aurelian 0.1.0__py3-none-any.whl

aurelian/agents/phenopackets/phenopackets_evals.py

@@ -0,0 +1,99 @@
+"""
+Evaluation module for the Phenopackets agent.
+
+This module implements evaluations for the Phenopackets agent using the pydantic-ai-evals framework.
+"""
+import asyncio
+import sys
+from typing import Optional, Any, Dict, Callable, Awaitable
+
+from aurelian.evaluators.model import MetadataDict, metadata
+from aurelian.evaluators.substring_evaluator import SubstringEvaluator
+from pydantic_evals import Case, Dataset
+from pydantic_evals.evaluators import LLMJudge
+
+from aurelian.agents.phenopackets.phenopackets_agent import phenopackets_agent
+from aurelian.agents.phenopackets.phenopackets_config import PhenopacketsDependencies
+
+class PhenopacketsMetadata(Dict[str, Any]):
+    """Simple metadata dictionary for Phenopackets evaluations."""
+    pass
+
+# Define individual evaluation cases
+case1 = Case(
+    name="liver_disease_patients",
+    inputs="What patients have liver disease?",
+    expected_output="hepat",  # Should mention hepatic/liver terms
+    metadata=metadata("medium", "phenotype_query")
+)
+
+case2 = Case(
+    name="metabolic_pathway_genes",
+    inputs="What phenopackets involve genes from metabolic pathways?",
+    expected_output="metabol",  # Should mention metabolic genes/pathways
+    metadata=metadata("hard", "gene_pathway_query"),
+    evaluators=[
+        LLMJudge(
+            rubric="""
+            Answer should:
+            1. Identify phenopackets containing genes involved in metabolic pathways
+            2. Link the metabolic genes to their corresponding phenotypes
+            3. Explain how these genes relate to metabolic pathways
+            4. Provide patient/case IDs where applicable
+            """,
+            include_input=True
+        )
+    ]
+)
+
+case3 = Case(
+    name="variant_effect_peroxisomal",
+    inputs="How does the type of variant affect phenotype in peroxisomal disorders?",
+    expected_output="peroxisom",  # Should discuss peroxisomal disorders
+    metadata=metadata("hard", "variant_phenotype_correlation")
+)
+
+case4 = Case(
+    name="skeletal_dysplasia_comparison",
+    inputs="Examine phenopackets for skeletal dysplasias and compare their phenotypes",
+    expected_output="skeletal",  # Should discuss skeletal terms
+    metadata=metadata("medium", "comparative_phenotype_analysis")
+)
+
+case5 = Case(
+    name="pnpla6_mutations",
+    inputs="Look up any patients with mutations in the PNPLA6 gene",
+    expected_output="PNPLA6",  # Should mention the PNPLA6 gene
+    metadata=metadata("easy", "gene_mutation_query")
+)
+
+def create_eval_dataset() -> Dataset[str, str, MetadataDict]:
+    """
+    Create a dataset for evaluating the Phenopackets agent.
+    
+    Returns:
+        Dataset of Phenopackets evaluation cases with appropriate evaluators
+    """
+    # Collect all cases
+    cases = [case1, case2, case3, case4, case5]
+    
+    # Dataset-level evaluators
+    evaluators = [
+        SubstringEvaluator(),
+        LLMJudge(
+            rubric="""
+            Evaluate the answer based on:
+            1. Accuracy in identifying relevant phenopackets based on the query
+            2. Correct interpretation of phenotype-genotype relationships
+            3. Proper use of HPO terms and gene identifiers
+            4. Comprehensive analysis of phenotypic data when requested
+            5. Clear presentation of results including patient/case identifiers when appropriate
+            """,
+            model="anthropic:claude-3-7-sonnet-latest"
+        )
+    ]
+    
+    return Dataset(
+        cases=cases,
+        evaluators=evaluators
+    )

aurelian/agents/phenopackets/phenopackets_gradio.py

@@ -0,0 +1,55 @@
+"""
+Gradio UI for the phenopackets agent.
+"""
+from typing import List, Optional
+
+import gradio as gr
+
+from aurelian.agents.phenopackets.phenopackets_agent import phenopackets_agent
+from aurelian.agents.phenopackets.phenopackets_config import PhenopacketsDependencies
+from aurelian.utils.async_utils import run_sync
+
+
+def chat(deps: Optional[PhenopacketsDependencies] = None, db_path: Optional[str] = None, collection_name: Optional[str] = None, **kwargs):
+    """
+    Initialize a chat interface for the phenopackets agent.
+    
+    Args:
+        deps: Optional dependencies configuration
+        db_path: Optional database path, defaults to MongoDB localhost
+        collection_name: Optional collection name, defaults to "main"
+        **kwargs: Additional arguments to pass to the agent
+        
+    Returns:
+        A Gradio chat interface
+    """
+    if deps is None:
+        deps = PhenopacketsDependencies()
+        
+    if db_path:
+        deps.db_path = db_path
+    if collection_name:
+        deps.collection_name = collection_name
+
+    def get_info(query: str, history: List[str]) -> str:
+        print(f"QUERY: {query}")
+        print(f"HISTORY: {history}")
+        if history:
+            query += "## History"
+            for h in history:
+                query += f"\n{h}"
+        result = run_sync(lambda: phenopackets_agent.run_sync(query, deps=deps, **kwargs))
+        return result.data
+
+    return gr.ChatInterface(
+        fn=get_info,
+        type="messages",
+        title="Phenopackets AI Assistant",
+        examples=[
+            ["What patients have liver disease?"],
+            ["What phenopackets involve genes from metabolic pathways?"],
+            ["How does the type of variant affect phenotype in peroxisomal disorders?"],
+            ["Examine phenopackets for skeletal dysplasias and compare their phenotypes"],
+            ["Look up any patients with mutations in the PNPLA6 gene"]
+        ]
+    )

aurelian/agents/phenopackets/phenopackets_mcp.py

@@ -0,0 +1,178 @@
+"""
+MCP tools for working with phenopacket databases.
+"""
+import os
+from typing import Dict, List
+
+from mcp.server.fastmcp import FastMCP
+
+import aurelian.agents.filesystem.filesystem_tools as fst
+from aurelian.agents.phenopackets.phenopackets_agent import SYSTEM
+import aurelian.agents.phenopackets.phenopackets_tools as pt
+from aurelian.agents.phenopackets.phenopackets_config import PhenopacketsDependencies
+from pydantic_ai import RunContext
+
+# Initialize FastMCP server
+mcp = FastMCP("phenopackets", instructions=SYSTEM)
+
+
+from aurelian.dependencies.workdir import WorkDir
+
+def deps() -> PhenopacketsDependencies:
+    deps = PhenopacketsDependencies()
+    # Set the location from environment variable or default
+    loc = os.getenv("AURELIAN_WORKDIR", "/tmp/aurelian")
+    deps.workdir = WorkDir(loc)
+    
+    # Get database connection parameters from environment if available
+    db_path = os.getenv("PHENOPACKETS_DB_PATH")
+    db_name = os.getenv("PHENOPACKETS_DB_NAME") 
+    collection_name = os.getenv("PHENOPACKETS_COLLECTION_NAME")
+    
+    if db_path:
+        deps.db_path = db_path
+    if db_name:
+        deps.db_name = db_name
+    if collection_name:
+        deps.collection_name = collection_name
+        
+    return deps
+
+def ctx() -> RunContext[PhenopacketsDependencies]:
+    rc: RunContext[PhenopacketsDependencies] = RunContext[PhenopacketsDependencies](
+        deps=deps(),
+        model=None, usage=None, prompt=None,
+    )
+    return rc
+
+
+@mcp.tool()
+async def search_phenopackets(query: str) -> List[Dict]:
+    """
+    Performs a retrieval search over the Phenopackets database.
+
+    The query can be any text, such as name of a disease, phenotype, gene, etc.
+
+    The objects returned are "Phenopackets" which is a structured representation
+    of a patient. Each is uniquely identified by a phenopacket ID (essentially
+    the patient ID).
+
+    The objects returned are summaries of Phenopackets; some details such
+    as phenotypes are omitted. Use `lookup_phenopacket` to retrieve full details.
+
+    Args:
+        query: The search query text
+
+    Returns:
+        List[Dict]: List of phenopackets matching the query
+    """
+    return await pt.search_phenopackets(ctx(), query)
+
+
+@mcp.tool()
+async def lookup_phenopacket(phenopacket_id: str) -> Dict:
+    """
+    Performs a lookup of an individual Phenopacket by its ID.
+
+    IDs are typically of the form PMID_nnn_PatientNumber, but this should not be assumed.
+
+    Args:
+        phenopacket_id: The ID of the Phenopacket to look up
+
+    Returns:
+        Dict: The phenopacket data
+    """
+    return await pt.lookup_phenopacket(ctx(), phenopacket_id)
+
+
+@mcp.tool()
+async def lookup_pmid(pmid: str) -> str:
+    """
+    Lookup the text of a PubMed article by its PMID.
+    
+    A PMID should be of the form "PMID:nnnnnnn" (no underscores).
+    
+    NOTE: Phenopacket IDs are typically of the form PMID_nnn_PatientNumber,
+    but this should not be assumed. To reliably get PMIDs for a phenopacket,
+    use `lookup_phenopacket` to retrieve and examine the `externalReferences` field.
+    
+    Args:
+        pmid: The PubMed ID to look up
+        
+    Returns:
+        str: Full text if available, otherwise abstract
+    """
+    return await pt.lookup_pmid(pmid)
+
+
+@mcp.tool()
+async def search_web(query: str) -> str:
+    """
+    Search the web using a text query.
+    
+    Args:
+        query: The search query
+        
+    Returns:
+        str: Search results with summaries
+    """
+    return await pt.search_web(query)
+
+
+@mcp.tool()
+async def retrieve_web_page(url: str) -> str:
+    """
+    Fetch the contents of a web page.
+    
+    Args:
+        url: The URL to fetch
+        
+    Returns:
+        str: The contents of the web page
+    """
+    return await pt.retrieve_web_page(url)
+
+
+@mcp.tool()
+async def inspect_file(data_file: str) -> str:
+    """
+    Inspect a file in the working directory.
+
+    Args:
+        data_file: name of file
+
+    Returns:
+        str: Contents of the file
+    """
+    return await fst.inspect_file(ctx(), data_file)
+
+
+@mcp.tool()
+async def list_files() -> str:
+    """
+    List files in the working directory.
+
+    Returns:
+        str: List of files in the working directory
+    """
+    return await fst.list_files(ctx())
+
+
+@mcp.tool()
+async def write_to_file(file_name: str, data: str) -> str:
+    """
+    Write data to a file in the working directory.
+
+    Args:
+        file_name: Name of the file to write
+        data: Data to write to the file
+
+    Returns:
+        str: Confirmation message
+    """
+    return await fst.write_to_file(ctx(), file_name, data)
+
+
+if __name__ == "__main__":
+    # Initialize and run the server
+    mcp.run(transport='stdio')

aurelian/agents/phenopackets/phenopackets_tools.py

@@ -0,0 +1,127 @@
+"""
+Tools for the phenopackets agent.
+"""
+from typing import List, Dict, Optional
+
+from pydantic_ai import RunContext, ModelRetry
+
+from aurelian.agents.phenopackets.phenopackets_config import PhenopacketsDependencies
+from aurelian.utils.data_utils import flatten
+from aurelian.agents.literature.literature_tools import (
+    lookup_pmid as literature_lookup_pmid,
+    search_literature_web,
+    retrieve_literature_page
+)
+
+
+async def search_phenopackets(ctx: RunContext[PhenopacketsDependencies], query: str) -> List[Dict]:
+    """
+    Performs a retrieval search over the Phenopackets database.
+
+    The query can be any text, such as name of a disease, phenotype, gene, etc.
+
+    The objects returned are "Phenopackets" which is a structured representation
+    of a patient. Each is uniquely identified by a phenopacket ID (essentially
+    the patient ID).
+
+    The objects returned are summaries of Phenopackets; some details such
+    as phenotypes are omitted. Use `lookup_phenopacket` to retrieve full details.
+
+    Args:
+        ctx: The run context
+        query: The search query text
+
+    Returns:
+        List[Dict]: List of phenopackets matching the query
+    """
+    print(f"SEARCH PHENOPACKETS: {query} // {ctx.deps}")
+    try:
+        qr = ctx.deps.collection.search(query, index_name="llm", limit=ctx.deps.max_results)
+        objs = []
+        for score, row in qr.ranked_rows:
+            obj = flatten(row, preserve_keys=["interpretations", "diseases"])
+            obj["relevancy_score"] = score
+            objs.append(obj)
+            print(f"RESULT: {obj}")
+        
+        if not objs:
+            raise ModelRetry(f"No phenopackets found matching the query: {query}. Try a different search term.")
+        
+        return objs
+    except Exception as e:
+        if "ModelRetry" in str(type(e)):
+            raise e
+        raise ModelRetry(f"Error searching phenopackets: {str(e)}")
+
+
+async def lookup_phenopacket(ctx: RunContext[PhenopacketsDependencies], phenopacket_id: str) -> Dict:
+    """
+    Performs a lookup of an individual Phenopacket by its ID.
+
+    IDs are typically of the form PMID_nnn_PatientNumber, but this should not be assumed.
+
+    Args:
+        ctx: The run context
+        phenopacket_id: The ID of the Phenopacket to look up
+
+    Returns:
+        Dict: The phenopacket data
+    """
+    print(f"LOOKUP PHENOPACKET: {phenopacket_id}")
+    try:
+        qr = ctx.deps.collection.find({"id": phenopacket_id})
+        if not qr.rows:
+            raise ModelRetry(f"Could not find phenopacket with ID {phenopacket_id}. The ID may be incorrect.")
+        return qr.rows[0]
+    except Exception as e:
+        if "ModelRetry" in str(type(e)):
+            raise e
+        raise ModelRetry(f"Error looking up phenopacket {phenopacket_id}: {str(e)}")
+
+
+async def lookup_pmid(pmid: str) -> str:
+    """
+    Lookup the text of a PubMed article by its PMID.
+    
+    A PMID should be of the form "PMID:nnnnnnn" (no underscores).
+    
+    NOTE: Phenopacket IDs are typically of the form PMID_nnn_PatientNumber,
+    but this should not be assumed. To reliably get PMIDs for a phenopacket,
+    use `lookup_phenopacket` to retrieve and examine the `externalReferences` field.
+    
+    Args:
+        pmid: The PubMed ID to look up
+        
+    Returns:
+        str: Full text if available, otherwise abstract
+    """
+    print(f"LOOKUP PMID FOR PHENOPACKET: {pmid}")
+    return await literature_lookup_pmid(pmid)
+
+
+async def search_web(query: str) -> str:
+    """
+    Search the web using a text query.
+    
+    Args:
+        query: The search query
+        
+    Returns:
+        str: Search results with summaries
+    """
+    print(f"PHENOPACKET WEB SEARCH: {query}")
+    return await search_literature_web(query)
+
+
+async def retrieve_web_page(url: str) -> str:
+    """
+    Fetch the contents of a web page.
+    
+    Args:
+        url: The URL to fetch
+        
+    Returns:
+        str: The contents of the web page
+    """
+    print(f"FETCH WEB PAGE FOR PHENOPACKET: {url}")
+    return await retrieve_literature_page(url)

aurelian/agents/rag/__init__.py

@@ -0,0 +1,40 @@
+"""
+RAG agent package for retrieval-augmented generation against document collections.
+"""
+
+# Constants
+COLLECTION_NAME = "main"
+
+# isort: skip_file
+from .rag_agent import rag_agent  # noqa: E402
+from .rag_config import RagDependencies, get_config  # noqa: E402
+from .rag_gradio import chat  # noqa: E402
+from .rag_tools import (  # noqa: E402
+    search_documents,
+    inspect_document,
+    lookup_pmid,
+    search_web,
+    retrieve_web_page,
+)
+
+__all__ = [
+    # Constants
+    "COLLECTION_NAME",
+    
+    # Agent
+    "rag_agent",
+    
+    # Config
+    "RagDependencies",
+    "get_config",
+    
+    # Tools
+    "search_documents",
+    "inspect_document",
+    "lookup_pmid",
+    "search_web",
+    "retrieve_web_page",
+    
+    # Gradio
+    "chat",
+]

aurelian/agents/rag/rag_agent.py

@@ -0,0 +1,84 @@
+"""
+Agent for retrieval-augmented generation (RAG) against document collections.
+"""
+from pydantic_ai import Agent, RunContext
+
+from .rag_config import RagDependencies
+from .rag_tools import search_documents, inspect_document, lookup_pmid, search_web, retrieve_web_page
+
+
+rag_agent = Agent(
+    model="openai:gpt-4o",
+    deps_type=RagDependencies,
+    result_type=str,
+    system_prompt=(
+        "You are an AI assistant that help explore a literature collection via RAG."
+        " You can use different functions to access the store, for example:"
+        "  - `search_documents` to find documents by text query"
+        "  - `inspect_document` to retrieve a specific document (by title/name)"
+        "You can also use `lookup_pmid` to retrieve the text of a PubMed ID, or `search_web` to search the web."
+    ),
+    defer_model_check=True,
+)
+
+
+@rag_agent.tool
+async def search_documents_tool(ctx: RunContext[RagDependencies], query: str):
+    """
+    Performs a retrieval search over the RAG database.
+    
+    The query can be any text, such as name of a disease, phenotype, gene, etc.
+    """
+    return await search_documents(ctx, query)
+
+
+@rag_agent.tool
+async def inspect_document_tool(ctx: RunContext[RagDependencies], query: str):
+    """
+    Returns the content of the document.
+    
+    Args:
+        query: E.g. title
+    """
+    return await inspect_document(ctx, query)
+
+
+@rag_agent.tool
+async def lookup_pmid_tool(ctx: RunContext[RagDependencies], pmid: str):
+    """
+    Lookup the text of a PubMed ID, using its PMID.
+    
+    A PMID should be of the form "PMID:nnnnnnn" (no underscores).
+    
+    NOTE: Phenopacket IDs are typically of the form PMID_nnn_PatientNumber,
+    but this should be be assumed. To reliably get PMIDs for a phenopacket,
+    use `lookup_phenopacket` to retrieve examine the `externalReferences`
+    field.
+    
+    Returns: full text if available, otherwise abstract
+    """
+    return await lookup_pmid(ctx, pmid)
+
+
+@rag_agent.tool
+async def search_web_tool(ctx: RunContext[RagDependencies], query: str):
+    """
+    Search the web using a text query.
+    
+    Note, this will not retrieve the full content, for that you
+    should use `retrieve_web_page`.
+    
+    Returns: matching web pages plus summaries
+    """
+    return await search_web(ctx, query)
+
+
+@rag_agent.tool
+async def retrieve_web_page_tool(ctx: RunContext[RagDependencies], url: str):
+    """
+    Fetch the contents of a web page.
+    
+    Returns:
+        The contents of the web page.
+    """
+    return await retrieve_web_page(ctx, url)

aurelian/agents/rag/rag_config.py

@@ -0,0 +1,80 @@
+"""
+Configuration for the RAG agent.
+"""
+from dataclasses import dataclass, field
+import os
+from typing import Optional
+
+from linkml_store import Client
+from linkml_store.api import Collection
+
+from aurelian.dependencies.workdir import HasWorkdir, WorkDir
+from . import COLLECTION_NAME
+
+
+@dataclass
+class RagDependencies:
+    """Configuration for the RAG agent."""
+    
+    # Required fields
+    db_path: str
+    
+    # Optional fields with defaults
+    collection_name: str = COLLECTION_NAME
+    max_results: int = 10
+    max_content_len: int = 5000
+    workdir: Optional[WorkDir] = None
+    _collection: Optional[Collection] = None
+    
+    def __post_init__(self):
+        """Initialize the config with default values."""
+        if self.workdir is None:
+            self.workdir = WorkDir()
+
+    @property
+    def collection(self) -> Collection:
+        """Get the database collection, initializing it if needed."""
+        if self._collection is None:
+            client = Client()
+            db_path = self.db_path
+            client.attach_database(db_path)
+            db = client.databases[db_path]
+            self._collection = db.get_collection(self.collection_name)
+        return self._collection
+
+
+def get_config(db_path: Optional[str] = None, collection_name: Optional[str] = None) -> RagDependencies:
+    """
+    Get the RAG configuration from environment variables or defaults.
+    
+    Args:
+        db_path: The database path to use (overrides environment variable)
+        collection_name: The collection name to use (overrides environment variable)
+        
+    Returns:
+        A RagDependencies instance
+    """
+    # Try to get from environment, then use provided values or defaults
+    env_db_path = os.environ.get("AURELIAN_RAG_DB_PATH", None)
+    env_collection = os.environ.get("AURELIAN_RAG_COLLECTION", COLLECTION_NAME)
+    
+    # Use provided values first, then environment, then defaults
+    final_db_path = db_path or env_db_path
+    final_collection = collection_name or env_collection
+    
+    # For testing purposes, if no DB path is provided, use a default one
+    # This is only used for running basic smoke tests
+    if not final_db_path:
+        if os.environ.get("TESTING", "0") == "1":
+            final_db_path = "memory://test"
+        else:
+            raise ValueError("Database path must be provided either as parameter or via AURELIAN_RAG_DB_PATH environment variable")
+    
+    workdir_path = os.environ.get("AURELIAN_WORKDIR", None)
+    workdir = WorkDir(location=workdir_path) if workdir_path else None
+    
+    return RagDependencies(
+        db_path=final_db_path,
+        collection_name=final_collection,
+        workdir=workdir,
+    )