Hamony

官网: https://github.com/immunogenomics/harmony

文章: https://www.nature.com/articles/s41592-019-0619-0

学习博文：

https://www.jianshu.com/p/7c43dc99c4b1
https://www.jianshu.com/p/d1cb5a7f19af

require(Seurat)
require(ggplot2)
require(dplyr)
require(harmony)
require(cowplot)

#----------------------------------------------------------------------------------------------#
object_used_1 <- Dmel_ovary_CD_filtered_by_3_200_feature_over_600_mito_less_20_kept_mito_gene_HVF_1100_dim_6_doublet_removed
object_used_1$orig.ident <- "CD"

object_used_2 <- Dmel_ovary_HSD_filtered_by_3_200_feature_over_600_mito_less_20_kept_mito_gene_HVF_1100_dim_6_doublet_removed
object_used_2$orig.ident <- "HSD"

#----------------------------------------------------------------------------------------------#
#Merge the two seurat objects to one
object_merged <- merge(object_used_1, object_used_2)

#----------------------------------------------------------------------------------------------#
#Normalize the merged seurat object
object_merged <- NormalizeData(object_merged, normalization.method = "LogNormalize", scale.factor = 10000)

#Call the high variable features of the merged seurat object based on the normalized UMI matrix
object_merged <- FindVariableFeatures(object_merged, selection.method = "vst", nfeatures = 2000)

#Scaled the normalized UMI matrix
genes_all <- rownames(object_merged)
object_merged <- ScaleData(object_merged, features = genes_all)

#PCA based on the scaled normalized UMI matrix
object_merged <- RunPCA(object_merged, features = VariableFeatures(object = object_merged))

#----------------------------------------------------------------------------------------------#
#Integrate two sample by RunHarmony() based on the PCA results of merged Seurat object; dims.use, the PC used; max.iter.harmony, the iteration number of RunHarmony()
object_harmonied <- RunHarmony(object_merged, group.by.vars = "orig.ident", plot_convergence = TRUE)

#----------------------------------------------------------------------------------------------#
#Downstream analysis of integrated seurat object
object_harmonied <- RunTSNE(object_harmonied, reduction = "harmony", dims = 1:20)
object_harmonied <- RunUMAP(object_harmonied, reduction = "harmony", dims = 1:20)

object_harmonied <- FindNeighbors(object_harmonied, reduction = "harmony", dims = 1:20)
object_harmonied <- FindClusters(object_harmonied, resolution = 0.7)

1. 说明一

• 1. RunHarmony() 进行的数据整合，是依赖merge 但未整合数据的 PC 进行的
• 2. RunHarmony() 并不会改变未整合前的任何数据，而是产生一个新的降维数据 harmony；此数据放在 @reductions$harmony@feature.loadings 这个 slot 中
• 3. 因为RunHarmony() 并未改变未整合前的任何数据，所以用 Harmony 进行数据整合后并不会改变 差异分析 等分析结果

2. 说明二

@reductions$harmony@feature.loadings 中数据解读:

• 1. @reductions$harmony@cell.embeddings: 各个细胞在 harmony 这个降维方式中的坐标
• 2. @reductions$harmony@feature.loadings: 各个 feature(gene) 在 harmony 中的 embedding，数值大小代表 feature 对这个 harmony 维度的贡献程度；数值越大贡献越大
• 3. 数据整合后，后续的 降维 和 聚类 分析都在 @reductions$harmony@cell.embeddings 的基础上进行