Methods in brief / 方法概要
Program definition. Direction-consistent differentially expressed genes shared by the MD discovery cohort (GSE202657, 9 vs 9) and the SLE discovery cohort (GSE50772, 61 vs 20; limma, adj. P < 0.05 and |logFC| > 0.585 in both) define the 123-gene shared set; grading by independent-cohort direction replication (T0 = all three layers, 13 genes; T1 = two, 52; T2 = one, 40; T3 = none, 18) and bootstrap stability selection define the locked 65-gene core program (36 up, 29 down).
程序定义：MD 发现队列（GSE202657，9/9）与 SLE 发现队列（GSE50772，61/20）中方向一致的差异基因构成 123 基因共享集（两病均 limma adj. P<0.05 且 |logFC|>0.585）；按独立队列方向复现层数分级（T0 三层 13 个、T1 两层 52、T2 一层 40、T3 零层 18），经自助稳定性选择锁定 65 基因核心程序（上调 36、下调 29）。
Locked-formula scoring. Per-sample ssGSEA scores (GSVA v2) for the up and down arms, computed separately, unchanged in every cohort. Cohort statistic: S = Cliff's δ(up) − Cliff's δ(down) with 10,000 label permutations. Declared input transforms: arrays as deposited; counts as log2(CPM+1); RPKM as log2(x+1).
锁定公式打分：逐样本 ssGSEA（GSVA v2）分别计算上调臂与下调臂，所有队列不作任何改动。队列统计量 S = Cliff's δ（上调）− Cliff's δ（下调），10,000 次标签置换。预先声明的输入变换：芯片用提交值；计数用 log2(CPM+1)；RPKM 用 log2(x+1)。
Evidence layers. Single-cell localisation: donor-level pseudobulk of the CELLxGENE lupus atlas (program strongest in the myeloid/DC axis; cDC per-gene concordance 55/65). Communication: CellChat — no program gene is a ligand/receptor in SLE; NAMPT (VISFATIN) is the sole MD exception. Perturbation: CellOracle and scTenifoldKnk on upstream regulators (see target table); Geneformer in-silico deletion in SLE classical monocytes — Impact_component_percent: CEBPB 40.28%, TNFAIP3 35.65%, STAT1 28.71% (highest for CEBPB; see the target cards for the full-precision values).
证据层：单细胞定位——CELLxGENE 狼疮图谱供体级伪 bulk，程序最强于髓系/DC 轴（cDC 逐基因一致 55/65）。细胞通讯——CellChat 显示 SLE 中无程序基因作配体/受体，MD 中仅 NAMPT（VISFATIN）例外。扰动——CellOracle 与 scTenifoldKnk 作用于上游调控因子（见靶点表）；Geneformer 虚拟敲除（SLE 经典单核细胞）——Impact_component_percent：CEBPB 40.28%、TNFAIP3 35.65%、STAT1 28.71%（CEBPB 最高；完整精度数值见各靶点卡片）。
Reading the numbers. In the evidence matrix, NA means “not tested under the pre-specified design”, never “tested and negative”. Overlap statistics on the Analyze page are descriptive research aids, not diagnostics.
读数规则：证据矩阵中 NA 表示“按预先设计未测”，不是“测了为阴性”。“算”页面的重叠统计是描述性研究辅助，不是诊断工具。
Downloads / 下载
- evidence_matrix_123.csv — the 123-gene evidence matrix (v2 columns plus Geneformer impact-%)
- perturbation_targets.csv — the 14 perturbation targets
- step7_specificity_map_full.csv — the 14-cohort specificity map
- score_program.R — locked-formula scoring in R (GSVA v2)
- Archive of record: Zenodo DOI 10.5281/zenodo.23680382 (CC BY 4.0)
How to cite / 引用：Lei, P. Multi-layer evidence matrix for the 123 shared genes of Ménière's disease and systemic lupus erythematosus. Zenodo, 2026. doi:10.5281/zenodo.23680382.