GeneVAR：面向自回归基因到 WSI 瓦片合成的因果 MeanFlow
GeneVAR: Causal MeanFlow for Autoregressive Gene-to-WSI Tile Synthesis
把 RNA-Seq 条件通过反事实干预约束成伪影不变的自回归粗到细生成过程，可为视觉中「条件信号弱、易被染色/对比度等非语义因素劫持」的可控生成提供可复用的因果正则思路；多尺度矢量量化离散形态表示也可迁移到其他病理或遥感瓦片生成。
原文依据 leverages counterfactual interventions to suppress non-biological variations (e.g., staining or contrast artifacts)
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论理解转录组程序如何塑造组织形态仍是计算病理学的核心挑战。基因到 WSI 瓦片合成提供了一个原则性的生成框架，可将分子谱转化为组织学图像。然而，多数现有方法将 RNA-Seq 压缩为单一全局嵌入并在初始化时一次性注入，这种过度简化的设计削弱了转录组信号并引入虚假的、非生物学的关联。我们提出 GeneVAR，一种自回归基因到 WSI 模型，将合成重构为迭代的、由粗到细的生成过程。其核心是新颖的因果 MeanFlow 模块，利用反事实干预抑制非生物学变异（如染色或对比度伪影）并强制实现伪影不变的生成过程。同时，它在多个阶段强化转录组引导，在整个生成轨迹中保持生物学保真度。结合用于紧凑基因嵌入的 b-VAE 和用于离散形态表示的多尺度矢量量化器，GeneVAR 生成的 H&E 染色 WSI 瓦片既视觉逼真又转录组忠实。在五个 TCGA 癌症基准上的大量实验表明其持续达到最优性能，在生成保真度和下游分类准确率上均超越先验方法。
英文原文摘要 1417 字符
Understanding how transcriptomic programs shape tissue morphology remains a central challenge in computational pathology. Gene-to-WSI tile synthesis offers a principled generative framework to translate molecular profiles into histological images. However, most existing methods compress RNA-Seq into a single global embedding injected once at initialization, an oversimplified design that weakens transcriptomic signals and induces spurious, non-biological associations. We present GeneVAR, an Autoregressive Gene-to-WSI model that reformulates synthesis as an iterative, coarse-to-fine generative process. At its core is a novel Causal MeanFlow module, which leverages counterfactual interventions to suppress non-biological variations (e.g., staining or contrast artifacts) and enforce an artifact-invariant generation process. Concurrently, it reinforces transcriptome-informed guidance across multiple stages, preserving biological fidelity throughout the generative trajectory. Combined with a b-VAE for compact gene embeddings and a multi-scale vector quantizer for discrete morphology representation, GeneVAR generates H&E-stained WSI tiles that are both visually realistic and transcriptomically faithful. Extensive experiments across five TCGA cancer benchmarks demonstrate consistent state-of-the-art performance, surpassing prior methods in both generative fidelity and downstream classification accuracy.