Skip to content

Configuration Guide 🛠️

This guide details all configuration options available in 🧪 MetaboT 🍵, helping you customize the system to your specific needs.

Configuration Overview 📁

🧪 MetaboT 🍵 uses several configuration files located in the app/config/ directory:

  • params.ini: Language model configurations
  • sparql.ini: SPARQL query templates and settings
  • logging.ini: Logging configuration
  • .env: Environment variables (created by user)

LLMs Configuration 🤖

Located in params.ini, this configuration controls the behavior of different language models used in the system.

[llm_preview]
id = gpt-4-0125-preview
temperature = 0
max_retries = 3

[llm_o]
id = gpt-4o
temperature = 0
max_retries = 3


[llm_mini]
id = gpt-4o-mini
temperature = 0
max_retries = 3

[llm_o3_mini]
id = o3-mini-2025-01-31
temperature = 1
max_retries = 3

[llm_o1]
id = o1-2024-12-17
temperature = 1
max_retries = 3

[deepseek_deepseek-chat]
id = deepseek-chat
temperature = 0
max_retries = 3
base_url = https://api.deepseek.com

[deepseek_deepseek-reasoner]
id = deepseek-reasoner
temperature = 0
max_retries = 3
base_url = https://api.deepseek.com

[ovh_Meta-Llama-3_1-70B-Instruct]
id = Meta-Llama-3_1-70B-Instruct
temperature = 0
max_retries = 3
base_url = https://llama-3-1-70b-instruct.endpoints.kepler.ai.cloud.ovh.net/api/openai_compat/v1

Available Sections

  • Main LLM ([llm_o]) Production-optimized GPT-4o 2024.08.06 used by most agents.
  • Preview Model ([llm_preview]) Latest model versions with cutting-edge capabilities.
  • Mini Model ([llm_mini]) Lightweight GPT-4o variant for basic tasks.
  • O3 Mini ([llm_o3_mini]) Specialized model for experimental tasks with higher creativity.

  • O1 Core ([llm_o1]) Advanced model for research and development.

  • DeepSeek Chat ([deepseek_deepseek-chat]) Conversational model from DeepSeek for interactive queries.

  • DeepSeek Reasoner ([deepseek_deepseek-reasoner]) Analytical model from DeepSeek for enhanced reasoning.
  • OVH Meta-Llama ([ovh_Meta-Llama-3_1-70B-Instruct]) Instructive model providing robust language understanding.

🧪 MetaboT 🍵 supports both OpenAI-compatible API endpoints and various other LLM providers through LiteLLM integration. Models can be configured in two ways:

  1. OpenAI-compatible endpoints using sections like [llm_o], [deepseek_deepseek-chat], etc.
  2. Other providers through LiteLLM using sections starting with [llm_litellm_], for example: 
    [llm_litellm_openai]
id = gpt-4
temperature = 0

[llm_litellm_deepseek]
id = deepseek/deepseek-chat

[llm_litellm_claude]
id = claude-3-opus-20240229

[llm_litellm_gemini]
id = gemini/gemini-1.5-pro

For a complete list of supported providers and their model identifiers, see the LiteLLM Providers Documentation.

You can configure different models for each agent in your workflow. For detailed instructions on agent-specific model configuration, refer to the Language Model Configuration Guide.

Parameters

  • id: Model identifier (format depends on provider)
  • For OpenAI-compatible: e.g., gpt-4o, gpt-3.5-turbo
  • For LiteLLM: provider/model-name (e.g., deepseek/deepseek-chat)
  • temperature: Randomness in responses (0-1)
  • max_retries: Number of retry attempts (optional)
  • base_url: Custom API endpoint URL (optional)

SPARQL Configuration 🔍

The sparql.ini file contains SPARQL query templates and settings essential for interacting with the knowledge graph. These configurations are used by the RdfGraph class to dynamically retrieve the schema from the knowledge graph when no local schema file is provided.

Query Templates

[sparqlQueries]
# Class information query
CLS_RDF = SELECT DISTINCT ?cls ?com ?label
        WHERE {
            ?cls a rdfs:Class .
            OPTIONAL { ?cls rdfs:comment ?com }
            OPTIONAL { ?cls rdfs:label ?label }
        }
        GROUP BY ?cls ?com ?label
This query retrieves all classes along with their optional comments and labels. The results form the foundation for constructing the dynamic schema.

# Class relationships query
CLS_REL_RDF = SELECT ?property (SAMPLE(COALESCE(?type, STR(DATATYPE(?value)), "Untyped")) AS ?valueType) 
        WHERE {...}
This query is executed for each class retrieved by CLS_RDF. It retrieves properties associated with instances of the specified class and determines a representative value type for each property.

Excluded URIs

[excludedURIs]
uris = http://www.w3.org/1999/02/22-rdf-syntax-ns#type,
       http://www.w3.org/2000/01/rdf-schema#comment,
       http://www.w3.org/2000/01/rdf-schema#Class,
       http://xmlns.com/foaf/0.1/depiction
The list of excluded URIs defines properties that are filtered out during the schema retrieval process.

Logging Configuration 📝

The logging.ini file controls the logging behavior of 🧪 MetaboT 🍵.

Logger Settings

[loggers]
keys=root

[logger_root]
level=INFO
handlers=consoleHandler,fileHandler

Handler Configuration

  • Console Handler
    [handler_consoleHandler]
class=StreamHandler
level=INFO
formatter=simpleFormatter
args=(sys.stdout,)
  • File Handler
    [handler_fileHandler]
class=logging.handlers.RotatingFileHandler
level=INFO
formatter=detailedFormatter
args=('./app/config/logs/app.log', 'w', 100000, 5)

Formatter Settings

  • Simple Formatter (Console)
    [formatter_simpleFormatter]
format=%(name)s - %(levelname)s - %(message)s
  • Detailed Formatter (File)
    [formatter_detailedFormatter]
format=%(asctime)s - %(name)s - %(levelname)s - %(message)s
datefmt=%Y-%m-%d %H:%M:%S

Environment Variables 🔧

Create a .env file in the project root with these variables:

# LLM API Configuration
OPENAI_API_KEY=your_openai_api_key          # If using OpenAI models
DEEPSEEK_API_KEY=your_deepseek_api_key      # If using DeepSeek models
CLAUDE_API_KEY=your_claude_api_key          # If using Claude models
GEMINI_API_KEY=your_gemini_api_key          # If using Gemini models
OVHCLOUD_API_KEY=your_ovh_api_key           # If using Llama on OVHcloud

# Knowledge Graph Configuration
KG_ENDPOINT_URL=https://enpkg.commons-lab.org/graphdb/repositories/ENPKG
SPARQL_USERNAME=your_username               # If endpoint requires authentication
SPARQL_PASSWORD=your_password               # If endpoint requires authentication

# LangSmith Configuration (Optional)
LANGCHAIN_API_KEY=your_langsmith_api_key
LANGCHAIN_PROJECT=MetaboT
LANGCHAIN_ENDPOINT=https://api.smith.langchain.com

Best Practices 📘

Language Model Selection

  • Use llm_o for complex queries requiring high accuracy.
  • Use llm_o3_mini for faster, cost-effective operations.
  • Consider llm_o3 or llm_o1and its variants for complex questions.

Logging Configuration

  • Keep the default INFO level for production.
  • Use the DEBUG level during development.
  • Monitor log file sizes (default 1MB per file, 5 files max).

SPARQL Optimization

Environment Security

  • Never commit the .env file to version control.
  • Rotate API keys regularly.
  • Use different keys for development and production.

Troubleshooting 🚨

Common Issues

  • Language Model Errors

    • Check API key validity.
    • Verify model availability.
    • Review rate limits.

  • Logging Issues

    • Ensure write permissions for the log directory.
    • Check disk space.
    • Verify log rotation settings.

  • SPARQL Problems

    • Validate endpoint accessibility.
    • Check query syntax.
    • Review timeout settings.

Default Dataset and Data Conversion 📊

Note: By default, 🧪 MetaboT 🍵 connects to the public ENPKG endpoint which hosts an open, annotated mass spectrometry dataset derived from a chemodiverse collection of 1,600 plant extracts. This default dataset enables you to explore all features of 🧪 MetaboT 🍵 without the need for custom data conversion immediately. To use 🧪 MetaboT 🍵 on your mass spectrometry data, the processed and annotated results must first be converted into a knowledge graph format using the ENPKG tool.

  • Update the KG_ENDPOINT_URL in your .env file to point to your custom knowledge graph endpoint.