1, DprE1, InhA, AtpE, MmpL3, QcrB) "
+    "are biological targets, NOT compounds.\n"
+    "- Rv locus tags (Rv3790, Rv1484), UniProt IDs (P9WPS1), and PDB codes "
+    "are accession_number, not target or gene_name.\n"
+    "- Compound registry and programme identifiers (CHEMBL4521987, ZINC000012345678, "
+    "DB00945, SACC-3060, TBDA-01187, GSK3036656) are compound_name; only Rv locus "
+    "tags, UniProt IDs, PDB codes and RefSeq IDs are accession_number.\n"
+    "- A compound with both a document-local label (a number with an optional letter: 7a, "
+    "12, 'Compound 9b') and a longer identifier (registry ID; programme, partner or vendor "
+    "code: CHEMBL…, SACC-…, a ChemBridge number) is extracted once, as the identifier, with "
+    "the label in synonyms; its bioactivities use the identifier as compound_name. A label "
+    "is the compound_name only when the document gives nothing else.\n"
+    "- Enzyme descriptions like 'enoyl-ACP reductase' are product, not target.\n"
+    "- 'cell wall', 'lipid metabolism' are functional_category, not mechanism_of_action.\n"
+    "- 'fragment screening', 'biochemical assay' are screening_method, not assay.\n"
+    "- A disease or infection is a disease entity wherever it is named, in prose, a heading "
+    "or a table cell (tuberculosis, HIV infection, malaria, schistosomiasis, glioblastoma in 'U87 (glioblastoma)'), "
+    "even when its organism also fills a bioactivity's strain.\n"
+    "- Use target for proteins in a drug-targeting context, gene_name for loci, "
+    "protein_name for non-drug-target proteins.\n\n"
+    "Extract only what is explicitly stated; do not infer or generate values."
+)
+
+TB_CHEMISTRY_PROMPT = (
+    "Extract chemical entities from this tuberculosis drug discovery text. "
+    "Include compound names, SMILES (only if explicitly written), CAS numbers, "
+    "and molecular formulas. "
+    "Mycobacterial proteins (ClpC1, DprE1, InhA, AtpE, etc.) are NOT compounds. "
+    "Extract only what is explicitly stated; do not infer or generate values."
+)
+
+TB_BIOLOGY_PROMPT = (
+    "Extract biological entities from this tuberculosis research text. "
+    "Use target for proteins in a drug-targeting context, gene_name for loci, "
+    "protein_name for non-drug-target proteins. "
+    "Rv locus tags and UniProt IDs are accession_number. "
+    "Enzyme descriptions (e.g. 'enoyl-ACP reductase') are product. "
+    "Protein functional categories (e.g. 'cell wall', 'lipid metabolism') "
+    "are functional_category. "
+    "Extract only what is explicitly stated; do not infer or generate values."
+)
+
+# Framing for NERExtractor.extract(context=...), rendered by langextract ahead of
+# the examples.
+DOCUMENT_CONTEXT = (
+    "Document context, from elsewhere in the same document. It is not the text to extract "
+    "from: take no entity and no value from it, and do not pair a value in the text with a "
+    "number or claim made here. Use it only to fill a bioactivity attribute the text leaves "
+    "unstated, such as the target or organism of a table, when the context names one for "
+    "the whole document.\n"
+)

{structflo_ner-0.5.0 → structflo_ner-0.7.0}/structflo/ner/extractor.py

@@ -3,15 +3,43 @@
 from __future__ import annotations
 
 import logging
+import re
 
 import langextract as lx
 
+from structflo.ner import _prompts
 from structflo.ner._entities import NERResult
 from structflo.ner._mapping import annotated_doc_to_result
 from structflo.ner.profiles import FULL, EntityProfile
 
 logger = logging.getLogger(__name__)
 
+# A table cell holding only a value: '0.94', '>32', '2.4*', '0.62µM'.
+_VALUE_CELL = re.compile(r"^[<>≤≥~=]*\s*\d[\d,]*\.?\d*\s*\*?\s*(%|[nµu]M|µg/mL|mg/kg)?$")
+
+
+def _bioactivity_count(doc: lx.data.AnnotatedDocument) -> int:
+    return sum(1 for ext in doc.extractions if ext.extraction_class == "bioactivity")
+
+
+def _table_values_dropped(text: str, doc: lx.data.AnnotatedDocument) -> bool:
+    """True when fewer than half of a page's table values came back as bioactivities.
+
+    Models now and then return a complete, valid answer that lists a table's
+    compounds but leaves out its values. On the ChEMBL-gold decks this fires on
+    ~2% of pages and catches ~60% of such dropouts.
+    """
+    # ponytail: markdown tables only; values in prose and qualitative results
+    # are not checked, so those dropouts still pass silently.
+    cells = sum(
+        1
+        for line in text.splitlines()
+        if line.lstrip().startswith("|")
+        for cell in line.strip().strip("|").split("|")
+        if _VALUE_CELL.match(cell.strip())
+    )
+    return cells >= 5 and _bioactivity_count(doc) < 0.5 * cells
+
 
 class NERExtractor:
     """Extract drug discovery entities from text with zero configuration.
@@ -90,12 +118,17 @@ class NERExtractor:
         self,
         text: str | list[str],
         profile: EntityProfile | None = None,
+        context: str | None = None,
     ) -> NERResult | list[NERResult]:
         """Extract drug discovery entities from text.
 
         Args:
             text: Input text (or list of texts) to process.
             profile: Override the default profile for this call only.
+            context: Text from elsewhere in the same document (its title or
+                summary). The model reads it to fill attributes the text leaves
+                unstated, such as the target of a table whose target is named
+                only on another page; no entities are extracted from it.
 
         Returns:
             A :class:`NERResult` for a single string input, or a list of
@@ -107,7 +140,13 @@ class NERExtractor:
 
         results = []
         for single_text in texts:
-            doc = self._run_extraction(single_text, active_profile)
+            doc = self._run_extraction(single_text, active_profile, context)
+            if "bioactivity" in active_profile.entity_classes and _table_values_dropped(
+                single_text, doc
+            ):
+                logger.warning("Table values missing from extraction; retrying once.")
+                retry = self._run_extraction(single_text, active_profile, context)
+                doc = max(doc, retry, key=_bioactivity_count)
             results.append(annotated_doc_to_result(doc, single_text))
 
         return results if is_batch else results[0]
@@ -198,6 +237,7 @@ class NERExtractor:
         self,
         text: str,
         profile: EntityProfile,
+        context: str | None = None,
     ) -> lx.data.AnnotatedDocument:
         """Call lx.extract and return the AnnotatedDocument."""
         examples = self._build_examples(profile)
@@ -206,6 +246,8 @@ class NERExtractor:
         kwargs: dict = dict(self._langextract_kwargs)
         kwargs.setdefault("use_schema_constraints", True)
         kwargs.setdefault("show_progress", False)
+        if context:
+            kwargs["additional_context"] = _prompts.DOCUMENT_CONTEXT + context
 
         if self._provider is not None:
             # Explicit provider → deterministic routing via ModelConfig.
@@ -236,4 +278,29 @@ class NERExtractor:
 
         # Post-process: drop extractions with hallucinated entity classes
         allowed = set(profile.entity_classes)
-        return self._filter_extractions(doc, allowed)
+        doc = self._filter_extractions(doc, allowed)
+        if context:
+            doc = self._drop_context_copies(doc, context)
+        return doc
+
+    @staticmethod
+    def _drop_context_copies(
+        doc: lx.data.AnnotatedDocument,
+        context: str,
+    ) -> lx.data.AnnotatedDocument:
+        """Drop extractions the model took from the context instead of the text.
+
+        The prompt forbids it, but models still extract from the context now and
+        then. Such an extraction does not align to the text, and its text occurs
+        in the context but not in the text (an unaligned one that is in the text
+        is a missed alignment, not a copy).
+        """
+        ctx, text = context.casefold(), doc.text.casefold()
+        kept = [
+            ext
+            for ext in doc.extractions
+            if ext.char_interval is not None
+            or ext.extraction_text.casefold() not in ctx
+            or ext.extraction_text.casefold() in text
+        ]
+        return lx.data.AnnotatedDocument(text=doc.text, extractions=kept)

{structflo_ner-0.5.0 → structflo_ner-0.7.0}/structflo/ner/fast/README.md

@@ -66,7 +66,7 @@ so a deployment's own gazetteer decides which ones apply:
 | Seed | Auto-derived Pattern | Matches |
 |---|---|---|
 | `Rv0005` | `Rv\d{4}[c]?` | All Rv locus tags |
-| `MT0005` | `MT\w+` | Mycobrowser IDs |
+| `MT0005` | `MT[A-Z]{0,2}\d{2,}\w*` | Mycobrowser IDs (not MTT, MTX) |
 | `P9WGR1` | `[OPQ][0-9][A-Z0-9]{3}[0-9]` | UniProt accessions |
 | `4TZK` | `(?=[A-Z0-9]{0,3}[A-Z])[1-9][A-Z0-9]{3}` | PDB codes |
 | `WP_003407354` | `WP_\d+` | NCBI RefSeq proteins |

{structflo_ner-0.5.0 → structflo_ner-0.7.0}/structflo/ner/fast/_loader.py

@@ -89,8 +89,13 @@ _COMPOUND_ID_PATTERNS: list[IdPattern] = [
 _ACCESSION_PATTERNS: list[tuple[re.Pattern[str], re.Pattern[str], str]] = [
     # Rv locus tags: Rv0005, Rv3854c
     (re.compile(r"^Rv\d{4}[c]?$"), re.compile(r"\bRv\d{4}[c]?\b"), "Rv locus tag"),
-    # Mycobrowser MT IDs: MT0005, MTCI00.01
-    (re.compile(r"^MT\w+$"), re.compile(r"\bMT\w+\b"), "Mycobrowser ID"),
+    # Mycobrowser MT IDs: MT0005, MT18B_0001, MTB000001. The digit run keeps
+    # MTT (viability assay), MTX and MTD from matching.
+    (
+        re.compile(r"^MT[A-Z]{0,2}\d{2,}\w*$"),
+        re.compile(r"\bMT[A-Z]{0,2}\d{2,}\w*\b"),
+        "Mycobrowser ID",
+    ),
     # UniProt accessions: P9WGR1, O53617
     (
         re.compile(r"^[OPQ][0-9][A-Z0-9]{3}[0-9]$"),

{structflo_ner-0.5.0 → structflo_ner-0.7.0}/structflo/ner/fast/_matcher.py

@@ -275,8 +275,9 @@ def _build_position_map(original: str, normalized: str) -> list[int]:
 
         # Advance original pointer to find the matching character
         while orig_idx < len(original):
-            orig_lower = original[orig_idx].lower()
-            if orig_lower == norm_char:
+            # normalize() the char too: a newline became " " and an en dash "-",
+            # and comparing raw chars would desync every position after them.
+            if (normalize(original[orig_idx]) or " ") == norm_char:
                 position_map.append(orig_idx)
                 orig_idx += 1
                 break

{structflo_ner-0.5.0 → structflo_ner-0.7.0}/tests/test_extractor.py

@@ -7,6 +7,7 @@ from unittest.mock import MagicMock
 import langextract as lx
 
 from structflo.ner import (
+    BIOACTIVITY,
     BIOLOGY,
     CHEMISTRY,
     FULL,
@@ -121,6 +122,50 @@ class TestNERExtractorExtract:
         result = extractor.extract("My source text")
         assert result.source_text == "My source text"
 
+    _TABLE = "| cpd | MIC | CC50 |\n| --- | --- | --- |\n" + "".join(
+        f"| C{i} | {i}.5 | >50 |\n" for i in range(5)
+    )
+
+    def _bio_doc(self, n: int) -> lx.data.AnnotatedDocument:
+        return _make_annotated_doc(
+            [
+                lx.data.Extraction(extraction_class="bioactivity", extraction_text=f"{i}.5")
+                for i in range(n)
+            ]
+        )
+
+    def test_retries_once_when_table_values_are_missing(self):
+        extractor = NERExtractor(profile=TB)
+        extractor._run_extraction = MagicMock(side_effect=[self._bio_doc(0), self._bio_doc(10)])
+        result = extractor.extract(self._TABLE)
+        assert extractor._run_extraction.call_count == 2
+        assert len(result.bioactivities) == 10
+
+    def test_no_retry_when_table_values_came_back(self):
+        extractor = NERExtractor(profile=TB)
+        extractor._run_extraction = MagicMock(return_value=self._bio_doc(10))
+        extractor.extract(self._TABLE)
+        assert extractor._run_extraction.call_count == 1
+
+    def test_no_retry_for_profiles_without_bioactivity(self):
+        extractor = NERExtractor(profile=CHEMISTRY)
+        extractor._run_extraction = MagicMock(return_value=self._bio_doc(0))
+        extractor.extract(self._TABLE)
+        assert extractor._run_extraction.call_count == 1
+
+    def test_null_attributes_are_absent_not_none_strings(self):
+        # OpenAI strict mode returns every schema key, null when unstated.
+        extractor = self._extractor_with_mock(
+            [
+                lx.data.Extraction(
+                    extraction_class="bioactivity",
+                    extraction_text="IC50 = 3 nM",
+                    attributes={"value": "3", "strain": None, "assay": None},
+                )
+            ]
+        )
+        assert extractor.extract("text").bioactivities[0].attributes == {"value": "3"}
+
 
 class TestProviderRouting:
     """Explicit provider selection routes deterministically via ModelConfig,
@@ -180,6 +225,24 @@ class TestProviderRouting:
         assert pk["base_url"] == "http://ollama:11434"
         assert pk["num_ctx"] == 8192
 
+    def test_context_reaches_langextract_framed(self, monkeypatch):
+        from structflo.ner import extractor as extractor_mod
+
+        captured = {}
+
+        def fake_extract(**kwargs):
+            captured.update(kwargs)
+            return _make_annotated_doc([])
+
+        monkeypatch.setattr(extractor_mod.lx, "extract", fake_extract)
+        NERExtractor().extract("IC50s table", context="Deck on InhA inhibitors.")
+        ctx = captured["additional_context"]
+        assert ctx.startswith("Document context") and ctx.endswith("Deck on InhA inhibitors.")
+
+    def test_no_context_passes_no_additional_context(self, monkeypatch):
+        captured = self._run_and_capture(monkeypatch)
+        assert "additional_context" not in captured
+
     def test_cloud_provider_kwargs_carry_api_key(self, monkeypatch):
         captured = self._run_and_capture(
             monkeypatch, provider="openai", model_id="gpt-4o", api_key="sk-test"
@@ -234,6 +297,27 @@ class TestFilterExtractions:
         assert len(filtered.extractions) == 0
 
 
+class TestDropContextCopies:
+    def test_drops_only_unaligned_extractions_found_in_context(self):
+        aligned = lx.data.Extraction(
+            extraction_class="target",
+            extraction_text="InhA",
+            char_interval=lx.data.CharInterval(start_pos=0, end_pos=4),
+        )
+        copied = lx.data.Extraction(extraction_class="compound_name", extraction_text="Isoniazid")
+        unaligned = lx.data.Extraction(extraction_class="compound_name", extraction_text="7a")
+        # In the context too, but also on the page: a missed alignment, not a copy.
+        on_page = lx.data.Extraction(extraction_class="disease", extraction_text="malaria")
+        doc = lx.data.AnnotatedDocument(
+            text="InhA IC50s for 7a; malaria panel",
+            extractions=[aligned, copied, unaligned, on_page],
+        )
+        kept = NERExtractor._drop_context_copies(
+            doc, "InhA inhibitors that match isoniazid, and a malaria screen"
+        )
+        assert kept.extractions == [aligned, unaligned, on_page]
+
+
 class TestBuildExamples:
     def test_extra_examples_appended(self):
         extra = lx.data.ExampleData(text="extra", extractions=[])
@@ -284,8 +368,33 @@ class TestTBProfile:
         for cls in TB_BIOLOGY.entity_classes:
             assert cls in TB.entity_classes
 
+    def test_bioactivity_schema_has_assay(self):
+        """langextract builds the output schema from example attribute keys, so an
+        `assay` slot exists only if some bioactivity example carries one."""
+        for profile in (TB, BIOACTIVITY, FULL):
+            keys = {
+                k
+                for ex in profile.examples
+                for e in ex.extractions
+                if e.extraction_class == "bioactivity"
+                for k in (e.attributes or {})
+            }
+            assert {"value", "unit", "assay_type", "compound_name", "assay"} <= keys, profile.name
+
+    def test_tb_bioactivity_schema_has_context_slots(self):
+        """TB gives what a value was measured against (target, strain) and with
+        (combination) their own slots, apart from assay."""
+        keys = {
+            k
+            for ex in TB.examples
+            for e in ex.extractions
+            if e.extraction_class == "bioactivity"
+            for k in (e.attributes or {})
+        }
+        assert {"target", "strain", "combination"} <= keys
+
     def test_tb_examples_count(self):
-        assert len(TB.examples) == 5
+        assert len(TB.examples) == 7
         assert len(TB_CHEMISTRY.examples) == 2
         assert len(TB_BIOLOGY.examples) == 2
 

{structflo_ner-0.5.0 → structflo_ner-0.7.0}/tests/test_fast.py

@@ -180,6 +180,13 @@ class TestDeriveIdPatterns:
         assert not pdb.regex.search("in 2019 the trial")
         assert not pdb.regex.search("n = 1234")
 
+    def test_mycobrowser_pattern_needs_locus_digits(self):
+        """MTT (viability assay), MTX and MTD are not Mycobrowser loci."""
+        mt = next(p for p in _accessions(["MT18B_0001"]) if p.description == "Mycobrowser ID")
+        for locus in ["MT0005", "MT18B_0001", "MTB000001"]:
+            assert mt.regex.fullmatch(locus), locus
+        assert not mt.regex.search("HepG2 MTT, MTX and MTD read-outs")
+
     def test_cas_check_digit_rejects_dates(self):
         cas = next(p for p in derive_id_patterns({}) if p.description == "CAS number")
         assert cas.validate("50-78-2")  # aspirin
@@ -246,6 +253,13 @@ class TestGazetteerMatcher:
         m = inha_matches[0]
         assert text[m.char_start : m.char_end] == "InhA"
 
+    def test_char_offsets_survive_newlines(self):
+        """normalize() folds a newline into a space; offsets after it must not drift."""
+        matcher = self._simple_matcher(fuzzy_threshold=0)
+        text = "Targets:\nINHA and\n\nDPRE1."
+        found = {m.canonical: text[m.char_start : m.char_end] for m in matcher.match(text)}
+        assert found == {"InhA": "INHA", "DprE1": "DPRE1"}
+
     def test_regex_accession_matching(self):
         matcher = GazetteerMatcher(
             gazetteers={"target": ["InhA"]},

{structflo_ner-0.5.0 → structflo_ner-0.7.0}/uv.lock

@@ -917,7 +917,7 @@ wheels = [
 
 [[package]]
 name = "langextract"
-version = "1.6.0"
+version = "1.7.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "absl-py" },
@@ -940,9 +940,9 @@ dependencies = [
     { name = "tqdm" },
     { name = "typing-extensions" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/49/c2/30216c49a53419bf3c124cb3ea6daba95a617a62e5ea4964b9fb11784f1e/langextract-1.6.0.tar.gz", hash = "sha256:ac6412152b173bfb0d6b4e08bf6ace7319b1e734c4f9c6f33d39f0c5915e640d", size = 142637, upload-time = "2026-07-02T06:24:00.632Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/e3/e7/6cf877cdd3d5d3a479c0688ff2b63c665c4319ae35ba00734c9fef36a1d2/langextract-1.7.0.tar.gz", hash = "sha256:cb6dc515583b0fdc0f9eff9dce61b2a3c7c84b0668add4dd4e7be364f4c188ba", size = 147514, upload-time = "2026-09-13T18:58:06.376Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/d7/e0/3475ffd1c7e2a4a49f3d8c186c27869a53c710335742964707dbc499590e/langextract-1.6.0-py3-none-any.whl", hash = "sha256:ffa2d584c29ea25c73a2723cec756f21a76799e8ab76a1d9d92841e27504b057", size = 150147, upload-time = "2026-07-02T06:23:59.25Z" },
+    { url = "https://files.pythonhosted.org/packages/ae/ec/2883a93872701045971ea77e30434d72556a557f188df575b154d179854c/langextract-1.7.0-py3-none-any.whl", hash = "sha256:908f8ec696ab13578cdc3862efcf1a4fda929ae426410d99e86977d50e52136d", size = 154297, upload-time = "2026-09-13T18:58:04.896Z" },
 ]
 
 [[package]]
@@ -2170,7 +2170,7 @@ dev = [
 
 [package.metadata]
 requires-dist = [
-    { name = "langextract", specifier = ">=1.6.0" },
+    { name = "langextract", specifier = ">=1.7.0" },
     { name = "pandas", marker = "extra == 'dataframe'", specifier = ">=1.5" },
     { name = "pyyaml", specifier = ">=6.0" },
     { name = "rapidfuzz", specifier = ">=3.0" },

structflo_ner-0.5.0/structflo/ner/_prompts.py

@@ -1,95 +0,0 @@
-"""Prompt strings for each built-in EntityProfile."""
-
-CHEMISTRY_PROMPT = (
-    "Extract chemical entities from the text. "
-    "Include: compound names (generic names, IUPAC names, code names like 'Compound 5b', "
-    "brand names, and abbreviations), SMILES strings (the exact SMILES notation as written), "
-    "CAS registry numbers (e.g. '50-78-2'), and molecular formulas (e.g. 'C9H8O4'). "
-    "Do not infer or generate SMILES — only extract them if explicitly present in the text."
-)
-
-BIOLOGY_PROMPT = (
-    "Extract biological target entities from the text. "
-    "Include: protein targets (e.g. 'EGFR', 'CDK4/6', 'PD-L1'), gene names (e.g. 'KRAS', 'TP53', "
-    "'BRCA1'), receptor names, enzyme names, and pathway names. "
-    "For each target, capture the gene symbol if mentioned alongside a protein name. "
-    "Capture the organism context if specified (e.g. 'human', 'mouse')."
-)
-
-BIOACTIVITY_PROMPT = (
-    "Extract bioactivity measurements and assay data from the text. "
-    "Include: potency values (IC50, MIC, EC50, Ki, Kd, GI50, CC50), selectivity ratios, "
-    "percent inhibition values, and Hill coefficients. "
-    "For each value, capture the numeric value, unit (nM, µM, mM, µg/mL, ng/mL), and measurement type. "
-    "Also capture the compound — its name, code name, or identifier (e.g. 'Compound 7', 'BTZ043', "
-    "'CHEMBL1234', 'SACC-3000') — that the measurement applies to, as a 'compound_name' attribute, "
-    "if it can be identified from the surrounding context. "
-    "Also extract assay descriptions: cell lines used (e.g. 'HeLa', 'A549'), assay formats "
-    "(e.g. 'cell viability', 'binding assay', 'enzymatic assay'), and organisms."
-)
-
-DISEASE_PROMPT = (
-    "Extract disease names and clinical indications from the text. "
-    "Include: cancer types (e.g. 'non-small cell lung cancer', 'AML', 'NSCLC'), "
-    "non-oncology diseases (e.g. 'type 2 diabetes', 'rheumatoid arthritis'), "
-    "and therapeutic areas (e.g. 'oncology', 'CNS'). "
-    "Capture both full names and abbreviations. "
-    "For each disease, note the therapeutic area if discernible from context."
-)
-
-FULL_PROMPT = (
-    "Extract all drug discovery entities from the text. "
-    "This includes:\n"
-    "- Chemical entities: compound names (generic, IUPAC, code names, brand names), "
-    "SMILES strings (only if explicitly written), CAS numbers, molecular formulas.\n"
-    "- Biological targets: protein names, gene names, receptor names, enzyme names, pathways.\n"
-    "- Bioactivity data: IC50, MIC, EC50, Ki, Kd, and other potency/selectivity measurements "
-    "with their numeric values, units, and the compound name or identifier they apply to "
-    "(captured as a 'compound_name' attribute).\n"
-    "- Assay information: cell lines, assay formats, experimental organisms.\n"
-    "- Diseases and indications: cancer types, disease names, therapeutic areas.\n"
-    "- Mechanisms of action: binding modes, inhibition types, selectivity descriptions.\n"
-    "Extract only what is explicitly stated; do not infer or generate values."
-)
-
-# ── Tuberculosis early drug discovery prompts ──────────────────────────
-
-TB_PROMPT = (
-    "Extract drug discovery entities from this tuberculosis research text.\n\n"
-    "For each bioactivity measurement (IC50, MIC, MIC90, EC50, CC50, etc.), capture the "
-    "compound name, code, or identifier it belongs to as a 'compound_name' attribute "
-    "(e.g. 'BTZ043 showed a MIC of 1 ng/mL' → compound_name: 'BTZ043').\n\n"
-    "DISAMBIGUATION RULES:\n"
-    "- Mycobacterial proteins (e.g. ClpC1, DprE1, InhA, AtpE, MmpL3, QcrB) "
-    "are biological targets, NOT compounds.\n"
-    "- Rv locus tags (Rv3790, Rv1484), UniProt IDs (P9WPS1), and PDB codes "
-    "are accession_number, not target or gene_name.\n"
-    "- Compound registry and programme identifiers (CHEMBL4521987, ZINC000012345678, "
-    "DB00945, SACC-3060, TBDA-01187, GSK3036656) are compound_name; only Rv locus "
-    "tags, UniProt IDs, PDB codes and RefSeq IDs are accession_number.\n"
-    "- Enzyme descriptions like 'enoyl-ACP reductase' are product, not target.\n"
-    "- 'cell wall', 'lipid metabolism' are functional_category, not mechanism_of_action.\n"
-    "- 'fragment screening', 'biochemical assay' are screening_method, not assay.\n"
-    "- Use target for proteins in a drug-targeting context, gene_name for loci, "
-    "protein_name for non-drug-target proteins.\n\n"
-    "Extract only what is explicitly stated; do not infer or generate values."
-)
-
-TB_CHEMISTRY_PROMPT = (
-    "Extract chemical entities from this tuberculosis drug discovery text. "
-    "Include compound names, SMILES (only if explicitly written), CAS numbers, "
-    "and molecular formulas. "
-    "Mycobacterial proteins (ClpC1, DprE1, InhA, AtpE, etc.) are NOT compounds. "
-    "Extract only what is explicitly stated; do not infer or generate values."
-)
-
-TB_BIOLOGY_PROMPT = (
-    "Extract biological entities from this tuberculosis research text. "
-    "Use target for proteins in a drug-targeting context, gene_name for loci, "
-    "protein_name for non-drug-target proteins. "
-    "Rv locus tags and UniProt IDs are accession_number. "
-    "Enzyme descriptions (e.g. 'enoyl-ACP reductase') are product. "
-    "Protein functional categories (e.g. 'cell wall', 'lipid metabolism') "
-    "are functional_category. "
-    "Extract only what is explicitly stated; do not infer or generate values."
-)