MAAS

Overview

Single-cell multimodal analysis enables highly accurate delineation of clinically relevant tumor cell subpopulations

Quick start

To use MAAS functions for integration, please run the following commands in R. We provide example data of cell similarities for the three modalities formatted as list R object.

library(Rcpp)
for (i in list.files("./src/", pattern="MAAS*")) {
  sourceCpp(paste0("./src/", i))
}

data <- readRDS("example.data/example.similarity.rds")
maas.test <- MAAS(data$Peak, df$CNV, df$SNV, dims = 2:5)
saveRDS(maas.test, "maas.res.all.rds")

Then we can do clustering based on the consensus latent factors

#### Determine the optimal clustering strategy
maas.res <- readRDS("maas.res.test.rds")
barcode.list <- rownames(data$Peak)
set.seed(1)
barcode.list <- sample(barcode.list, 400)

#### Using the optimal clustering strategy
clusPerformance <- data.frame(matrix(nrow = length(maas.res)-1, ncol = 5),
                              row.names = paste0("dims=", 2:length(maas.res)))
colnames(clusPerformance) <- paste0("k=", 2:6)
for(i in 1:(length(maas.res)-1)){
  for(j in 2:6){
    df <- as.data.frame(maas.res[[i]]$W)
    rownames(df) <- barcode.list
    maas.tmp.clu <- withr::with_seed(2, kmeans(df, centers = j)$cluster)
    clusPerformance[i,j-1] <- clusteringMetric(maas.res[[i]]$W, clu = maas.tmp.clu, disMethod = "cosine")
  }
}

#### Re-running clustering with the optimal performance
df <- as.data.frame(maas.res[[4]]$W)
rownames(df) <- barcode.list
maas.clu <- data.frame(Cluster = withr::with_seed(2, kmeans(df, centers = 2)$cluster))
maas.clu$Cluster <- as.factor(maas.clu$Cluster)
saveRDS(ataclone.clu, "example.MAAS.clu.rds")

Visualization using UMAP plot

umap.axis <- withr::with_seed(2, uwot::umap(df, n_neighbors = 50, metric = "manhattan", min_dist = 0.1, n_threads = 30))
umap.axis <- as.data.frame(umap.axis); umap.axis$Cluster <- ataclone.clu$Cluster
colnames(umap.axis) <- c("UMAP-1", "UMAP-2", "Cluster")
ggplot(umap.axis, aes(`UMAP-1`, `UMAP-2`))+
  geom_point(aes(color = Cluster), size = 1.75)+
  scale_color_manual(values = my_color)+
  theme_dr()+
  labs(x = "UMAP-1", y = "UMAP-2")+
  theme(panel.grid = element_blank(),
        axis.title = element_text(size = 14),
        axis.text = element_blank(),
        legend.text = element_text(size = 12))

Documentation and tutorials (full data preparation and integration) can be found at https://larrycpan.github.io/MAAS/. The example data (cell similarity) has been uploaded to the folder example.data, which can be used for the MAAS function directly.

We provide the source code of SNV calling and MAAS in the github repository, including peak adjusting and multimodal integration. To perform SNV RPCA analysis, you should install Matlab in advance, which always finishes in a few seconds. This step is optional for users that raw SNV matrix is also adopted by MAAS.

In addition, we highly recommend installing openBLAS to speed matrix operations, before starting your MAAS analysis.

Getting help

If you encounter a bug or have a feature request, please open an Issues.

If you would like to discuss questions related to single-cell analysis, you can open a Discussions.

Related packages

• Signac
• epiAneufinder
• SComatic