{
  "id": 18107847,
  "title": "SPaRTA: Spatial Reasoning in Text-to-SQL for Omics Analysis",
  "url": "https://urgent.news/2026/11/04/sparta-spatial-reasoning-in-text-to-sql-for-omics-analysis",
  "topic": "science",
  "section": "Science",
  "published": "2026-11-04T00:00:00.000Z",
  "source": {
    "name": "bioRxiv",
    "slug": "biorxiv",
    "url": "https://www.biorxiv.org/content/10.64898/2026.10.31.763119v1?rss=1"
  },
  "original_language": "en",
  "account": "Spatial-omics technology offers high-resolution maps of transcriptomic, proteomic, and metabolomic activity within tissues. However, these complex datasets can be difficult to query without specialized computational knowledge, limiting accessibility to those with programming experience. To address this issue, researchers have developed SPaRTA (Spatial Reasoning in Text-to-SQL for Omics Analysis), a natural language interface that converts free-text questions into executable spatial SQL queries for such data.\n\nThe team has released SPaRTA-Bench, a benchmark dataset containing database schemas specific to each dataset, explicit query conventions, and 7,500 question-SQL pairs. These pairs have been validated through LLM-assisted verification and targeted human review, covering breast cancer, pancreatic cancer, and melanoma tissue sections. The benchmark identifies 15 failure modes where a query may run but return an incorrect answer.\n\nWhen using open-weight large language models (LLMs) of various families, such as Llama, Mistral, Qwen, Gemma, and Phi, zero-shot execution accuracy ranges from 10-40%. With reasoning and execution feedback, accuracy improves to at most 61%. Fine-tuning further increases accuracy to 63-77%, but the models still rarely achieve perfect accuracy on unseen tissue data. However, execution feedback can help repair queries that fail to run, though it rarely resolves issues where queries run correctly but return incorrect answers.",
  "summary": "Spatial-omics technologies provide clinicians, pathologists, and biomedical researchers with high-resolution maps of transcriptomic, proteomic, and metabolomic activity within tissues. However, querying these complex datasets often requires specialized computational expertise, limiting their accessibility to domain experts without programming experience. In the current manuscript, we develop…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}