Development of a Prognostic Compound Immunoscore for Head and Neck Cancer

项目摘要 来自最近临床试验的知识表明，超过80%的头颈癌（HNC） 低免疫原性冷肿瘤和对免疫检查点受体（ICR）阻断无应答。与 由于新出现的针对冷癌的组合策略，精确识别这组肿瘤对于 选择最佳的治疗方案。然而，没有一致的算法可用于评估 HNC的整体免疫特征。目前大多数免疫评分方法是基于免疫组化 (IHC)有限的一组生物标志物的染色，这阻止了对肿瘤景观的精确注释， 浸润淋巴细胞（TIL）。IHC方法具有技术敏感性，可能存在机构间和 病理学家之间的差异。此外，目前的免疫评分仅侧重于少数T细胞亚群， 不整合调节肿瘤对免疫杀伤反应的癌症基因组特征。事实上， 有证据表明，I型干扰素（IFN-I）途径在HNC应答中起重要作用， 效应免疫细胞因此，利用全球TIL谱和癌症基因组特征提供了一种新的治疗方法。 这是基于HNC的免疫原性对其进行分类的前所未有的机会。目前的稳健方法 细胞去卷积对离群值敏感，这在整个肿瘤RNA-Seq中经常观察到 数据集。我们最近的研究表明，一种新的机器学习工具，快速和鲁棒的DEconcolution 自适应检测和去除离群值的表达谱（FARDEEP）在以下方面表现出优越的上级准确性： 免疫细胞去卷积与FOA完全一致，该项目的首要假设是， 整合FARDEEP辅助的TIL去卷积和癌症基因组学的复合免疫评分可以 有效识别冷HNC。为达致这个目标，我们紧接着要采取的两个步骤是：**（1）我们会制订 强大的无模型方法来识别TLR驱动的致癌途径; **（2）我们将构建一种化合物 通过整合癌症基因组特征和TIL谱的免疫评分来鉴定冷HNC。这些研究将 开发一种新的“适用于分析全基因组数据的统计方法”，并提供“统计数据”。 对现有的全基因组数据进行分析”。该项目将完善一个强大的， 新的免疫细胞去卷积机器学习工具，并表征转移的中枢致癌通路 TL景观。新的免疫基因组学算法将简化免疫评分方法， 有效地分层HNC，并有助于组合治疗的精度选择。

PROJECT SUMMARY Knowledge from the recent clinical trials suggests that over 80% of head and neck cancer (HNC) are hypo-immunogenic cold tumors and non-responsive to immune checkpoint receptors (ICR) blockade. With the emerging combinatorial strategies for cold cancer, precise identification of this group of tumors is essential for the selection of optimal treatment protocols. However, there is no consistent algorithm available to assess the global immune profile of HNC. Most of the current immunoscore methods are based on immunohistochemical (IHC) staining of a limited panel of biomarkers, which prevents a precise annotation of the landscape of tumor- infiltrating lymphocytes (TIL). The IHC method is technically sensitive, and may present inter-institutional and inter-pathologists variations. Moreover, the current immunoscore only emphasizes on a few T-cell subsets, and does not integrate cancer genomic features that modulate tumor response to immune killing. In fact, strong evidence suggests that the type I interferon (IFN-I) pathway plays a fundamental role in HNC response to effector immune cells. Thus, leveraging global TIL profiles and cancer genomic features offers an unprecedented opportunity to classify HNC based on its immunogenicity. The current robust methods for cellular deconvolution are sensitive to outliers, which are frequently observed in the whole tumor RNA-Seq datasets. Our recent studies show that a novel machine learning tool Fast And Robust DEconcolution of Expression Profiles (FARDEEP), which adaptively detects and removes outliers, exhibits superior accuracy in immune cell deconvolution. In precise alignment with the FOA, the overarching hypothesis of this project is that a compound immunoscore integrating FARDEEP-assisted TIL deconvolution and cancer genomics can effectively identify cold HNC. To achieve this goal, our two immediate next steps are: **(1) We will develop a robust model-free approach to identify TIL-driving oncogenic pathways; **(2) We will construct a compound immunoscore integrating cancer genomic features and TIL profiles to identify cold HNC. These studies will develop a novel “statistical methodology appropriate for analyzing genome-wide data” and provide “statistical analysis of existing genome-wide data” for an NIDCR priority disease. This project will refine a robust and novel immune-cell deconvolution machine learning tool and characterize central oncogenic pathways that shift the TIL landscape. The new immunogenomics algorithms will streamline the immunoscoring method to effectively stratify HNC and contribute to the precision selection of combinatorial treatments.