WGCNA Signed网络分析出现负kME值的原因咨询

我正在开展WGCNA分析，遇到了诸多难题。我在bwnet分析中采用了Signed网络（TOMType = "signed"），原本认为这意味着仅考虑正相关，负相关会被设为0。但当我绘制kME与Gene Significance的关系图时，发现大量基因与模块特征基因表现相反（存在负kME值），我无法理解该现象的原因，希望能得到帮助，目前我对所有数据都产生了质疑。

网络构建代码 #

#Network construction

bwnet <- blockwiseModules(norm_counts,
                            maxBlockSize = 18456,
                            TOMType = "signed",
                            power = soft_power,
                            mergeCutHeight = 0.25,
                            numericLabels = FALSE,
                            randomSeed = 1234,
                            verbose = 3)
  
  cor <- temp_cor

kME与Trait相关性绘图代码 #

#Plotting kME vs Correlation to trait
# load needed libraries----
      library(ggplot2)
      library(WGCNA)
      library(dplyr)
      library(tidyr)
      library(patchwork)  # or use facets in ggplot
      
      
      #load all files needed and prepare data----
      load("results_V4/norm_counts.RData")
      load("results_V4/bwnet.RData")
      data <- read.csv("count_data_df_JT.csv", sep= ";", row.names = "genes")
      meta <- read.csv("sample_info.csv", sep =";", row.names = "sampletype")
      growth_condition <- meta$growth
      
      # Convert growth condition to numeric (1 for 'c', 0 for 'sg')
      growth_condition_numeric <- ifelse(growth_condition == "c", 1, 0)
      
      
      # Transpose norm_counts if needed genes(columns) , samples (rows)
      norm_counts_t <- t(norm_counts)
      
      # Prepare eigengenes
      module_eigengenes <- bwnet$MEs
      module_colors <- bwnet$colors
      unique_modules <- setdiff(unique(module_colors), "grey")
      
      # Recalculate module membership (kME)
      module.membership.measure <- cor(norm_counts_t, module_eigengenes, use = "p")
      
      # Get gene significance (GS) for growth
      GS <- apply(norm_counts, 1, function(x) cor(x, growth_condition_numeric)) 
      
      # Prepare plotting dataframe
      plot_df <- data.frame()
      
      for (mod in unique_modules) {
        # Get genes belonging to the current module
        module_genes <- names(module_colors[module_colors == mod])
        
        # Get the module membership (kME) for the genes in the current module
        module_kME <- module.membership.measure[module_genes, paste0("ME", mod)]
        
        # Get Gene Significance (GS) for the current module
        GS_module <- GS[module_genes]
        
        # Create a dataframe to store the module's kME and GS
        df <- data.frame(
          Module = mod,
          kME = module_kME,
          GS = GS_module
        )
        
        # Calculate Pearson correlation and p-value for the TS-MM
        corr <- cor(df$kME, df$GS, use = "complete.obs")
        pval <- corPvalueStudent(corr, nrow(df))  # Calculate p-value for correlation
        
        # Add correlation label and number of genes in the module
        df$corr_label <- paste0("r = ", round(corr, 2), ", p = ", format.pval(pval, digits = 2))
        df$n_genes <- length(module_genes)
        
        # Append to the plot dataframe
        plot_df <- rbind(plot_df, df)
      }
      
      # Step 5: Plotting using ggplot2
      pdf("module_membership_vs_trait_significance.pdf", width = 10, height = 8) 
      ggplot(plot_df, aes(x = kME, y = GS, color = Module)) +
        geom_point(alpha = 0.7) +  # Scatter plot of kME vs GS
        geom_smooth(method = "lm", color = "black") +  # Regression line for TS-MM
        facet_wrap(~Module, scales = "free", ncol = 4) +  # Create a facet for each module
        labs(x = "Module Membership (kME)",
             y = "Gene Significance for Growth Condition",
             title = "Module Membership vs Trait Significance (TS-MM)") +
        theme_minimal() +
        geom_text(
          data = plot_df %>% distinct(Module, corr_label, n_genes),
          aes(x = -Inf, y = Inf, label = paste0("n=", n_genes, "\n", corr_label)),
          hjust = -0.1, vjust = 1.2,
          inherit.aes = FALSE
        )
      dev.off() 

核心误解澄清 #

你对TOMType = "signed"的功能理解有误：

• signed模式不会将负相关置为0：它通过公式(1 + cor)/2计算邻接矩阵，负相关会被转换为0~0.5的低值，而非直接丢弃。该模式的作用是让模块划分优先聚合正共表达的基因，但不排除模块内存在表达模式相反的基因。
• 若需要完全忽略负相关、仅基于正相关构建网络，应使用TOMType = "signed hybrid"，该模式会直接将负相关设为0。

负kME值的合理性 #

模块特征基因（ME）是模块内所有基因表达的第一主成分，它捕捉的是模块内最显著的整体表达趋势。即使模块由正共表达基因聚合而成，少数基因（比如模块内的负调控因子、表达量波动与整体趋势相反的基因）仍可能与ME呈现负相关，从而出现负kME值——这是正常的生物学现象，不代表分析流程出错。

代码优化与验证建议 #

1. 规范kME计算：你手动计算kME的方式是正确的，但可以使用WGCNA内置的signedKME函数，结果更贴合WGCNA的分析逻辑：

# 替换手动计算kME的代码
module.membership.measure <- signedKME(norm_counts_t, module_eigengenes, outputColumnName = "kME")

2. 验证模块表达模式：绘制模块内基因的表达热图，确认模块内大部分基因的表达模式是否一致，少数负kME基因是否属于合理例外：

# 替换为你的目标模块颜色
target_module <- "blue"
module_genes <- names(module_colors[module_colors == target_module])
module_expr <- norm_counts[module_genes, ]
heatmap(t(module_expr), scale = "row", col = colorRampPalette(c("blue", "white", "red"))(100), main = paste("Expression Heatmap of", target_module, "Module"))

内容的提问来源于stack exchange，提问作者Laura