{"id":507,"title":"Visual LLM-guided consensus spatial domain detection with L-STAR","title_zh":"L-STAR：用视觉大语言模型比较空间域划分结果并构建共识图谱","url":"https://doi.org/10.1038/s41467-026-78923-w","doi":"10.1038/s41467-026-78923-w","pmid":null,"journal":"Nature Communications","kind":"paper","lang":"en","published_at":"2026-11-03T00:00:00+00:00","fetched_at":"2026-11-08T03:45:21+00:00","authors":["Changyue Zhao","Zhicheng Ji"],"primary_topic":"ethics-policy","topics":[{"slug":"ai-compute","score":4.0,"hits":["large language model","llm"]},{"slug":"ethics-policy","score":4.0,"hits":["consensus"]},{"slug":"omics-methods","score":1.0,"hits":["spatial transcriptomics"]}],"topic_score":4.0,"preference_score":0.0,"preference_labels":[],"llm_score":72.0,"final_score":80.0,"score_dims":{"novelty":86,"rigor":75,"impact":68,"relevance":42,"translation":58},"recommend_reason":"将视觉大语言模型引入空间转录组域划分的共识构建，方法思路新、可解释性有亮点，对使用空间转录组的生殖组织研究（子宫内膜、卵巢、胚胎）有方法学借鉴价值；但非生殖核心问题，属于工具类论文。","keywords":["空间转录组","空间域划分","大语言模型","共识分析","可解释性"],"selected":1,"story_id":115,"llm_provider":"deepseek","llm_model":"deepseek-flash","llm_effort":"low","source_key":"crossref-cns","source_name":"Crossref · CNS 与大子刊新上线（领域过滤）","source_tier":1,"mention_count":1,"has_interpretation":false,"journal_key":"nat-commun","journal_tier":"cns-sub","journal_if":14.7,"is_mock":false,"summary_zh":"该研究针对空间转录组分析中的关键步骤——组织空间域划分，指出不同算法在不同数据集上表现差异大、缺乏普遍最优方法。作者提出 L-STAR，利用大语言模型的视觉能力对多种候选方法生成的空间域图进行比较与排序，并基于排名最高的结果构建共识空间域图谱。在涵盖不同组织、物种与技术平台的数据集中，L-STAR 的划分结果比单一方法或固定方法组合的共识分析更接近专家注释；其排序具有可重复性，且在图像配色与呈现顺序改变时总体稳定。L-STAR 还能给出比较依据，并为检出的空间域赋予有生物学意义的名称，使分析更自适应、稳健且可解释。","summary":"Abstract Spatial transcriptomics measures gene activity while preserving the locations of cells within tissue. A key analysis step is to divide the tissue into spatial domains, but the performance of computational methods varies considerably across datasets, and no single method consistently performs best. Here we show that a large language model can compare spatial domain maps and identify the better-performing methods. We developed L-STAR, which visually compares and ranks candidate methods, then builds a consensus map from the highest-ranked results. Across datasets representing different tissues, species, and technologies, L-STAR matched expert annotations more closely than individual methods or consensus analysis using a fixed set of methods. Its rankings were reproducible and generally stable when image colors or presentation order changed. L-STAR also provided rationales for its comparisons and assigned biologically meaningful names to the detected domains. These findings show that large language models can make spatial transcriptomics analysis more adaptive, robust, and interpretable.","interpretation_basis":null,"fulltext_status":null,"interpretation":null,"fulltext":{"available":false,"status":null,"reason":"尚未抓取全文（未入选、分数未达阈值，或还没跑 fulltext 步骤）","pmcid":null,"license":null,"chars":0,"source":null,"fetched_at":null}}