默赛生基因科技(重庆)有限公司 Mosaisheng Gene Technology (Chongqing) Co., Ltd.
以 DNA 甲基化谱为核心,攻坚中枢神经系统肿瘤的分类难题与算法模型解决方案 DNA methylation profiling at the core — algorithms, models and research collaboration for CNS tumor classification
向下浏览Scroll
科研实例与合作Research in action

经同行评议的甲基化分类研究Peer-reviewed methylation classification research

创始人以第一作者发表的代表性研究,与分类器的真实输出——可作为科研合作能力的直接佐证。Selected first-author studies and real classifier output — direct evidence of our research-collaboration capability.

PGNT t-SNE 聚类
乳头状胶质神经元肿瘤(PGNT)的特征性甲基化谱 t-SNE 聚类Characteristic methylation t-SNE clustering of papillary glioneuronal tumor (PGNT)
Hou Y, et al. Acta Neuropathologica, 2019 · 第一作者first author
CIC 融合肿瘤 t-SNE 聚类
CIC 融合中枢神经系统肿瘤的甲基化谱 t-SNE 分析Methylation t-SNE analysis of CIC-fused CNS tumors
Hou Y, et al. Acta Neuropathologica Communications, 2024 · 第一作者first author
AICNS v2 报告示例
AICNS v2 分类器输出报告示例Sample output report from the AICNS v2 classifier
分层分类 + 三态判读 + t-SNE/CNV(科研用途·示例数据)hierarchical class + three-state readout + t-SNE/CNV (research use · demo data)

第一作者代表性论著Selected first-author publications

  • Hou Y, Pinheiro J, Sahm F, et al. Papillary glioneuronal tumor exhibits a characteristic methylation profile and fusions involving PRKCA. Acta Neuropathologica, 2019.
  • Hou Y, Du Y, et al. Pediatric CNS tumor with CIC::LEUTX fusion: a diagnostic challenge. Acta Neuropathologica Communications, 2024.
  • Xu Y, Hou Y, Gao X, et al. High-grade neuroepithelial neoplasms harbouring EP300::BCOR fusions. Brain Pathology, 2023.
  • Hou Y, Sahm F. What the neuropathologist needs to tell the clinician concerning WHO CNS5. Glioma, 2023.
核心服务Core services

从 AI 分类器到科研方法学,全程支持From the AI classifier to methodology, end to end

两条互补的科研服务线:一套面向科研的脑肿瘤 AI 甲基化分类器,以及面向研究团队的临床甲基化研究全流程方法学辅导。Two complementary research service lines: an AI methylation classifier for CNS tumors, and full-pipeline methodology advisory for clinical methylation-research teams.

01

AICNS · 中枢神经系统肿瘤 AI 甲基化分类器AICNS · AI Methylation Classifier for CNS Tumors

AI-based CNS Tumor Methylation Classification

基于冻结特征面板的 AI 集成分类器,覆盖中枢神经系统肿瘤参照分类,输出校准化的三态判读与结构化报告(科研用途)。An AI ensemble classifier on a frozen feature panel, covering CNS tumor reference classes, with calibrated three-state readout and structured reporting (research use).

  • 双平台输入:Illumina 甲基化芯片(450K/EPIC/EPICv2)与 Oxford Nanopore 全探针测序。Dual-platform input: Illumina arrays (450K/EPIC/EPICv2) and Oxford Nanopore full-probe sequencing.
  • 精细亚型:104 个参照甲基化群,含髓母细胞瘤 13 亚型、室管膜瘤 8 亚型等罕见类识别。Fine-grained subtyping: 104 reference classes, including 13 medulloblastoma and 8 ependymoma subtypes plus rare entities.
  • 辅助分析:MGMT-STP27、拷贝数变异(CNV)、UMAP / t-SNE 定位、肿瘤微环境解卷积。Ancillary analyses: MGMT-STP27, copy-number variation (CNV), UMAP/t-SNE localization and tumor-microenvironment deconvolution.
  • 结构化报告:分层分类卡、定位图、CNV 全基因组图与整合判读说明的中文 PDF(科研用途)。Structured report: a PDF with hierarchical classification cards, localization plots, genome-wide CNV and integration notes (research use).
02

临床甲基化研究方法学辅导Clinical methylation research advisory

Research & Methodology Consulting

面向开展 CNS 肿瘤甲基化研究的团队,提供从原始数据到模型落地的全流程思路辅导与方法学指导。For teams running CNS-tumor methylation studies — guidance from raw data to a deployable model, with rigorous methodology throughout.

  • 数据处理:IDAT / bedMethyl 到 beta 矩阵的预处理、质控、平台对齐与探针归一。Data processing: IDAT / bedMethyl to beta-matrix preprocessing, QC, cross-platform harmonization and probe normalization.
  • 模型训练:特征选择、集成建模、概率温度校准、离群与弃判阈值设计。Model training: feature selection, ensemble modeling, temperature calibration, OOD and abstention-threshold design.
  • 方法学指导:研究设计、罕见类原型策略、独立验证方案与对标 WHO CNS5 的分类思路。Methodology guidance: study design, rare-class prototype strategies, independent validation and WHO CNS5-aligned classification.
  • 纳米孔衍生:面向 ONT 测序的靶向面板、覆盖自适应转换与跨平台锚定方案咨询。Nanopore derivations: targeted panels for ONT sequencing, coverage-adaptive conversion and cross-platform anchoring.
技术与方法学Technology & methodology

严谨、可校准、对不确定性诚实Rigorous, calibrated and honest about uncertainty

我们的分类引擎不仅追求准确率,更强调可校准的置信度与对不确定性的诚实表达——这正是高质量甲基化分类算法的核心。Our engine optimizes not only accuracy but calibrated confidence and an honest treatment of uncertainty — the core of a high-quality methylation-classification algorithm.

104
参照甲基化分类群reference methylation classes
2 ×
检测平台:芯片 + 纳米孔platforms: array + Nanopore
84.2%
EPICv2 独立验证 · 家族级准确率family-level accuracy, independent EPICv2 validation
15K
冻结 CpG 特征面板CpG frozen feature panel
平台稳健建模Platform-robust modeling

集成分类 + 概率校准Calibrated ensemble classification

先剔除平台间系统性差异最大的 CpG,再以最近质心(平台稳健的相关系数)与随机森林双读集成,并经温度校准给出可比概率。EPICv2 严格作为独立验证集,不参与训练或调参。After removing the CpGs most affected by cross-platform bias, a nearest-centroid and random-forest dual read are ensembled and temperature-calibrated. EPICv2 is held out strictly for independent validation.

辅助证据链Ancillary evidence

CNV · MGMT · 解卷积CNV · MGMT · deconvolution

在分类之外提供拷贝数变异、MGMT-STP27 启动子甲基化状态、UMAP/t-SNE 临近群定位,以及肿瘤纯度与微环境解卷积,构成完整、可解释的证据链。Beyond classification: copy-number variation, MGMT-STP27 status, UMAP/t-SNE localization, and deconvolution of tumor purity and microenvironment — a complete, interpretable evidence chain.

可报告Reportable
score > 0.85
置信度充分,可作为甲基化分类结果报告。Confidence sufficient to report the methylation class.
仅供参考Reference only
0.60 – 0.85
提示性结果,需结合更多证据综合判读。Suggestive only; integrate with additional evidence.
不适用Not applicable
< 0.60 / OOD
置信不足或离群,明确弃判,不强行给出分类。Low confidence or out-of-distribution — abstained, no forced call.
关于我们About us

专注甲基化谱,破解神经肿瘤的分类难题Solving CNS tumor classification through the methylation profile

默赛生基因科技专注于基于 DNA 甲基化谱的肿瘤分类研究。我们整合全球公开甲基化数据集与本地临床样本,构建肿瘤甲基化表型分类模型,致力于攻克中枢神经系统(CNS)肿瘤、尤其是儿童脑肿瘤的分类难题Mosaisheng Gene Technology focuses on DNA-methylation-based tumor classification research. By integrating globally available reference datasets with locally curated clinical samples, we build methylation-phenotype classification models to tackle the classification challenges of central nervous system (CNS) tumors — pediatric brain tumors in particular.

通过整合尖端甲基化检测、前沿计算方法与深厚的神经病理专业知识,我们提供精确、可靠、可解释的肿瘤分类科研解决方案与算法模型Combining advanced methylation assays, state-of-the-art computation and deep neuropathology expertise, we deliver accurate, reliable and interpretable classification research solutions and algorithmic models.

依托与临床及科研伙伴的广泛合作,我们致力于把前沿方法学转化为可操作的科研课题合作与分类模型方案,推动神经肿瘤学研究的创新。Through extensive collaboration with clinical and research partners, we translate frontier methodology into actionable research collaborations and classification-model solutions that advance neuro-oncology research.

海德堡溯源Heidelberg lineage

方法学承袭 Heidelberg / DKFZ 神经病理甲基化分类体系,对标国际标准。Methodology rooted in the Heidelberg / DKFZ neuropathology methylation-classification ecosystem.

本地化样本Locally grounded

结合国内临床样本进行验证与校准,贴合本地科研与样本场景。Validated and calibrated with domestic clinical samples for local research settings.

可解释 · 可信Interpretable & honest

校准概率 + 三态弃判 + 离群检测,测不准即明确弃判。Calibrated probabilities, three-state abstention and out-of-distribution detection — uncertainty is reported, not hidden.

创始人Founder

深耕神经病理与甲基化分类At the intersection of neuropathology and methylation classification

侯仰昊 博士Dr. Yanghao Hou
创始人 · 公司法定代表人Founder & Legal Representative
  • 医学博士(Dr. med., Magna cum laude),德国海德堡大学神经病理科;完成神经病理亚专科实习轮转Dr. med. (Magna cum laude), Dept. of Neuropathology, Heidelberg University; completed subspecialty rotation
  • 博士师从 WHO 肿瘤分类指南编委 Andreas von Deimling 教授Doctoral training under Prof. Andreas von Deimling, a WHO tumor-classification editorial-board member
  • 重庆医科大学基础医学院病理学系 讲师;分子医学检测中心 临床病理医师、甲基化平台负责人Lecturer, Dept. of Pathology, Chongqing Medical University; clinical pathologist & methylation-platform lead, Molecular Medicine Testing Center
  • Acta Neuropathologica 第一作者;累计影响因子 > 30;重庆市英才计划 C 类First author in Acta Neuropathologica; cumulative IF > 30; Chongqing Talent Program (Class C)

从海德堡到重庆的转化之路From Heidelberg to Chongqing

侯仰昊博士在德国海德堡大学神经病理科完成医学博士训练,师从甲基化分类体系奠基人之一、WHO 肿瘤分类指南编委 Andreas von Deimling 教授,系统掌握中枢神经系统肿瘤的甲基化分型、组织学与分子诊断,并具备 R 语言甲基化数据分析(Nanopore / Illumina / NGS)能力。Dr. Hou completed his doctoral training at Heidelberg University's Department of Neuropathology under Prof. Andreas von Deimling, mastering CNS-tumor methylation classification, histology and molecular diagnostics, with R-based methylation analysis across Nanopore, Illumina and NGS.

现任重庆医科大学基础医学院病理学系讲师,并担任分子医学检测中心临床病理医师与甲基化平台负责人,持续从事中枢神经系统肿瘤甲基化谱分类与 AI 算法研究,将国际前沿方法学与本地临床样本相结合——这正是默赛生分类服务的根基。He is now a lecturer in Pathology at Chongqing Medical University and serves as clinical pathologist and methylation-platform lead at the Molecular Medicine Testing Center, continuing research on CNS-tumor methylation classification and AI algorithms that bridges international methodology with local clinical samples — the foundation of Mosaisheng's services.

学术任职Academic appointments

国家神经疾病医学中心胶质瘤 MDT 专科联盟理事 · 欧洲神经肿瘤组委会成员 · 欧洲神经病理学组会委员 · 中国微循环学会转化医学专业青年委员会委员Council member, National Glioma MDT Alliance · member, European neuro-oncology committee · member, European neuropathology working group · youth committee, Chinese Society for Microcirculation (Translational Medicine)

海德堡神经病理团队
显微镜下阅片
4000+
例脑肿瘤及疑难罕见病例诊断经验brain-tumor & rare cases reviewed
亚专科会诊Subspecialty consultation

脑肿瘤亚专科精准诊断咨询Brain-tumor subspecialty diagnostic consultation

  • 3 年德国海德堡大学神经病理学临床诊断经历,完成神经病理亚专科实习轮转3 years of neuropathology clinical-diagnostic training at Heidelberg University, with a completed subspecialty rotation
  • 累计 4000+ 例脑肿瘤及疑难罕见病例的病理与分子病理诊断经验Experience across 4,000+ brain-tumor and difficult/rare cases in pathology and molecular pathology
  • 基于甲基化谱分类与 NGS 检测结果的整合诊断经验Integrated diagnosis combining methylation-profile classification and NGS results
  • 师从 WHO 肿瘤分类指南编委 Andreas von Deimling 教授Trained under Prof. Andreas von Deimling, a WHO tumor-classification editorial-board member
  • 欧洲神经肿瘤组委会成员 · 欧洲神经病理学组会委员 · 国家神经疾病医学中心胶质瘤 MDT 专科联盟理事European neuro-oncology & neuropathology committee member · council member, national glioma MDT alliance
申请会诊 会诊专属邮箱:Consultation email: yanghao.hou@outlook.de

说明:本会诊为神经病理亚专科的专业意见咨询服务,所出具的意见供临床医师与患者参考,不替代主诊医师及出具报告机构的最终诊断与诊疗决策。Note: this is a professional second-opinion neuropathology consultation. Opinions provided are for reference by clinicians and patients and do not replace the final diagnosis or treatment decisions of the attending physician and reporting institution.

会诊申请Consultation request

推荐通过在线申请表提交(支持切片 / 影像上传、自动通知);也可直接填写下方表单生成邮件。受理后将反馈资料要求、时限与费用。Submit via the online form (supports slide / imaging uploads and auto-notification); or use the form below to draft an email. We will reply with required materials, turnaround and fees.

在线会诊申请表(推荐)Online consultation form (recommended)
支持切片 / 影像等资料上传,提交后自动通知,可在表内完成报价与付费。Upload slides / imaging, get auto-notified, and handle quote & payment in-form.
填写在线申请表
或直接发送至 or email yanghao.hou@outlook.de

开启一次甲基化科研合作Start a methylation collaboration

无论是分类算法合作、数据分析方法学,还是脑肿瘤亚专科会诊,欢迎与我们联系。Whether for classification-algorithm collaboration, data-analysis methodology, or brain-tumor subspecialty consultation — we would be glad to hear from you.