Clustering CMC D1 data

Here we use two strategies to clustering:
1. Seurat: Detect communitis in Shared Nearest Neighbors network;
2. SIMLR: Single-cell Interpretation via Multi-kernel LeaRning.
SIMLR is often much precise but slower than Seurat.

Look through the data

Violin plot below shows:
- nGene: Number of genes detected (UMI>0) in each cell.
- nUMI: Number of Unique Molecular Identity identified in each cell.
- percent.mito: Percentage of mitochondrial genes UMIs in all total UMIs.
Note that large number of cells express high level of mitochondrial genes. Is it a reasonable phenomenon for cardiac muscle cells?

Seurat clustering

The cells were clustering using Seurat,2016. The top 8 PCs were used as imput. No mitochondrion genes were detected as HVG.

Determine statistically significant principal components. We select top 8 pcs for clustering.

tSNE and UMAP are two non-linear dimention reduction algorithm. The clustering results of Seurat are showed. (Resolution=0.2)

Cardica myoblast markers and hemoglobin

Hematopoietic markers

Find markers in each group

Top 10 markers of each cluster.

Differentially expression stats.

Enrichment

Reactome pathway

2. SIMLR clustering

The cells were clustering using SIMLR,2017. The top 20 PCs were used as imput.

2-D embedding.

Find markers in each group

Heatmap of the top 10 markers ranked by logFC using bimod test.

Differentially expression stats.