Add DISGENET API integration as a new database tool

[MoiraClimentGispert]

Summary

This PR by MedBioinformatics adds DISGENET as a new data source in Biomni, following the Adding New Data (web API) contribution guidelines.

DISGENET™(https://disgenet.com/) is an integrated, evidence-scored knowledge layer that connects genes, variants, diseases, traits and therapeutics into a unified semantic framework designed for computational use, clinical interpretation and translational research.

What's included

Core tool (new data source via web API):

• biomni/tool/database.py — New query_disgenet_api() function that translates natural-language prompts into DISGENET REST API calls. Includes automatic entity normalization (disease names → UMLS CUI, gene names → NCBI Gene ID), dynamic endpoint selection, and structured result parsing.
• biomni/tool/literature.py — New query_disgenet_evidence() evidence retrieval tool for DISGENET literature-backed association queries.
• biomni/tool/tool_description/database.py — Tool description for query_disgenet_api following existing format.
• biomni/tool/tool_description/literature.py — Tool description for query_disgenet_evidence() following the existing format.

Agent integration:

• biomni/agent/a1.py — Added DISGENET_API_KEY preflight check during agent initialization (warns early if key is missing; optionally prompts in interactive sessions). Moved load_dotenv call earlier to ensure env vars are available before imports.
• biomni/config.py — Added DISGENET-related configuration options.
• .env.example — Added DISGENET_API_KEY entry.

Key features of the DISGENET tool

• Natural language → API endpoint translation via LLM
• Automatic entity normalization (disease names, gene symbols → standard identifiers)
• Support for ordering/filtering by DISGENET score, DSI, DPI, EI, pLI, pmYear
• GDA, VDA, and DDA association queries
• Disease class queries
• Verbose mode for debugging (shows normalization steps, resolved endpoint, full metadata)

Test prompt

from biomni.agent import A1

agent = A1(path='./data', llm='claude-sonnet-4-20250514')
agent.go("What are the top genes associated with Parkinson's disease ordered by DISGENET score?")
agent.go("Show variant-disease associations for BRCA1 filtered by evidence index > 0.8")
agent.go("Find diseases associated with TP53 in the Musculoskeletal Diseases class")

Requirements if you want to use DISGENET tool

• A DISGENET API key (set as DISGENET_API_KEY environment variable)
• No additional pip dependencies beyond what Biomni already requires

Test plan

• Tested query_disgenet_api() with GDA, VDA, and DDA queries
• Verified entity normalization (disease names, gene symbols)
• Confirmed agent selects DISGENET tool appropriately via natural language
• Ran benchmark evaluation (results included in biomni/eval/benchmark_results/)
• Verified no conflicts with existing tools or upstream changes