单核分析脚本
作用:08_single_nucleus 流程的 Python 与 R 实现脚本,由上一级目录的 7 个 *.sh 按相对路径调用。
入口:无独立入口,由上一级入口脚本驱动(python ./Script/<name>.py、Rscript ./Script/<name>.R)。
输入:../02_processData/ 下的 *_filtered.h5ad、data_clusterd_all_gene_{n}.h5ad、zz_00_pea.rds、All_species_merged_by_GifuT2T.SNF_Symbol.tsv;../04_Anno/ 下的 marker CSV 与 Gifu T2T GTF。
输出:../02_processData/ 的 h5ad/rds/CSV、../03_figures/ 的图、../04_Anno/ 的 Seurat 对象与 marker 表,以及每个基因目录下的 01_runs/、02_summaries/、03_plots/。
运行:bash ../run_singlecell.sh、bash ../05_cluster_DEGs_finder.sh 等;无入口脚本的成员手动运行,目录名 Script 不可更改。
工具:scanpy、anndata、scvi-tools、leidenalg、rpy2(Python 3.9.19);Seurat、scTenifoldKnk、DESeq2、clustree(R 4.4.1)。
English
Purpose: The Python and R implementation scripts of the 08_single_nucleus workflow, called by relative path from the seven *.sh entry scripts in the parent directory.
Entry point: none of their own; they are driven by the parent entry scripts (python ./Script/<name>.py, Rscript ./Script/<name>.R).
Inputs: *_filtered.h5ad, data_clusterd_all_gene_{n}.h5ad, zz_00_pea.rds and All_species_merged_by_GifuT2T.SNF_Symbol.tsv under ../02_processData/; the marker CSVs and the Gifu T2T GTF under ../04_Anno/.
Outputs: h5ad/rds/CSV files in ../02_processData/, plot files in ../03_figures/, Seurat objects and marker tables in ../04_Anno/, and 01_runs/, 02_summaries/, 03_plots/ inside each gene directory.
Run: bash ../run_singlecell.sh, bash ../05_cluster_DEGs_finder.sh, and so on; the members without an entry script are run by hand, and the Script directory name must not change.
Tools: scanpy, anndata, scvi-tools, leidenalg, rpy2 (Python 3.9.19); Seurat, scTenifoldKnk, DESeq2, clustree (R 4.4.1).
以下为该步骤目录中的实际文件(同步快照,与该目录内容一致)。
01_annaDataQC.py · Python · 40 行
import sysimport anndataimport pandas as pdimport numpy as npimport scanpy as scimport matplotlib.pyplot as pltimport seaborn as sns
Dpi00_1 =anndata.read_h5ad("{PROJ_LOTUS_HX}/07_singleCell/02_processData/Dpi00_1_filtered.h5ad")Dpi00_2 =anndata.read_h5ad("{PROJ_LOTUS_HX}/07_singleCell/02_processData/Dpi00_2_filtered.h5ad")Dpi02_1 =anndata.read_h5ad("{PROJ_LOTUS_HX}/07_singleCell/02_processData/Dpi02_1_filtered.h5ad")Dpi02_2 =anndata.read_h5ad("{PROJ_LOTUS_HX}/07_singleCell/02_processData/Dpi02_2_filtered.h5ad")
Dpi00 = Dpi00_1.concatenate(Dpi00_2)Dpi02 = Dpi02_1.concatenate(Dpi02_2)
samples=["Dpi00", "Dpi02"]fig, aex = plt.subplots(ncols=2, nrows=2, figsize=(50, 30))for idx, data in enumerate([Dpi00, Dpi02]): data_concatenated = data data.write_h5ad(f"02_processData/{samples[idx]}_concatenated.h5ad") data.layers['counts'] = data.X.copy() sc.pp.normalize_total(data,target_sum=1e4,inplace=True) sc.pp.log1p(data) data_scaled = data.copy() data.write_h5ad(f"02_processData/{samples[idx]}_scaled.h5ad") sns.histplot(data_concatenated.obs["total_counts"], bins=100,kde=True,ax=aex[idx,0]) aex[idx,0].set_title(f"{samples[idx]}_Total counts") sns.histplot(data_scaled.X.sum(1), bins=100,kde=True,ax=aex[idx,1]) aex[idx,1].set_title(f"{samples[idx]}_Shifted logarithm")plt.savefig("03_figures/hist_counts.png")
adata_list =[]for sample in samples: ad = anndata.read_h5ad(f"02_processData/{sample}_scaled.h5ad") ad.obs["Sample"] = sample adata_list.append(ad)
data_all = anndata.concat(adata_list)data_all.write_h5ad("02_processData/data_all.h5ad")02_integrate.py · Python · 56 行
import sysimport anndataimport scanpy as scimport scvi
# 读取数据data_scaled = anndata.read_h5ad("./02_processData/data_all.h5ad")
# 循环运行不同的 n_top_genesfor n in [1000, 2000, 3000, 4000, 5000]:
print(f"==== Running n_top_genes = {n} ====")
# 每次都复制数据,避免被上一次处理修改 data = data_scaled.copy()
# 选择高变基因 sc.experimental.pp.highly_variable_genes( data, flavor="pearson_residuals", layer="counts", batch_key="Sample", n_top_genes=n, subset=True, inplace=True, )
# 设置 scVI anndata scvi.model.SCVI.setup_anndata( data, layer="counts", batch_key="Sample", categorical_covariate_keys=["Sample"], continuous_covariate_keys=["total_counts"] )
# 固定随机种子 scvi.settings.seed = 7
# 初始化模型 model = scvi.model.SCVI(data)
# 训练模型 model.train()
# 保存 latent space data.obsm['X_scVI'] = model.get_latent_representation()
# 保存 normalized expression data.layers['scvi_normalized'] = model.get_normalized_expression(library_size=1e4)
# 输出文件名 out_file = f"02_processData/data_integrated_{n}.h5ad" data.write_h5ad(out_file)
print(f"==== Saved: {out_file} ====\n")03_1-cluster_statitcs.py · Python · 123 行
#!/usr/bin/env python3# -*- coding: utf-8 -*-
import sysimport anndataimport pandas as pdimport numpy as npimport matplotlib.pyplot as pltimport os
# =============================# 1. 读取命令行参数# =============================if len(sys.argv) < 2: print("请提供 n 值,例如:python 03_1-cluster_statitcs.py 4000") sys.exit(1)
n = sys.argv[1]
# =============================# 2. 读取 .h5ad 数据# =============================h5_file = f"02_processData/data_clusterd_{n}.h5ad"adata = anndata.read_h5ad(h5_file)
df = adata.obs.copy()
# =============================# 3. 提取 cluster & Sample# =============================df["cluster"] = df["leiden_0.4"].astype(str)df["sample"] = df["Sample"].astype(str)
# =============================# 4. cluster 指定顺序# =============================cluster_order = ['9', '3', '4', '7', '2', '6', '8', '0', '1', '5']df["cluster"] = pd.Categorical(df["cluster"], categories=cluster_order, ordered=True)
# =============================# 5. 每个 cluster 的总细胞数# =============================cluster_counts = df.groupby("cluster").size().reset_index(name="total_cells")
# =============================# 6. 各 Sample 在每 cluster 的真实细胞数(raw counts)# =============================cluster_sample_count = df.groupby(["cluster", "sample"]).size().reset_index(name="count")
cluster_sample_count = cluster_sample_count.merge(cluster_counts, on="cluster", how="left")
# =============================# 7. pivot 创建 raw count 堆叠矩阵# =============================plot_df = cluster_sample_count.pivot_table( index="cluster", columns="sample", values="count", fill_value=0)
plot_df = plot_df.reindex(cluster_order)
# =============================# 8. 绘图(横向 raw count stacked bar)# =============================fig, ax = plt.subplots(figsize=(10, 8), dpi=300)
# 自定义颜色palette = ["#1F77B4", "#FF7F0E"]samples = plot_df.columns.tolist()colors = (palette * (len(samples) // len(palette) + 1))[:len(samples)]
# 横向堆叠条形图(使用真实数量)plot_df.plot( kind="barh", stacked=True, ax=ax, color=colors, width=0.7)
# =============================# 9. 添加 cluster 总细胞数注释(右侧)# =============================for idx, row in cluster_counts.iterrows(): cluster = row["cluster"] total_cells = row["total_cells"] y_pos = cluster_order.index(cluster)
ax.text( plot_df.sum(axis=1).max() * 1.02, # 在右侧一点点 y_pos, str(total_cells), va="center", ha="left", fontsize=9, fontweight="bold" )
# =============================# 10. 删除外边框,仅保留坐标轴# =============================for spine in ["top", "right"]: ax.spines[spine].set_visible(False)
ax.tick_params(axis='both', which='both', length=4)
# =============================# 图形标签# =============================ax.set_title(f"Cluster Composition (raw cell counts, n_top_genes = {n})")ax.set_xlabel("Number of Cells")ax.set_ylabel("Cluster")ax.legend(samples, title="Sample", bbox_to_anchor=(1.05, 1), loc="upper left")
plt.tight_layout()
# =============================# 11. 保存 PDF# =============================os.makedirs("03_figures", exist_ok=True)out_file = f"03_figures/cluster_stacked_raw_{n}.pdf"plt.savefig(out_file, bbox_inches="tight")
print("完成绘图:", out_file)03_1-dimenRedu_adjusted.py · Python · 122 行
import syssys.path.append('{CONDA_SINGLECELL}/lib/python3.9/site-packages/Scripts')import anndataimport pandas as pdimport numpy as npimport scanpy as scimport scipy.sparse as spimport scviimport Scriptsimport matplotlib.pyplot as pltimport seaborn as snsimport rpy2.robjects as rofrom rpy2.robjects import pandas2rifrom rpy2.robjects.packages import importr
pandas2ri.activate()clustree = importr('clustree')ggplot2 = importr('ggplot2')
# 循环运行不同的 n_top_genesfor n in [1000, 2000, 3000, 4000, 5000]:
print(f"==== Running n_top_genes = {n} ====")
in_file = f"02_processData/data_integrated_{n}.h5ad"
data_integrated = anndata.read_h5ad(in_file) data = data_integrated
palette = ["#1F77B4","#FF7F0E"]
categories = data.obs['Sample'].cat.categories color_map = {category: color for category, color in zip(categories, palette)}
###降维 ## UMAP # 通过降低n_neighbors和min_dist让各类之间的边界更清晰 sc.pp.neighbors(data, n_neighbors=10, use_rep='X_scVI') sc.tl.umap(data, min_dist=0.3) data.obsm["scvi_umap"]=data.obsm["X_umap"]
fig,ax = plt.subplots(figsize=(10,6),dpi=300) sc.pl.embedding(data,basis='scvi_umap',color="Sample",palette=color_map,show=False, ax=ax) plt.subplots_adjust(left=0.1, right=0.8, top=0.9, bottom=0.1) handles, labels = ax.get_legend_handles_labels() ax.legend(handles, labels, title='Sample', bbox_to_anchor=(1.05, 1), loc='upper left') out_file = f"03_figures/dimenRedu_{n}_adjusted.png" plt.savefig(out_file, bbox_inches='tight') plt.close()
##聚类 resolutions = [0.2, 0.4, 0.6, 0.8, 1.0, 1.2]
for res in resolutions: sc.tl.leiden(data, resolution=res, key_added=f'leiden_{res}',random_state=0)
fig,axes=plt.subplots(3,2,figsize=(30,15)) plt.subplots_adjust(left=0.1,bottom=0.1,top=0.9,right=0.8,hspace=0.2,wspace=0.4) sc.pl.embedding(data,basis='scvi_umap',ax=axes[0,0],color="leiden_0.2") sc.pl.embedding(data,basis='scvi_umap',ax=axes[0,1],color="leiden_0.4") sc.pl.embedding(data,basis='scvi_umap',ax=axes[1,0],color="leiden_0.6") sc.pl.embedding(data,basis='scvi_umap',ax=axes[1,1],color="leiden_0.8") sc.pl.embedding(data,basis='scvi_umap',ax=axes[2,0],color="leiden_1.0") sc.pl.embedding(data,basis='scvi_umap',ax=axes[2,1],color="leiden_1.2") out_file = f"03_figures/cluster_{n}_adjusted.png" plt.savefig(out_file) plt.close()
cluster_df = pd.DataFrame({f'leiden_{res}': data.obs[f'leiden_{res}'] for res in resolutions}) r_cluster_df = pandas2ri.py2rpy(cluster_df) ro.globalenv['cluster_df'] = r_cluster_df ro.globalenv['n'] = n r_code = """ library(clustree) library(ggplot2) p <- clustree(cluster_df, prefix = "leiden_") out_file = paste0("03_figures/clustree_plot_", n, ".pdf") ggsave(out_file, plot = p, device = "pdf", width=16, height=8) """ ro.r(r_code)
# final fig,axes=plt.subplots(2,1,figsize=(8,10),dpi=300) plt.subplots_adjust(left=0.1,bottom=0.1,top=0.9,right=0.8,hspace=0.2,wspace=0.25) sc.pl.embedding(data,basis='scvi_umap',ax=axes[0],color="Sample",palette=color_map,show=False) sc.pl.embedding(data,basis='scvi_umap',ax=axes[1],color="leiden_0.4",legend_loc="on data",show=False) axes[0].set_title("Samples") axes[1].set_title("Clusters") out_file = f"03_figures/final_cluster_{n}_adjusted.png" plt.savefig(out_file) plt.close()
data.write_h5ad(f"02_processData/data_clusterd_{n}_adjusted.h5ad")
# cluster_statitcs clusters = data.obs["leiden_0.4"] samples = data.obs['Sample'] cluster_counts = clusters.value_counts().sort_index() cluster_sample_counts=pd.crosstab(clusters,samples) cluster_sample_props = cluster_sample_counts.div(cluster_sample_counts.sum(axis=1), axis=0)
fig, axs = plt.subplots(nrows=2, figsize=(12, 16), dpi=300) cluster_counts.plot(kind='bar', ax=axs[0], color='grey') axs[0].set_title('Total Number of Cells in Each Cluster') axs[0].set_xlabel('Cluster') axs[0].set_ylabel('Number of Cells')
cluster_sample_props.plot(kind='bar', stacked=True, color=[color_map[color] for color in categories], ax=axs[1]) axs[1].set_title('Proportion of Samples in Each Cluster') axs[1].set_xlabel('Cluster') axs[1].set_ylabel('Proportion')
handles, labels = axs[1].get_legend_handles_labels() axs[1].legend(handles, labels, title='Sample', bbox_to_anchor=(1.05, 1), loc='upper left')
out_file = f"03_figures/cluster_statitcs_{n}_adjusted.png" plt.tight_layout() plt.savefig(out_file, bbox_inches='tight') plt.close()03_2-fliter_to_all_gene_h5ad.py · Python · 20 行
import anndata
# 循环运行不同的 n_top_genesfor n in [1000, 2000, 3000, 4000, 5000]:
data_concatenated = anndata.read_h5ad("./02_processData/data_all.h5ad")
print(f"==== Running n_top_genes = {n} ====")
in_file = f"02_processData/data_clusterd_{n}.h5ad" data_cluster = anndata.read_h5ad(in_file)
# Copy 'obs' and 'var' if necessary data_concatenated.obs = data_cluster.obs data_concatenated.uns = data_cluster.uns data_concatenated.obsm = data_cluster.obsm #data_concatenated.layers = data_cluster.layers data_concatenated.obsp = data_cluster.obsp
data_concatenated.write_h5ad(f"02_processData/data_clusterd_all_gene_{n}.h5ad")05_1-Seurat_Exp_h5ad_to_rds.py · Python · 37 行
import anndataimport osimport matplotlib.pyplot as pltimport scanpy as scimport scipy.io as sio
os.makedirs("04_Anno", exist_ok=True)
data_cluster = anndata.read_h5ad("./02_processData/data_clusterd_all_gene_4000.h5ad")selected_data = data_clusterselected_data.obs_names_make_unique()#############cell typecluster2annotation = { '0': '0', '1': '1', '2': '2', '3': '3', '4': '4', '5': '5', '6': '6', '7': '7', '8': '8', '9': '9'}
selected_data.obs['celltype'] = selected_data.obs['leiden_0.4'].map(cluster2annotation).astype('category')#############################################
cellinfo=selected_data.obsgeneinfo=selected_data.varmtx2=selected_data.layers['counts'].Tmtx1=selected_data.X.T
cellinfo.to_csv("04_Anno/cluster_01-cellinfo.csv")geneinfo.to_csv("04_Anno/cluster_02-geneinfo.csv")sio.mmwrite("04_Anno/cluster_03-sparse_matrix_counts.mtx",mtx2)sio.mmwrite("04_Anno/cluster_04-sparse_matrix.mtx",mtx1)05_2-Exp_h5ad_to_rds_cluster.R · R · 74 行
# 加载必要的包# dplyr: 数据操作# Seurat: 单细胞数据分析# patchwork: 用于将多个图形组合成一个布局# Biobase: 生物信息学分析工具library(dplyr)library(Seurat)library(patchwork)library(Biobase)
# 读取注释好的 h5ad 文件转换后的数据# 读取细胞信息(cellinfo)、基因信息(geneinfo)和表达矩阵(counts)cellinfo <- read.csv("04_Anno/cluster_01-cellinfo.csv", row.names = 1) # 细胞元数据geneinfo <- read.csv("04_Anno/cluster_02-geneinfo.csv", row.names = 1) # 基因元数据counts <- Matrix::readMM(file = "04_Anno/cluster_03-sparse_matrix_counts.mtx") # 稀疏表达矩阵
# 为稀疏矩阵添加行名(基因)和列名(细胞)rownames(counts) <- rownames(geneinfo)colnames(counts) <- rownames(cellinfo)
# 创建 Seurat 对象# 使用表达矩阵 counts 和细胞元数据 cellinfoAll.merge <- CreateSeuratObject(counts = counts, meta.data = cellinfo)
# 将 Seurat 对象保存为 pbmc(更常用的命名)pbmc <- All.merge
# 数据归一化# NormalizeData 对基因表达数据进行归一化处理pbmc <- NormalizeData(pbmc)
# 筛选高变基因# FindVariableFeatures 用于识别在不同细胞间具有高度可变性的基因pbmc <- FindVariableFeatures(pbmc)
# 计算不同聚类的平均表达量# group.by = "celltype" 按照列 celltype 分组dat <- AverageExpression(pbmc, group.by = "celltype")$RNA
# 将计算的平均表达量保存为 CSV 文件write.csv(dat, file = "04_Anno/cluster_05-AverageExpression_celltype.csv", row.names = TRUE)
### 计算两个cluster的差异表达情况#Idents(pbmc) <- pbmc@meta.data$celltype#cluster5.markers <- FindMarkers(pbmc, ident.1 = c(5,6), ident.2 = c(0,3))
saveRDS(pbmc, file = "04_Anno/cluster_00_pea.rds")
################################################################################library(dplyr)library(Seurat)library(patchwork)library(Biobase)#读取注释好的 h5ad 文件转换后的数据seurat_obj <- readRDS("04_Anno/cluster_00_pea.rds")### 计算cluster的marker geneIdents(pbmc) <- pbmc@meta.data$leiden_0.4##所有clusterall_markers <- FindAllMarkers( object = pbmc, min.pct = 0.25, logfc.threshold = 0.25, only.pos = TRUE)
#写出marker genewrite.csv(all_markers, file = "04_Anno/cluster_06_cluster_markers_gene.csv", row.names = FALSE)
all_markers <- FindAllMarkers( object = pbmc)all_markers <- all_markers[all_markers$p_val_adj<0.05&abs(all_markers$avg_log2FC)>1,]#写出marker genewrite.csv(all_markers, file = "04_Anno/cluster_06_cluster_markers_gene1.csv", row.names = FALSE)05_3-markerAnno_Ps_v5_own.py · Python · 103 行
import sysimport anndataimport pandas as pdimport numpy as npimport scanpy as scimport scipy.sparse as spimport scviimport scriptsimport matplotlib.pyplot as plt
data_clustered = anndata.read_h5ad("./02_processData/data_clusterd_all_gene_4000.h5ad")data=data_clustered
marker_genes = { "Cortex": [ "Lotus_GifuT2T_01G0461100", "Lotus_GifuT2T_02G1487100", ], "Epidermis": [ "Lotus_GifuT2T_05G2864900" ], "Meristem": [ "Lotus_GifuT2T_03G1828600", "Lotus_GifuT2T_04G2253600", "Lotus_GifuT2T_04G2253800" ], "Pericycle": [ "Lotus_GifuT2T_02G1146000", "Lotus_GifuT2T_04G2574000" ], "Phloem": [ "Lotus_GifuT2T_03G2130600", "Lotus_GifuT2T_05G2958500", ], "Vascular": [ "Lotus_GifuT2T_01G0541600", # "Lotus_GifuT2T_01G0093700", "Lotus_GifuT2T_01G0703300", "Lotus_GifuT2T_01G0626500" ], "Xylem": [ "Lotus_GifuT2T_01G0273300", "Lotus_GifuT2T_04G2644400", ], "Infected cell": [ "Lotus_GifuT2T_04G2692200", "Lotus_GifuT2T_01G0784100", "Lotus_GifuT2T_03G1515300", "Lotus_GifuT2T_05G3052700", ], "Nodule primordium": [ "Lotus_GifuT2T_01G0025100", "Lotus_GifuT2T_04G2544700" ], "LjLb": [ "Lotus_GifuT2T_05G2767100", "Lotus_GifuT2T_05G2767300", "Lotus_GifuT2T_05G2792500" ]}
marker_genes_in_data = dict()for ct, markers in marker_genes.items(): markers_found = list() for marker in markers: if marker in data.var.index: markers_found.append(marker) marker_genes_in_data[ct] = markers_found
sc.tl.dendrogram(data,'leiden_0.4',use_rep="X_scVI")
print(marker_genes_in_data)
# 指定聚类顺序,按实际需求调整列表顺序# cluster_order = ['3', '2', '6','12','10','7','8','0','1','4','5','9','11','13','14'] # 替换为你的实际顺序# cluster_order = ['8', '6', '1', '0', '5', '4', '3', '2', '7', '9']cluster_order = ['0', '1', '5', '8', '6', '2', '7', '3', '4', '9']
# 将 leiden_0.4 列转换为有序分类data.obs['leiden_0.4'] = pd.Categorical( data.obs['leiden_0.4'], categories=cluster_order, # 强制指定顺序 ordered=True)
# 创建画布fig, axes = plt.subplots(2, 1, figsize=(16, 12), dpi=300)
# 绘制点图(无需 groups_order 参数)sc.pl.dotplot( data, ax=axes[0], groupby="leiden_0.4", var_names=marker_genes_in_data, dendrogram=False, standard_scale="var")
axes[1].axis("off")
plt.savefig("03_figures/markerAnno_filter_dot_v5_own_0.4.pdf")05_3-marker_cluster_DEGs_own_order.py · Python · 187 行
#!/usr/bin/env python3# -*- coding: utf-8 -*-
import matplotlibimport matplotlib.pyplot as pltfrom matplotlib import font_managerimport os
# --------- 目录路径(你的 Arial 字体存放目录) ------------font_dir = "{HOME_HX}/.fonts/Arial"for f in os.listdir(font_dir): if f.endswith(".ttf") or f.endswith(".TTF"): font_manager.fontManager.addfont(os.path.join(font_dir, f))
# --------- 强制 matplotlib 使用 Arial 字体 ------------plt.rcParams['font.family'] = 'Arial'matplotlib.rcParams['pdf.fonttype'] = 42 # 可编辑文本matplotlib.rcParams['ps.fonttype'] = 42
# ------------ 保证AI中可编辑的Arial字体 ------------matplotlib.rcParams['pdf.fonttype'] = 42 # 不要转轮廓matplotlib.rcParams['ps.fonttype'] = 42plt.rcParams['font.family'] = 'Arial'# ------------------------------------------------------
import osimport pandas as pdimport anndataimport scanpy as scimport numpy as np
# ==========================# 0. Read h5ad# ==========================data = anndata.read_h5ad("./02_processData/data_clusterd_all_gene_4000.h5ad")
# ==========================# 1. Cluster order# ==========================# cluster_order = ['9', '3', '4', '7', '2', '6', '8', '0', '1', '5']cluster_order = ['5', '1', '0', '8', '6', '2', '7', '4', '3', '9']
data.obs["leiden_0.4"] = pd.Categorical( data.obs["leiden_0.4"], categories=cluster_order, ordered=True)
# ==========================# 2. Read marker gene CSV# ==========================csv_path = "./04_Anno/cluster_06_cluster_markers_gene.csv"df = pd.read_csv(csv_path)
df["cluster"] = df["cluster"].astype(str)
# ---- 统一 gene 名格式:替换 - 为 _ ----df["gene"] = df["gene"].str.replace("-", "_")
# ==========================# 3. special cluster 8 with corrected gene names# ==========================special_cluster8_genes_raw = [ "Lotus_GifuT2T_04G2692200", "Lotus-GifuT2T-01G0002200", "Lotus-GifuT2T-02G1393700"]
# 替换为与矩阵一致的格式special_cluster8_genes = [g.replace("-", "_") for g in special_cluster8_genes_raw]
cluster_top_genes = {}
# ==========================# 4. Process each cluster# ==========================for cl in cluster_order: sub = df[df["cluster"] == cl]
# Cluster8 special handling if cl == "8": cluster_top_genes[cl] = special_cluster8_genes continue
# Apply filtering rules sub_filtered = sub[ (sub["avg_log2FC"] > 0.25) & (sub["pct.1"] > 0.25) & (sub["p_val_adj"] < 0.05) ]
if sub_filtered.shape[0] == 0: print(f"⚠️ Cluster {cl} has no genes passing filter") continue
# Sort by p_val_adj → avg_log2FC sub_filtered = sub_filtered.sort_values( by=["p_val_adj", "avg_log2FC"], ascending=[True, False] )
# Select top3 cluster_top_genes[cl] = list(sub_filtered["gene"].head(3))
# print final used genesprint("\n最终用于绘图的基因:")for cl in cluster_order: if cl in cluster_top_genes: print(f"Cluster {cl}: {cluster_top_genes[cl]}") else: print(f"Cluster {cl}: 无基因符合条件")
# ==============================================================# 5. DOTPLOT ALL CLUSTERS# ==============================================================
all_genes = []for cl in cluster_order: if cl in cluster_top_genes: all_genes.extend(cluster_top_genes[cl])
# remove genes not in varall_genes = [g for g in all_genes if g in data.var.index]
# 如果仍然没有基因,直接退出防止报错if len(all_genes) == 0: raise ValueError("❌ dotplot 无基因可绘制,检查基因名格式是否都正确匹配 data.var.index")
# 自动根据基因数调整画布宽度n_genes = len(all_genes)fig_width = max(12, n_genes * 0.8)
print(f"绘制 dotplot,基因数 {n_genes},figsize=({fig_width}, 10)")
plt.figure(figsize=(fig_width, 10), dpi=300)
sc.pl.dotplot( data, var_names=all_genes, groupby="leiden_0.4", standard_scale="var", dendrogram=False, show=False)
plt.savefig("./04_Anno/dotplot_all_clusters_top3.pdf", dpi=300)plt.close()
print("\nDOTPLOT Done!")
# ==============================================================# 6. UMAP for each cluster# ==============================================================
out_root = "./04_Anno/UMAP_top/"os.makedirs(out_root, exist_ok=True)
red_gray_cmap = plt.cm.colors.LinearSegmentedColormap.from_list( "redgray", ["lightgray", "red"])
for cl in cluster_order: if cl not in cluster_top_genes: continue
cl_dir = os.path.join(out_root, f"cluster_{cl}") os.makedirs(cl_dir, exist_ok=True)
for gene in cluster_top_genes[cl]: if gene not in data.var.index: print(f"⚠️ {gene} not found in data") continue
sc.pl.umap( data, color=gene, show=False, color_map=red_gray_cmap, s=10 )
plt.title(f"Cluster {cl} - {gene}") plt.savefig(os.path.join(cl_dir, f"{gene}_UMAP.pdf"), dpi=300) plt.close()
print("\nUMAP All Done!")05_3-marker_knowngene_lotus_own_order.py · Python · 123 行
import sysimport anndataimport pandas as pdimport numpy as npimport scanpy as scimport scipy.sparse as spimport scriptsimport matplotlib.pyplot as plt
data_clustered = anndata.read_h5ad("./02_processData/data_clusterd_all_gene_4000.h5ad")data=data_clustered
### marker基因注释marker_genes = {"":["Lotus_GifuT2T_06G3179800","Lotus_GifuT2T_01G0061000","Lotus_GifuT2T_01G0087900","Lotus_GifuT2T_01G0128800"]}
# check if the markers are in the datamarker_genes_in_data = dict()for ct, markers in marker_genes.items(): markers_found = list() for marker in markers: if marker in data.var.index: markers_found.append(marker) marker_genes_in_data[ct] = markers_found
sc.tl.dendrogram(data,'leiden_0.4',use_rep="X_scVI")
print(marker_genes_in_data)
# 指定聚类顺序,按实际需求调整列表顺序cluster_order = ['0', '1', '5', '8', '6', '2', '7', '3', '4', '9'] # 替换为你的实际顺序
# 将 leiden_0.4 列转换为有序分类data.obs['leiden_0.4'] = pd.Categorical( data.obs['leiden_0.4'], categories=cluster_order, # 强制指定顺序 ordered=True)
categories = data.obs['Sample'].cat.categoriesfor sample in categories: selected_cells = data.obs['Sample'] == sample sample_data = data[selected_cells].copy()
marker_genes_in_data = dict() for ct, markers in marker_genes.items(): markers_found = list() for marker in markers: if marker in sample_data.var.index: markers_found.append(marker) marker_genes_in_data[ct] = markers_found
sc.tl.dendrogram(sample_data,'leiden_0.4',use_rep="X_scVI")
fig, axes=plt.subplots(2,1,figsize=(16,12),dpi=300) sc.pl.dotplot( sample_data, ax=axes[0], groupby="leiden_0.4", var_names=marker_genes_in_data, dendrogram=False, standard_scale="var", ) axes[1].axis("off")
plt.savefig(f"03_figures/knowngene_dot_{sample}_1_own_0.4.pdf")
# ================================# 绘制 marker 基因的 UMAP 分布图# ================================# 四个 marker 基因列表genes_to_plot = [ "Lotus_GifuT2T_06G3179800", "Lotus_GifuT2T_01G0061000", "Lotus_GifuT2T_01G0087900", "Lotus_GifuT2T_01G0128800"]
# 逐个 sample 绘制 UMAPfor sample in categories: selected_cells = data.obs['Sample'] == sample sample_data = data[selected_cells].copy()
# 筛选当前 sample 中确实存在的基因 valid_genes = [g for g in genes_to_plot if g in sample_data.var.index]
if len(valid_genes) == 0: print(f"⚠️ Sample {sample} 中没有任何 marker 基因") continue
# ---- 创建一个 2×2 的 panel 图 ---- n = len(valid_genes) ncols = 2 nrows = int(np.ceil(n / ncols))
fig, axs = plt.subplots(nrows, ncols, figsize=(10, 10), dpi=300) axs = axs.flatten()
for i, gene in enumerate(valid_genes): sc.pl.umap( sample_data, color=gene, ax=axs[i], show=False, color_map="viridis", # 可换 plasma, magma s=10 ) axs[i].set_title(gene, fontsize=10)
# 隐藏多余子图 for j in range(i+1, len(axs)): axs[j].axis('off')
plt.tight_layout() plt.savefig(f"03_figures/knowngene_umap_{sample}_4genes.pdf") plt.close()
print("UMAP 分布图已全部绘制完成!")06_1-Seurat_Exp_h5ad_to_rds.py · Python · 38 行
import anndataimport osimport matplotlib.pyplot as pltimport scanpy as scimport scipy.io as sio
data_cluster = anndata.read_h5ad("./02_processData/data_clusterd_all_gene_4000.h5ad")data_cluster.obs_names_make_unique()selected_data = data_cluster
#############cell type#Xylem:1 Phloem:2 Pericycle:3 Vascular:4 Cortex:5 Meristem:6 Early symbiotic signaling cells:7cluster2annotation = { '0': '5', '1': '5', '2': '3', '3': '4', '4': '4', '5': '5', '6': '6', '7': '2', '8': '7', '9': '1'}
selected_data.obs['celltype'] = selected_data.obs['leiden_0.4'].map(cluster2annotation).astype('category')
#############################################
cellinfo=selected_data.obsgeneinfo=selected_data.varmtx2=selected_data.layers['counts'].Tmtx1=selected_data.X.T
cellinfo.to_csv("./02_processData/zz_01-cellinfo.csv")geneinfo.to_csv("./02_processData/zz_02-geneinfo.csv")sio.mmwrite("./02_processData/zz_03-sparse_matrix_counts.mtx",mtx2)sio.mmwrite("./02_processData/zz_04-sparse_matrix.mtx",mtx1)06_2-Exp_h5ad_to_rds_cell_type.R · R · 146 行
# 加载必要的包# dplyr: 数据操作# Seurat: 单细胞数据分析# patchwork: 用于将多个图形组合成一个布局# Biobase: 生物信息学分析工具library(dplyr)library(Seurat)library(patchwork)library(Biobase)library(tidyverse)
# 读取注释好的 h5ad 文件转换后的数据# 读取细胞信息(cellinfo)、基因信息(geneinfo)和表达矩阵(counts)cellinfo <- read.csv("./02_processData/zz_01-cellinfo.csv") # 细胞元数据geneinfo <- read.csv("./02_processData/zz_02-geneinfo.csv", row.names = 1) # 基因元数据counts <- Matrix::readMM(file = "./02_processData/zz_03-sparse_matrix_counts.mtx") # 稀疏表达矩阵
#------------- Step 1. 读取 barcode -------------cell00dpi1 <- read_tsv("00dpi_1_barcodes.tsv", col_names = FALSE)$X1cell00dpi2 <- read_tsv("00dpi_2_barcodes.tsv", col_names = FALSE)$X1cell02dpi1 <- read_tsv("02dpi_1_barcodes.tsv", col_names = FALSE)$X1cell02dpi2 <- read_tsv("02dpi_2_barcodes.tsv", col_names = FALSE)$X1
# 合并为两个集合set00dpi <- unique(c(cell00dpi1, cell00dpi2))set02dpi <- unique(c(cell02dpi1, cell02dpi2))
#------------- Step 2. 清洗 cellinfo$X(去掉第二个 "-" 及后内容)-------------# 原格式如:AAACCCACAACAGCCC-1-0cellinfo$X_clean <- sub("^([^-]+-[^-]+).*", "\\1", cellinfo$X)
#------------- Step 3. 判断属于哪一类并加后缀 -------------cellinfo$X_final <- case_when( cellinfo$X_clean %in% set00dpi ~ paste0(cellinfo$X_clean, "-0"), cellinfo$X_clean %in% set02dpi ~ paste0(cellinfo$X_clean, "-1"), TRUE ~ cellinfo$X_clean # 不在任何列表就保持原样(一般不会出现))
#------------- Step 4. 处理重复并设置行名 -------------cellinfo$X_final_unique <- make.unique(cellinfo$X_final, sep = "_")rownames(cellinfo) <- cellinfo$X_final_uniquecellinfo <- cellinfo %>% select(-X, -X_clean, -X_final, -X_final_unique)
# 为稀疏矩阵添加行名(基因)和列名(细胞)rownames(counts) <- rownames(geneinfo)colnames(counts) <- rownames(cellinfo)
# 创建 Seurat 对象# 使用表达矩阵 counts 和细胞元数据 cellinfoAll.merge <- CreateSeuratObject(counts = counts, meta.data = cellinfo)
# 将 Seurat 对象保存为 pbmc(更常用的命名)pbmc <- All.merge
# 数据归一化# NormalizeData 对基因表达数据进行归一化处理pbmc <- NormalizeData(pbmc)
# 筛选高变基因# FindVariableFeatures 用于识别在不同细胞间具有高度可变性的基因pbmc <- FindVariableFeatures(pbmc)
# 计算不同聚类的平均表达量# group.by = "celltype" 按照列 celltype 分组dat <- AverageExpression(pbmc, group.by = "celltype")$RNA
# 将计算的平均表达量保存为 CSV 文件write.csv(dat, file = "./02_processData/zz_05-AverageExpression_celltype.csv", row.names = TRUE)
### 计算两个cluster的差异表达情况#Idents(pbmc) <- pbmc@meta.data$celltype#cluster5.markers <- FindMarkers(pbmc, ident.1 = c(5,6), ident.2 = c(0,3))
### 将对象拆分为不同时期处理,比较不同时期的表达差异# 假设 pbmc 是一个 Seurat 对象,提取元数据metadata <- pbmc@meta.data
# 提取细胞名称,并获取最后一位数字cell_names <- rownames(metadata)last_digits <- as.numeric(substr(cell_names, nchar(cell_names), nchar(cell_names)))
# 动态调整 cell_type 的值 0-9 10-19 20-29 30-39 40-49metadata$adjusted_celltype <- metadata$celltype + 10 * last_digits
# 将修改后的元数据更新到 Seurat 对象pbmc@meta.data <- metadata
# 查看调整后的结果#head(metadata[, c("leiden_0.6", "adjusted_leiden_0.6")])
# 计算基因表达dat <- AverageExpression(pbmc, group.by = "adjusted_celltype")$RNAwrite.csv(dat, file = "./02_processData/zz_06-AverageExpression_celltype_difftime.csv", row.names = TRUE)
saveRDS(pbmc, file = "./02_processData/zz_00_pea.rds")
################################################################################library(dplyr)library(Seurat)library(patchwork)library(Biobase)#读取注释好的 h5ad 文件转换后的数据seurat_obj <- readRDS("./02_processData/zz_00_pea.rds")### 计算cluster的marker geneIdents(pbmc) <- pbmc@meta.data$adjusted_celltype##所有clusterall_markers <- FindAllMarkers( object = pbmc, min.pct = 0.25, logfc.threshold = 0.25, only.pos = TRUE)#写出marker genewrite.csv(all_markers, file = "./02_processData/zz_07_celltype_difftime_markers_gene.csv", row.names = FALSE)
### 计算cluster的marker geneIdents(pbmc) <- pbmc@meta.data$celltype##所有clusterall_markers <- FindAllMarkers( object = pbmc, min.pct = 0.25, logfc.threshold = 0.25, only.pos = TRUE)#写出marker genewrite.csv(all_markers, file = "./02_processData/zz_07_celltype_markers_gene.csv", row.names = FALSE)
################################################################################library(dplyr)library(Seurat)library(patchwork)library(Biobase)#读取注释好的 h5ad 文件转换后的数据seurat_obj <- readRDS("./02_processData/zz_00_pea.rds")### 计算cluster的marker geneIdents(pbmc) <- pbmc@meta.data$leiden_0.4##所有clusterall_markers <- FindAllMarkers( object = pbmc, min.pct = 0.25, logfc.threshold = 0.25, only.pos = TRUE)#写出marker genewrite.csv(all_markers, file = "./02_processData/zz_08_cluster_markers_gene.csv", row.names = FALSE)06_3-agg_to_fpkm.R · R · 119 行
# 加载必要的包# dplyr: 数据操作# Seurat: 单细胞数据分析# patchwork: 用于将多个图形组合成一个布局# Biobase: 生物信息学分析工具library(dplyr)library(Seurat)library(patchwork)library(Biobase)library(tidyverse)
# 读取注释好的 h5ad 文件转换后的数据# 读取细胞信息(cellinfo)、基因信息(geneinfo)和表达矩阵(counts)cellinfo <- read.csv("./02_processData/zz_01-cellinfo.csv") # 细胞元数据geneinfo <- read.csv("./02_processData/zz_02-geneinfo.csv", row.names = 1) # 基因元数据counts <- Matrix::readMM(file = "./02_processData/zz_03-sparse_matrix_counts.mtx") # 稀疏表达矩阵
#------------- Step 1. 读取 barcode -------------cell00dpi1 <- read_tsv("00dpi_1_barcodes.tsv", col_names = FALSE)$X1cell00dpi2 <- read_tsv("00dpi_2_barcodes.tsv", col_names = FALSE)$X1cell02dpi1 <- read_tsv("02dpi_1_barcodes.tsv", col_names = FALSE)$X1cell02dpi2 <- read_tsv("02dpi_2_barcodes.tsv", col_names = FALSE)$X1
# 合并为两个集合set00dpi <- unique(c(cell00dpi1, cell00dpi2))set02dpi <- unique(c(cell02dpi1, cell02dpi2))
#------------- Step 2. 清洗 cellinfo$X(去掉第二个 "-" 及后内容)-------------# 原格式如:AAACCCACAACAGCCC-1-0cellinfo$X_clean <- sub("^([^-]+-[^-]+).*", "\\1", cellinfo$X)
#------------- Step 3. 判断属于哪一类并加后缀 -------------cellinfo$X_final <- case_when( cellinfo$X_clean %in% set00dpi ~ paste0(cellinfo$X_clean, "-0"), cellinfo$X_clean %in% set02dpi ~ paste0(cellinfo$X_clean, "-1"), TRUE ~ cellinfo$X_clean # 不在任何列表就保持原样(一般不会出现))
#------------- Step 4. 处理重复并设置行名 -------------cellinfo$X_final_unique <- make.unique(cellinfo$X_final, sep = "_")rownames(cellinfo) <- cellinfo$X_final_uniquecellinfo <- cellinfo %>% select(-X, -X_clean, -X_final, -X_final_unique)
# 为稀疏矩阵添加行名(基因)和列名(细胞)rownames(counts) <- rownames(geneinfo)colnames(counts) <- rownames(cellinfo)
# 创建 Seurat 对象# 使用表达矩阵 counts 和细胞元数据 cellinfoAll.merge <- CreateSeuratObject(counts = counts, meta.data = cellinfo)
# 将 Seurat 对象保存为 pbmc(更常用的命名)pbmc <- All.merge
### 将对象拆分为不同时期处理,比较不同时期的表达差异# 假设 pbmc 是一个 Seurat 对象,提取元数据metadata <- pbmc@meta.data
# 提取细胞名称,并获取最后一位数字cell_names <- rownames(metadata)last_digits <- as.numeric(substr(cell_names, nchar(cell_names), nchar(cell_names)))
# 动态调整 cell_type 的值 0-8 9-17 18-26 27-35 36-44# 动态调整 cell_type 的值 0-9 10-19 20-29 30-39 40-49metadata$adjusted_celltype <- metadata$celltype + 10 * last_digits
# 将修改后的元数据更新到 Seurat 对象pbmc@meta.data <- metadata
# 计算不同聚类的平均表达量# group.by = "celltype" 按照列 celltype 分组# 通过AggregateExpression生成伪样本计数矩阵pseudobulk <- Seurat::AggregateExpression( pbmc, group.by = "adjusted_celltype", # 分组字段 slot = "counts", # 使用原始counts fun = "sum" # 求和聚合)count_matrix <- pseudobulk$RNA # 提取计数矩阵
# 从伪样本列名解析元数据(假设列名为"sample_celltype")sample_info <- data.frame( celltype = colnames(count_matrix), # 直接使用列名 row.names = colnames(count_matrix))
# 从GTF文件加载基因长度(需替换为实际数据)library(GenomicFeatures)txdb <- txdbmaker::makeTxDbFromGFF("{PROJ_LOTUS_HX}/00_assembly_result/GifuT2T/Lotus_GifuT2T.gtf")gene_lengths <- transcriptLengths(txdb, with.cds_len=TRUE) %>% group_by(gene_id) %>% summarise(length = max(cds_len)) # 取最长转录本长度
# 确保基因名称与计数矩阵一致rownames(gene_lengths) <- gene_lengths$gene_idgene_lengths <- gene_lengths[rownames(count_matrix), ]
library(DESeq2)
# 创建DESeqDataSetdds <- DESeqDataSetFromMatrix( countData = count_matrix, colData = sample_info, design = ~ celltype # 根据分析目标设定design)
# 添加基因长度信息mcols(dds)$basepairs <- gene_lengths$length
# 方法1:直接计算(未标准化)fpkm_matrix <- fpkm(dds, robust = TRUE) # robust=TRUE对长度做中位数校正
# 方法2:先运行标准化(推荐)dds <- estimateSizeFactors(dds)fpkm_matrix <- fpkm(dds, robust = TRUE)
# 保存为CSVwrite.csv(fpkm_matrix, "./02_processData/00_02_pseudobulk_fpkm.csv")06_4-DEGs_cell_type.R · R · 183 行
library(dplyr)library(Seurat)library(patchwork)library(Biobase)library(ggplot2)
#读取注释好的 h5ad 文件转换后的数据pbmc <- readRDS("./02_processData/zz_00_pea.rds")### 计算cluster的marker gene#Idents(pbmc) <- pbmc@meta.data$adjusted_celltype##所有cluster#all_markers <- FindAllMarkers(# object = pbmc#)#写出marker gene#write.csv(all_markers, file = "zz_07_celltype_difftime_markers_gene.csv", row.names = FALSE)
### 计算cluster的marker gene# 定义 cluster 组别cluster_groups <- list( Xylem = c(1, 11), Phloem = c(2, 12), Pericycle = c(3, 13), Vascular = c(4, 14), Cortex = c(5, 15), Meristem = c(6, 16), Early = c(7, 17))
# 设置细胞标识Idents(pbmc) <- pbmc@meta.data$adjusted_celltype
# 初始化结果列表marker_results <- list()
# 计算组内差异基因,以dpi01为背景group_names <- names(cluster_groups)for (i in 1:(length(group_names))) { group_name <- names(cluster_groups)[i] clusters <- cluster_groups[[group_name]] markers <- FindMarkers( pbmc, ident.1 = clusters[1], ident.2 = clusters[2] ) markers <- markers %>% mutate(Significance = case_when( p_val_adj < 0.05 & avg_log2FC > 1 ~ "Upregulated", p_val_adj < 0.05 & avg_log2FC < -1 ~ "Downregulated", TRUE ~ "Not Significant" ))
# 保存结果 comparison_name <- paste(group_name, "02dpivs00dpi", sep = "_") marker_results[[comparison_name]] <- markers
ggplot(markers, aes(x = avg_log2FC, y = -log10(p_val_adj), color = Significance)) + geom_point(alpha = 0.8) + scale_color_manual(values = c("Upregulated" = "red", "Downregulated" = "blue", "Not Significant" = "grey")) + theme_minimal() + labs( title = "Volcano Plot", x = "Log2 Fold Change", y = "-Log10 Adjusted P-Value" )
ggsave( filename = paste0("./03_figures/", comparison_name, "_volcano_plots.pdf"), width = 6, height = 5)
# 输出结果到文件 write.csv(markers, file = paste0("./02_processData/", comparison_name, "_markers.csv"))
# 打印提示信息 cat("已完成比较:", comparison_name, "\n")}
# 初始化结果列表marker_results <- list()
# 遍历所有组合group_names <- names(cluster_groups)for (i in 1:(length(group_names) - 1)) { for (j in (i + 1):length(group_names)) { # 获取当前组合的组名和 cluster 编号 group1 <- group_names[i] group2 <- group_names[j] clusters1 <- cluster_groups[[group1]] clusters2 <- cluster_groups[[group2]]
# 计算差异基因 markers <- FindMarkers( pbmc, ident.1 = clusters1, ident.2 = clusters2 ) markers <- markers %>% mutate(Significance = case_when( p_val_adj < 0.05 & avg_log2FC > 1 ~ "Upregulated", p_val_adj < 0.05 & avg_log2FC < -1 ~ "Downregulated", TRUE ~ "Not Significant" ))
# 保存结果 comparison_name <- paste(group1, "vs", group2, sep = "_") marker_results[[comparison_name]] <- markers
# 绘制火山图 ggplot(markers, aes(x = avg_log2FC, y = -log10(p_val_adj), color = Significance)) + geom_point(alpha = 0.8) + scale_color_manual(values = c("Upregulated" = "red", "Downregulated" = "blue", "Not Significant" = "grey")) + theme_minimal() + labs( title = "Volcano Plot", x = "Log2 Fold Change", y = "-Log10 Adjusted P-Value" )
ggsave( filename = paste0("./03_figures/", comparison_name, "_volcano_plots.pdf"), width = 6, height = 5)
# 输出结果到文件 write.csv(markers, file = paste0("./02_processData/", comparison_name, "_markers.csv"))
# 打印提示信息 cat("已完成比较:", comparison_name, "\n") }}
################################################## 额外分析:Early symbiotic signaling cells vs 非 Early symbiotic signaling cells###############################################
cat("开始 Early symbiotic signaling cells vs Others 分析...\n")
# 定义 Early symbiotic signaling cells 的 clustersEarly_clusters <- cluster_groups$Early # 即 c(7, 17)
# 所有 clustersall_clusters <- unique(unlist(cluster_groups))
# 除 Early symbiotic signaling cells 外的其他 clustersother_clusters <- setdiff(all_clusters, Early_clusters)
# 执行差异分析(Early symbiotic signaling cells 为 ident.1,上调表示 Early symbiotic signaling cells 相对 others 上调)markers_inf <- FindMarkers( pbmc, ident.1 = Early_clusters, ident.2 = other_clusters)
markers_inf <- markers_inf %>% mutate(Significance = case_when( p_val_adj < 0.05 & avg_log2FC > 1 ~ "Upregulated", p_val_adj < 0.05 & avg_log2FC < -1 ~ "Downregulated", TRUE ~ "Not Significant" ))
# 保存结果comparison_name <- "Early_symbiotic_signaling_cells_vs_Others"marker_results[[comparison_name]] <- markers_inf
# 绘制火山图p_inf <- ggplot(markers_inf, aes(x = avg_log2FC, y = -log10(p_val_adj), color = Significance)) + geom_point(alpha = 0.8) + scale_color_manual(values = c("Upregulated" = "red", "Downregulated" = "blue", "Not Significant" = "grey")) + theme_minimal() + labs( title = "Volcano Plot: Early symbiotic signaling cells vs Others", x = "Log2 Fold Change", y = "-Log10 Adjusted P-Value" )
ggsave( filename = paste0("./03_figures/", comparison_name, "_volcano_plots.pdf"), width = 6, height = 5)
# 输出到 CSVwrite.csv(markers_inf, file = paste0("./02_processData/", comparison_name, "_markers.csv"))
cat("已完成比较:", comparison_name, "\n")07_batch_multiple_scTenifoldKnk.R · R · 579 行
#!/usr/bin/env Rscript
suppressPackageStartupMessages({ library(data.table) library(dplyr) library(Seurat) library(tidyverse) library(patchwork) library(ggplot2) library(ggrepel) library(remotes) library(scTenifoldKnk) library(glue) library(gghalves)})
options(stringsAsFactors = FALSE)
# ============================================================# 0) 参数# ============================================================args <- commandArgs(trailingOnly = TRUE)if (length(args) < 2) stop("用法:Rscript 10.scTenifoldKnk_replicates.R gKO", call. = FALSE)
gKO <- args[1]celltype <- args[2]
# 运行次数N_RUN <- 100
# 子集数量K_SET <- c(10, 30, 50, 70, 100)
# hit阈值(用于次数统计)LOG2FC_CUT <- 1PADJ_CUT <- 0.05
# 细胞选择(与你当前脚本一致)N_CELLS_USE <- 2000
# HVG数N_HVG <- 1000
# ============================================================# 1) 输出目录# ============================================================dir.create("01_runs", showWarnings = FALSE)dir.create("02_summaries", showWarnings = FALSE)dir.create("03_plots", showWarnings = FALSE)
# ============================================================# 2) 数据准备:提取count矩阵 + SNF重命名# ============================================================obj5 <- readRDS("zz_00_pea.rds")SNF <- read.table("All_species_merged_by_GifuT2T.SNF_Symbol.tsv", header = TRUE)
SNF_fix <- SNF %>% mutate(GifuT2T_fix = gsub("_", "-", GifuT2T))
Idents(obj5) <- "celltype"message("celltype table:")print(table(obj5$celltype))
# 提取该细胞类型的特异表达基因deg_ct <- FindMarkers( object = obj5, ident.1 = celltype, ident.2 = NULL)
deg_ct_sig <- deg_ct %>% dplyr::filter( p_val_adj < 0.05, abs(avg_log2FC) > 1 )
genes_ct_all <- rownames(deg_ct_sig)
# 早期信号响应细胞(与你当前一致:idents="7")sc.s <- subset(obj5, idents = celltype)
count <- GetAssayData(sc.s, slot = "counts")
# HVGshvg <- VariableFeatures(sc.s)if (length(hvg) < N_HVG) { warning(glue("HVG数量不足({length(hvg)}), 将使用全部HVG。")) hvg_use <- hvg} else { hvg_use <- hvg[1:N_HVG]}
genes_use <- unique(c(hvg_use, gKO, genes_ct_all,SNF_fix$GifuT2T_fix))cells_use <- colnames(count)[1:min(N_CELLS_USE, ncol(count))]
count_use <- count[genes_use, cells_use]
# --- 将SNF的基因号替换为GeneSymbol(处理同名重复加后缀) ---genes_count <- rownames(count_use)
snf_map <- SNF_fix %>% filter(GifuT2T_fix %in% genes_count) %>% select(GeneSymbol, GifuT2T_fix) %>% group_by(GeneSymbol) %>% mutate( index = row_number(), NewGene = if (n() > 1) paste0(GeneSymbol, "-", index) else GeneSymbol ) %>% ungroup()
rename_table <- data.frame(OldGene = genes_count, stringsAsFactors = FALSE) %>% left_join(snf_map %>% select(GifuT2T_fix, NewGene), by = c("OldGene" = "GifuT2T_fix")) %>% mutate(FinalGene = ifelse(is.na(NewGene), OldGene, NewGene))
count_use_renamed <- count_userownames(count_use_renamed) <- rename_table$FinalGene
write.table( rename_table, file = "02_summaries/SNF_gene_rename_map.tsv", sep = "\t", quote = FALSE, row.names = FALSE)
# 虚拟敲除的基因名称要和矩阵保持一致if (gKO %in% rename_table$OldGene){ gKO_raw <- gKO gKO <- rename_table[ rename_table$OldGene %in% gKO_raw, "FinalGene"]}
# ============================================================# 3) scTenifoldKnk重复运行:100次并保留结果# ============================================================run_one <- function(seed_i, count_mat, gKO) { set.seed(seed_i)
res <- scTenifoldKnk( countMatrix = count_mat, qc = TRUE, gKO = gKO, qc_mtThreshold = 0.1, qc_minLSize = 1000, nc_lambda = 0, nc_nNet = 10, nc_nCells = min(500, ncol(count_mat)), nc_nComp = 3, nc_scaleScores = TRUE, nc_symmetric = FALSE, nc_q = 0.9, td_K = 3, td_maxIter = 1000, td_maxError = 1e-5, td_nDecimal = 3, ma_nDim = 2, nCores = parallel::detectCores() )
df <- res[["diffRegulation"]] %>% as.data.frame() # 统一列名:确保 gene / FC / p.adj 存在 if (!("gene" %in% colnames(df))) stop("diffRegulation中缺少 gene 列") if (!("FC" %in% colnames(df))) stop("diffRegulation中缺少 FC 列") if (!("p.adj" %in% colnames(df))) stop("diffRegulation中缺少 p.adj 列")
df <- df %>% mutate( log2FC = log2(FC), logp.adj = -log10(pmax(p.adj, 1e-300)) ) %>% select(gene, distance, Z, FC, p.value, p.adj, log2FC, logp.adj, everything())
df}
# 主循环(带失败保护)all_runs <- vector("list", N_RUN)run_status <- tibble( run = 1:N_RUN, seed = 1000 + 1:N_RUN, ok = FALSE, msg = NA_character_)
message(glue("Start {N_RUN} runs ..."))for (i in 1:N_RUN) { seed_i <- 1000 + i message(glue("[Run {i}/{N_RUN}] seed={seed_i}"))
out_rds <- file.path("01_runs", glue("run_{sprintf('%03d', i)}.rds")) out_csv <- file.path("01_runs", glue("run_{sprintf('%03d', i)}.csv"))
df_i <- tryCatch( { df <- run_one(seed_i, count_use_renamed, gKO)
saveRDS(df, out_rds) write.csv(df, out_csv, row.names = FALSE)
run_status$ok[i] <- TRUE run_status$msg[i] <- "OK" df }, error = function(e) { run_status$ok[i] <- FALSE run_status$msg[i] <- as.character(e$message) warning(glue("[Run {i}] FAILED: {e$message}")) NULL } )
all_runs[[i]] <- df_i}
all_runs <- lapply( all_runs, function(df) { if (is.null(df)) return(NULL) df %>% dplyr::filter(gene != gKO) })
write.csv(run_status, "02_summaries/run_status.csv", row.names = FALSE)saveRDS(all_runs, "02_summaries/all_runs_list.rds")
ok_idx <- which(run_status$ok)if (length(ok_idx) < max(K_SET)) { warning(glue( "成功运行次数只有 {length(ok_idx)},不足以覆盖最大K={max(K_SET)}。将只对可用范围内的K做分析。" ))}# 按顺序取成功的前若干次all_runs_ok <- all_runs[ok_idx]N_OK <- length(all_runs_ok)
K_SET2 <- K_SET[K_SET <= N_OK]if (length(K_SET2) == 0) stop("没有足够的成功运行结果用于后续分析。")
# ============================================================# 4) (2) 相关性:各K下,K次结果两两FC Spearman相关 + 箱线图 + 均值拟合# ============================================================# 将每次结果转为 named vector: gene -> FCto_fc_vec <- function(df) { v <- df$FC names(v) <- df$gene v}
# 计算两次run的spearman相关(按gene对齐)spearman_fc <- function(v1, v2) { g <- intersect(names(v1), names(v2)) if (length(g) < 10) return(NA_real_) suppressWarnings(cor(v1[g], v2[g], method = "spearman", use = "pairwise.complete.obs"))}
cor_records <- list()mean_records <- list()
for (K in K_SET2) { runsK <- all_runs_ok[1:K] vecs <- lapply(runsK, to_fc_vec)
cors <- c() if (K >= 2) { for (i in 1:(K - 1)) { for (j in (i + 1):K) { cors <- c(cors, spearman_fc(vecs[[i]], vecs[[j]])) } } }
cor_df <- tibble( K = K, pair_id = seq_along(cors), rho = cors ) %>% filter(!is.na(rho))
cor_records[[as.character(K)]] <- cor_df
mean_records[[as.character(K)]] <- tibble( K = K, mean_rho = mean(cor_df$rho, na.rm = TRUE) )}
cor_all <- bind_rows(cor_records)mean_all <- bind_rows(mean_records)
write.csv(cor_all, "02_summaries/spearman_pairwise_all.csv", row.names = FALSE)write.csv(mean_all, "02_summaries/spearman_mean_byK.csv", row.names = FALSE)
cor_all <- cor_all %>% mutate(K = factor(K, levels = c(10, 30, 50, 70, 100)))
mean_all <- mean_all %>% mutate(K = factor(K, levels = c(10, 30, 50, 70, 100)))
mean_all <- mean_all %>% mutate(K_index = as.numeric(K))
## -----------------------------## 1. 数据准备(再次确认)## -----------------------------cor_all <- cor_all %>% mutate(K = factor(K, levels = c("10", "30", "50", "70", "100")))
mean_all <- mean_all %>% mutate( K = factor(K, levels = c("10", "30", "50", "70", "100")), K_index = as.numeric(K) # 仅用于趋势线定位 )
## -----------------------------## 2. 作图(CNS / PNAS 风格)## -----------------------------p_cor <- ggplot() +
## 箱线图(主体,黑色) geom_boxplot( data = cor_all, aes(x = K, y = rho), width = 0.55, outlier.size = 0.6, outlier.alpha = 0.25, linewidth = 0.4, color = "black", fill = NA ) +
## 均值点(小而克制) stat_summary( data = cor_all, aes(x = K, y = rho), fun = mean, geom = "point", size = 1.8, color = "black" ) +
## 均值趋势线(强调色,非回归) geom_line( data = mean_all, aes(x = K_index, y = mean_rho, group = 1), linewidth = 0.9, color = "#B2182B" # CNS 常用深红 ) +
## 均值锚点 geom_point( data = mean_all, aes(x = K_index, y = mean_rho), size = 2.2, color = "#B2182B" ) +
## X 轴:离散标签 scale_x_discrete( limits = c("10", "30", "50", "70", "100"), name = "Number of scTenifoldKnk runs (K)" ) +
## Y 轴 scale_y_continuous( name = "Spearman correlation of FC (pairwise)", expand = expansion(mult = c(0.05, 0.1)) ) +
## 主题(PNAS / CNS 通用) theme_classic(base_size = 11) + theme( axis.line = element_line(linewidth = 0.4), axis.ticks = element_line(linewidth = 0.4), axis.text = element_text(color = "black"), axis.title = element_text(color = "black"), plot.title = element_text( hjust = 0.5, face = "plain", size = 11 ) ) +
labs( title = glue("Reproducibility of inferred FC across repeated scTenifoldKnk runs (gKO = {gKO})") )
ggsave("03_plots/01_FC_spearman_boxplot_with_mean_fit.pdf", p_cor, width = 8, height = 5)
# ============================================================# 5) (3) 各K下:每个基因 log2FC>1 且 p.adj<0.05 的次数# ============================================================# 以及 p.adj<0.05 的次数(给第4步上色用)count_hits_for_K <- function(runs_list, K, log2fc_cut, padj_cut) { runsK <- runs_list[1:K]
# gene全集(通常一致,但稳妥起见取union) gene_all <- sort(unique(unlist(lapply(runsK, function(df) df$gene))))
hit_counts <- setNames(rep(0L, length(gene_all)), gene_all) sigp_counts <- setNames(rep(0L, length(gene_all)), gene_all)
for (df in runsK) { df2 <- df %>% select(gene, log2FC, p.adj)
hit_gene <- df2 %>% filter(log2FC < log2fc_cut, p.adj < padj_cut) %>% pull(gene) sig_gene <- df2 %>% filter(p.adj < padj_cut) %>% pull(gene)
hit_counts[hit_gene] <- hit_counts[hit_gene] + 1L sigp_counts[sig_gene] <- sigp_counts[sig_gene] + 1L }
tibble( gene = gene_all, K = K, hit_n = as.integer(hit_counts[gene_all]), sigp_n = as.integer(sigp_counts[gene_all]) )}
hit_tables <- lapply(K_SET2, function(K) { count_hits_for_K(all_runs_ok, K, LOG2FC_CUT, PADJ_CUT)})hit_all <- bind_rows(hit_tables)
write.csv(hit_all, "02_summaries/gene_hit_counts_byK.csv", row.names = FALSE)
# 同时输出宽表(便于查看)hit_wide <- hit_all %>% select(gene, K, hit_n) %>% pivot_wider(names_from = K, values_from = hit_n, names_prefix = "K_")
sigp_wide <- hit_all %>% select(gene, K, sigp_n) %>% pivot_wider(names_from = K, values_from = sigp_n, names_prefix = "K_")
write.csv(hit_wide, "02_summaries/gene_hit_counts_byK_wide.csv", row.names = FALSE)write.csv(sigp_wide, "02_summaries/gene_sigp_counts_byK_wide.csv", row.names = FALSE)
# ============================================================# 6) (4) 各K下:Top20基因云雨图(雨=散点,云=小提琴)# - 横轴:FC# - 纵轴:gene# - 基因排序:按FC均值排序# - 颜色:由“p.adj<0.05的次数(sigp_n)”控制(渐变)# ============================================================get_fc_long_for_K <- function(runs_list, K) { runsK <- runs_list[1:K] fc_long <- bind_rows(lapply(seq_along(runsK), function(i) { df <- runsK[[i]] df %>% transmute( run = i, gene = gene, FC = FC, log2FC = log2FC, p.adj = p.adj ) })) fc_long}
make_raincloud_plot <- function(K, fc_long, hit_tbl_K, out_pdf) { # Top20:按 hit_n 降序 top_genes <- hit_tbl_K %>% arrange(desc(hit_n), desc(sigp_n)) %>% slice_head(n = 20) %>% pull(gene)
dat <- fc_long %>% filter(gene %in% top_genes)
# gene排序:按FC均值(对这K次) gene_order <- dat %>% group_by(gene) %>% summarise(mean_FC = mean(FC, na.rm = TRUE), .groups = "drop") %>% arrange(mean_FC) %>% pull(gene)
# 合并颜色信息:sigp_n(p.adj<0.05次数) sigp_map <- hit_tbl_K %>% select(gene, sigp_n, hit_n)
dat2 <- dat %>% left_join(sigp_map, by = "gene") %>% mutate(gene = factor(gene, levels = gene_order))
# 云雨图:violin + boxplot + jitter p <- ggplot(dat2, aes(x = gene, y = log2FC)) +
## 上半部小提琴(云) geom_half_violin( aes(fill = sigp_n), side = "r", trim = TRUE, width = 0.9, alpha = 0.85, color = NA ) +
## 箱线(雨的“核心”) geom_boxplot( width = 0.12, outlier.shape = NA, alpha = 0.9, linewidth = 0.35 ) +
## 散点(雨) geom_jitter( aes(color = sigp_n), width = 0.08, height = 0, size = 1.2, alpha = 0.6 ) +
coord_flip() +
## CNS 风格连续渐变(小提琴 & 点共用) scale_fill_gradient( low = "grey85", high = "#B2182B" # Nature / CNS 常用暗红 ) + scale_color_gradient( low = "grey60", high = "#B2182B" ) +
labs( title = glue( "Top 20 genes ranked by hit frequency (log2FC > {LOG2FC_CUT}, adj. P < {PADJ_CUT})\n", "K = {K}, gKO = {gKO}" ), x = "Gene (ordered by mean log2FC)", y = "log2(FC)", fill = "Count (adj. P < 0.05)", color = "Count (adj. P < 0.05)" ) +
theme_classic(base_size = 11) + theme( axis.line = element_line(linewidth = 0.4), axis.ticks = element_line(linewidth = 0.4), axis.text.y = element_text(size = 8), axis.text.x = element_text(size = 9), legend.position = "right", plot.title = element_text( hjust = 0, size = 11, face = "plain" ) )
ggsave(out_pdf, p, width = 8.5, height = 6)
p}
for (K in K_SET2) { message(glue("Make raincloud plot for K={K}"))
fc_long_K <- get_fc_long_for_K(all_runs_ok, K)
hit_tbl_K <- hit_all %>% filter(K == !!K)
out_pdf <- file.path("03_plots", glue("02_raincloud_Top20_byHit_K{K}.pdf")) make_raincloud_plot(K, fc_long_K, hit_tbl_K, out_pdf)}
# ============================================================# 7) 额外输出:每个K下 Top20基因列表(方便检查)# ============================================================top20_list <- bind_rows(lapply(K_SET2, function(K) { (hit_all %>% filter(K == !!K) %>% arrange(desc(hit_n), desc(sigp_n)) %>% slice_head(n = 20) %>% mutate(rank = row_number()) %>% select(K, rank, gene, hit_n, sigp_n))}))
write.csv(top20_list, "02_summaries/top20_genes_byHit_eachK.csv", row.names = FALSE)
message("All done.")makesure_02dpi_samples_all_have_Early_signal_cell.py · Python · 141 行
#!/usr/bin/env python3
import anndataimport pandas as pdimport matplotlib.pyplot as plt
# ===============================# 1. 读取 h5ad# ===============================adata = anndata.read_h5ad("./02_processData/data_clusterd_all_gene_4000.h5ad")adata.obs_names_make_unique()
# ===============================# 2. 注释 celltype# ===============================cluster2annotation = { "0": "5", "1": "5", "2": "3", "3": "4", "4": "4", "5": "5", "6": "6", "7": "2", "8": "7", "9": "1"}
adata.obs["celltype"] = ( adata.obs["leiden_0.4"] .astype(str) .map(cluster2annotation) .astype("category"))
# ===============================# 3. 读取 barcode# ===============================cell02dpi1 = pd.read_csv( "02dpi_1_barcodes.tsv", header=None, sep="\t")[0].astype(str).tolist()
cell02dpi2 = pd.read_csv( "02dpi_2_barcodes.tsv", header=None, sep="\t")[0].astype(str).tolist()
# ===============================# 4. 处理 obs_names# adata.obs_names: AAAC...-1-0# barcode 文件: AAAC...-1# 去掉最后的 -0 / -1 / -2 ...# ===============================obs = adata.obs.copy()obs["cell_id"] = obs.indexobs["barcode"] = obs["cell_id"].str.replace(r"-[0-9]+$", "", regex=True)
obs["group"] = "other"obs.loc[obs["barcode"].isin(cell02dpi1), "group"] = "cell02dpi1"obs.loc[obs["barcode"].isin(cell02dpi2), "group"] = "cell02dpi2"
# ===============================# 5. 加入 UMAP 坐标# ===============================obs["UMAP_1"] = adata.obsm["X_umap"][:, 0]obs["UMAP_2"] = adata.obsm["X_umap"][:, 1]
# ===============================# 6. 统计 celltype == 7 数量# ===============================n_02dpi1_total = (obs["group"] == "cell02dpi1").sum()n_02dpi2_total = (obs["group"] == "cell02dpi2").sum()
n_02dpi1_ct7 = ((obs["group"] == "cell02dpi1") & (obs["celltype"].astype(str) == "7")).sum()n_02dpi2_ct7 = ((obs["group"] == "cell02dpi2") & (obs["celltype"].astype(str) == "7")).sum()
print(f"cell02dpi1 total cells: {n_02dpi1_total}")print(f"cell02dpi1 celltype==7 cells: {n_02dpi1_ct7}")
print(f"cell02dpi2 total cells: {n_02dpi2_total}")print(f"cell02dpi2 celltype==7 cells: {n_02dpi2_ct7}")
# ===============================# 7. 绘图函数# ===============================def plot_highlight_umap( obs, group_name, outfile_prefix): target = (obs["group"] == group_name) & (obs["celltype"].astype(str) == "7") other = ~target
plt.figure(figsize=(5, 4.5))
plt.scatter( obs.loc[other, "UMAP_1"], obs.loc[other, "UMAP_2"], s=2, c="lightgrey", alpha=0.5, linewidths=0 )
plt.scatter( obs.loc[target, "UMAP_1"], obs.loc[target, "UMAP_2"], s=6, c="#D7B76A", alpha=0.9, linewidths=0 )
plt.title(f"{group_name}: celltype 7", fontsize=14, fontweight="bold") plt.xlabel("UMAP 1") plt.ylabel("UMAP 2") plt.tight_layout()
plt.savefig(f"{outfile_prefix}.pdf") plt.savefig(f"{outfile_prefix}.png", dpi=300) plt.close()
# ===============================# 8. 分别绘制两个 UMAP# ===============================plot_highlight_umap( obs, group_name="cell02dpi1", outfile_prefix="cell02dpi1_celltype7_umap")
plot_highlight_umap( obs, group_name="cell02dpi2", outfile_prefix="cell02dpi2_celltype7_umap")
print("Done.")