# HCA human bone marrow (10X Genomics) 

<script>
document.addEventListener("click", function (event) {
    if (event.target.classList.contains("rebook-collapse")) {
        event.target.classList.toggle("active");
        var content = event.target.nextElementSibling;
        if (content.style.display === "block") {
            content.style.display = "none";
        } else {
            content.style.display = "block";
        }
    }
})
</script>

<style>
.rebook-collapse {
  background-color: #eee;
  color: #444;
  cursor: pointer;
  padding: 18px;
  width: 100%;
  border: none;
  text-align: left;
  outline: none;
  font-size: 15px;
}

.rebook-content {
  padding: 0 18px;
  display: none;
  overflow: hidden;
  background-color: #f1f1f1;
}
</style>

## Introduction

Here, we use an example dataset from the [Human Cell Atlas immune cell profiling project on bone marrow](https://preview.data.humancellatlas.org), which contains scRNA-seq data for 380,000 cells generated using the 10X Genomics technology.
This is a fairly big dataset that represents a good use case for the techniques in [Advanced Chapter 14](http://bioconductor.org/books/3.18/OSCA.advanced/dealing-with-big-data.html#dealing-with-big-data).

## Data loading

This dataset is loaded via the *[HCAData](https://bioconductor.org/packages/3.18/HCAData)* package, which provides a ready-to-use `SingleCellExperiment` object. 


```r
library(HCAData)
sce.bone <- HCAData('ica_bone_marrow', as.sparse=TRUE)
sce.bone$Donor <- sub("_.*", "", sce.bone$Barcode)
```

We use symbols in place of IDs for easier interpretation later.


```r
library(EnsDb.Hsapiens.v86)
rowData(sce.bone)$Chr <- mapIds(EnsDb.Hsapiens.v86, keys=rownames(sce.bone),
    column="SEQNAME", keytype="GENEID")

library(scater)
rownames(sce.bone) <- uniquifyFeatureNames(rowData(sce.bone)$ID,
    names = rowData(sce.bone)$Symbol)
```

## Quality control

Cell calling was not performed (see [here](https://s3.amazonaws.com/preview-ica-expression-data/Brief+ICA+Read+Me.pdf)) so we will perform QC using all metrics and block on the donor of origin during outlier detection.
We perform the calculation across multiple cores to speed things up.


```r
library(BiocParallel)
bpp <- MulticoreParam(8)
sce.bone <- unfiltered <- addPerCellQC(sce.bone, BPPARAM=bpp,
    subsets=list(Mito=which(rowData(sce.bone)$Chr=="MT")))

qc <- quickPerCellQC(colData(sce.bone), batch=sce.bone$Donor,
    sub.fields="subsets_Mito_percent")
sce.bone <- sce.bone[,!qc$discard]
```


```r
unfiltered$discard <- qc$discard

gridExtra::grid.arrange(
    plotColData(unfiltered, x="Donor", y="sum", colour_by="discard") +
        scale_y_log10() + ggtitle("Total count"),
    plotColData(unfiltered, x="Donor", y="detected", colour_by="discard") +
        scale_y_log10() + ggtitle("Detected features"),
    plotColData(unfiltered, x="Donor", y="subsets_Mito_percent",
        colour_by="discard") + ggtitle("Mito percent"),
    ncol=2
)
```

<div class="figure">
<img src="hca-bone-marrow_files/figure-html/unref-hca-bone-qc-1.png" alt="Distribution of QC metrics in the HCA bone marrow dataset. Each point represents a cell and is colored according to whether it was discarded." width="672" />
<p class="caption">(\#fig:unref-hca-bone-qc)Distribution of QC metrics in the HCA bone marrow dataset. Each point represents a cell and is colored according to whether it was discarded.</p>
</div>


```r
plotColData(unfiltered, x="sum", y="subsets_Mito_percent", 
    colour_by="discard") + scale_x_log10()
```

<div class="figure">
<img src="hca-bone-marrow_files/figure-html/unref-hca-bone-mito-1.png" alt="Percentage of mitochondrial reads in each cell in the HCA bone marrow dataset compared to its total count. Each point represents a cell and is colored according to whether that cell was discarded." width="672" />
<p class="caption">(\#fig:unref-hca-bone-mito)Percentage of mitochondrial reads in each cell in the HCA bone marrow dataset compared to its total count. Each point represents a cell and is colored according to whether that cell was discarded.</p>
</div>

## Normalization

For a minor speed-up, we use already-computed library sizes rather than re-computing them from the column sums.


```r
sce.bone <- logNormCounts(sce.bone, size_factors = sce.bone$sum)
```


```r
summary(sizeFactors(sce.bone))
```

```
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##    0.05    0.47    0.65    1.00    0.89   42.38
```

## Variance modeling

We block on the donor of origin to mitigate batch effects during HVG selection.
We select a larger number of HVGs to capture any batch-specific variation that might be present.


```r
library(scran)
set.seed(1010010101)
dec.bone <- modelGeneVarByPoisson(sce.bone, 
    block=sce.bone$Donor, BPPARAM=bpp)
top.bone <- getTopHVGs(dec.bone, n=5000)
```


```r
par(mfrow=c(4,2))
blocked.stats <- dec.bone$per.block
for (i in colnames(blocked.stats)) {
    current <- blocked.stats[[i]]
    plot(current$mean, current$total, main=i, pch=16, cex=0.5,
        xlab="Mean of log-expression", ylab="Variance of log-expression")
    curfit <- metadata(current)
    curve(curfit$trend(x), col='dodgerblue', add=TRUE, lwd=2)
}
```

<div class="figure">
<img src="hca-bone-marrow_files/figure-html/unref-hca-bone-var-1.png" alt="Per-gene variance as a function of the mean for the log-expression values in the HCA bone marrow dataset. Each point represents a gene (black) with the mean-variance trend (blue) fitted to the variances." width="576" />
<p class="caption">(\#fig:unref-hca-bone-var)Per-gene variance as a function of the mean for the log-expression values in the HCA bone marrow dataset. Each point represents a gene (black) with the mean-variance trend (blue) fitted to the variances.</p>
</div>

## Data integration

Here we use multiple cores, randomized SVD and approximate nearest-neighbor detection to speed up this step.


```r
library(batchelor)
library(BiocNeighbors)

set.seed(1010001)
merged.bone <- fastMNN(sce.bone, batch = sce.bone$Donor, subset.row = top.bone,
     BSPARAM=BiocSingular::RandomParam(deferred = TRUE), 
     BNPARAM=AnnoyParam(),
     BPPARAM=bpp)

reducedDim(sce.bone, 'MNN') <- reducedDim(merged.bone, 'corrected')
```

We use the percentage of variance lost as a diagnostic measure:


```r
metadata(merged.bone)$merge.info$lost.var
```

```
##      MantonBM1 MantonBM2 MantonBM3 MantonBM4 MantonBM5 MantonBM6 MantonBM7
## [1,]  0.006922  0.006392  0.000000  0.000000  0.000000  0.000000  0.000000
## [2,]  0.006380  0.006863  0.023049  0.000000  0.000000  0.000000  0.000000
## [3,]  0.005068  0.003084  0.005178  0.019496  0.000000  0.000000  0.000000
## [4,]  0.002009  0.001891  0.001901  0.001786  0.023105  0.000000  0.000000
## [5,]  0.002452  0.002003  0.001770  0.002926  0.002646  0.023852  0.000000
## [6,]  0.003167  0.003222  0.003169  0.002636  0.003362  0.003442  0.024650
## [7,]  0.001968  0.001701  0.002441  0.002045  0.001585  0.002312  0.002003
##      MantonBM8
## [1,]   0.00000
## [2,]   0.00000
## [3,]   0.00000
## [4,]   0.00000
## [5,]   0.00000
## [6,]   0.00000
## [7,]   0.03216
```

## Dimensionality reduction

We set `external_neighbors=TRUE` to replace the internal nearest neighbor search in the UMAP implementation with our parallelized approximate search.
We also set the number of threads to be used in the UMAP iterations.


```r
set.seed(01010100)
sce.bone <- runUMAP(sce.bone, dimred="MNN",
    external_neighbors=TRUE, 
    BNPARAM=AnnoyParam(),
    BPPARAM=bpp,
    n_threads=bpnworkers(bpp))
```

## Clustering

Graph-based clustering generates an excessively large intermediate graph so we will instead use a two-step approach with $k$-means.
We generate 1000 small clusters that are subsequently aggregated into more interpretable groups with a graph-based method.
If more resolution is required, we can increase `centers` in addition to using a lower `k` during graph construction.


```r
library(bluster)

set.seed(1000)
colLabels(sce.bone) <- clusterRows(reducedDim(sce.bone, "MNN"),
    TwoStepParam(KmeansParam(centers=1000), NNGraphParam(k=5)))

table(colLabels(sce.bone))
```

```
## 
##     1     2     3     4     5     6     7     8     9    10    11    12    13 
## 20331 11161 55464 47426 15731 10581 64721 26493 18703 15043 17097  4992  3157 
##    14    15 
##  3403  2422
```

We observe mostly balanced contributions from different samples to each cluster (Figure \@ref(fig:unref-hca-bone-ab)), consistent with the expectation that all samples are replicates from different donors.


```r
tab <- table(Cluster=colLabels(sce.bone), Donor=sce.bone$Donor)
library(pheatmap)
pheatmap(log10(tab+10), color=viridis::viridis(100))
```

<div class="figure">
<img src="hca-bone-marrow_files/figure-html/unref-hca-bone-ab-1.png" alt="Heatmap of log~10~-number of cells in each cluster (row) from each sample (column)." width="672" />
<p class="caption">(\#fig:unref-hca-bone-ab)Heatmap of log~10~-number of cells in each cluster (row) from each sample (column).</p>
</div>




```r
# TODO: add scrambling option in scater's plotting functions.
scrambled <- sample(ncol(sce.bone))

gridExtra::grid.arrange(
    plotUMAP(sce.bone, colour_by="label", text_by="label"),
    plotUMAP(sce.bone[,scrambled], colour_by="Donor")
)
```

<div class="figure">
<img src="hca-bone-marrow_files/figure-html/unref-hca-bone-umap-1.png" alt="UMAP plots of the HCA bone marrow dataset after merging. Each point represents a cell and is colored according to the assigned cluster (top) or the donor of origin (bottom)." width="672" />
<p class="caption">(\#fig:unref-hca-bone-umap)UMAP plots of the HCA bone marrow dataset after merging. Each point represents a cell and is colored according to the assigned cluster (top) or the donor of origin (bottom).</p>
</div>

## Differential expression

We identify marker genes for each cluster while blocking on the donor.


```r
markers.bone <- findMarkers(sce.bone, block = sce.bone$Donor, 
    direction = 'up', lfc = 1, BPPARAM=bpp)
```

We visualize the top markers for a randomly chosen cluster using a "dot plot" in Figure \@ref(fig:unref-hca-bone-dotplot).
The presence of upregulated genes like _LYZ_, _S100A8_ and _VCAN_ is consistent with a monocyte identity for this cluster.




```r
top.markers <- markers.bone[["4"]]
best <- top.markers[top.markers$Top <= 10,]
lfcs <- getMarkerEffects(best)

library(pheatmap)
pheatmap(lfcs, breaks=seq(-5, 5, length.out=101))
```

<div class="figure">
<img src="hca-bone-marrow_files/figure-html/unref-hca-bone-dotplot-1.png" alt="Heatmap of log~2~-fold changes for the top marker genes (rows) of cluster 4 compared to all other clusters (columns)." width="672" />
<p class="caption">(\#fig:unref-hca-bone-dotplot)Heatmap of log~2~-fold changes for the top marker genes (rows) of cluster 4 compared to all other clusters (columns).</p>
</div>



## Cell type classification

We perform automated cell type classification using a reference dataset to annotate each cluster based on its pseudo-bulk profile. 
This is faster than the per-cell approaches described in Chapter \@ref(cell-type-annotation) at the cost of the resolution required to detect heterogeneity inside a cluster.
Nonetheless, it is often sufficient for a quick assignment of cluster identity, and indeed, cluster 4 is also identified as consisting of monocytes from this analysis.


```r
se.aggregated <- sumCountsAcrossCells(sce.bone, id=colLabels(sce.bone), BPPARAM=bpp)

library(celldex)
hpc <- HumanPrimaryCellAtlasData()

library(SingleR)
anno.single <- SingleR(se.aggregated, ref = hpc, labels = hpc$label.main,
    assay.type.test="sum")
anno.single
```

```
## DataFrame with 15 rows and 4 columns
##                             scores           labels  delta.next
##                           <matrix>      <character>   <numeric>
## 1   0.366050:0.741975:0.637800:...              GMP   0.3366255
## 2   0.399229:0.715314:0.628516:... Pro-B_cell_CD34+   0.0690916
## 3   0.325812:0.654869:0.570039:...          T_cells   0.1292127
## 4   0.296455:0.742466:0.529404:...         Monocyte   0.3099254
## 5   0.345773:0.565378:0.479722:...          T_cells   0.4943226
## ...                            ...              ...         ...
## 11  0.326882:0.646229:0.561060:...          T_cells 0.027835817
## 12  0.380710:0.684123:0.784540:...       BM & Prog. 0.000499754
## 13  0.368546:0.652935:0.580330:...           B_cell 0.201098545
## 14  0.294361:0.706019:0.527282:...         Monocyte 0.355212614
## 15  0.339786:0.687074:0.569933:...              GMP 0.131830602
##        pruned.labels
##          <character>
## 1                GMP
## 2   Pro-B_cell_CD34+
## 3            T_cells
## 4           Monocyte
## 5            T_cells
## ...              ...
## 11           T_cells
## 12        BM & Prog.
## 13                NA
## 14          Monocyte
## 15               GMP
```



<!-- ```{r saveRDS, include=FALSE} -->
<!-- ## TEMP FIX -->
<!-- saveRDS(sce.bone, 'sce.bone.rds') -->
<!-- saveRDS(markers.bone, 'markers.bone.rds') -->

<!-- sce.bone <- readRDS('sce.bone.rds') -->
<!-- markers.bone <- readRDS('markers.bone.rds') -->
<!-- ``` -->

<!--
## Gene set Enrichment Analysis

To interpret differential expression (DE) results, gene set enrichment analysis (GSEA) can be performed to annotate DE results. The first step is to acquire a mapping of ENTREZID to gene symbol.


```r
## Get mappings of ENTREZID to Symbol
library(org.Hs.eg.db)
keys_entrez <- keys(org.Hs.eg.db, 'ENTREZID')
mapping_es <- AnnotationDbi::select(org.Hs.eg.db,
                                    keys = keys_entrez,
                                    columns = c('ENTREZID', 'SYMBOL'),
                                    keytype = 'ENTREZID')
mapping_es$ENTREZID <- as.integer(mapping_es$ENTREZID)
```

The next is to parse the gene sets (here we use a "tidy" representation of MSigDB gene sets) and remap them.


```r
## Get pathways of interest - convert to list with symbol
## devtools::install_github('stephenturner/msigdf') # "tidy" MSigDB genesets
library(msigdf)
library(dplyr)
mdb <- dplyr::inner_join(msigdf.human, mapping_es,
                         by = c('entrez' = 'ENTREZID')) %>%
    dplyr::filter(collection == 'c7') %>%
    dplyr::select(-collection, -entrez) %>%
    dplyr::group_nest(geneset)
pathways <- purrr::map(mdb$data, function(x) {
    as.character(x$SYMBOL)
})
names(pathways) <- mdb$geneset
```

Finally, the DE results from the `findMarkers()` function are used to test for differential gene set enrichment using the `fgsea` package. Here we compare cluster 3 against cluster 15, which are at opposite ends of a UMAP group.


```r
## Get stats based on markers search
## Compare cluster 3 (Dendritic cells) and 15 (HSCs)
stats <- markers.bone[[3]]$logFC.15
names(stats) <- rownames(markers.bone[[3]])

## Run fast gene set enrichment analysis
## Returns a data.table object
library(fgsea)
gse <- fgsea(pathways = pathways, 
             stats = stats,
             minSize = 15,
             maxSize = 500,
             nperm = 5000)
```

The enrichment scores of the top up and down pathways can then be plotted as a table:


```r
## Get top up and down pathways based on NES
gse.up <- gse[order(gse$NES, decreasing = TRUE)[1:2], ]$pathway
gse.dn <- gse[order(gse$NES)[1:2], ]$pathway
topPathways <- c(gse.up, rev(gse.dn))

## Plot the GSEA table
plotGseaTable(pathways[topPathways], stats,
              gse, 
              gseaParam = 0.5)
```

Or select pathways can also be plotted as a more traditional running score plot:


```r
## Traditional GSEA plot
plotEnrichment(pathways[[topPathways[1]]],
               stats)
```

### Trajectory analysis

Slingshot relies on providing a set of clusters on which to construct a trajectory.


```r
library(slingshot)

## Clusters to attempt trajectory on:
slc.clusters <- c(15, 22, 11, 7, 1, 3)

## Subsample dataset for speed
slc.dat <- sce.bone[, sce.bone$cluster %in% slc.clusters]
slc.subdat <- slc.dat[, sample(ncol(slc.dat), 20000)]

## Run slingshot
slc <- slingshot(slc.subdat,
                 clusterLabels = 'cluster',
                 reducedDim = 'UMAP')
```

The calculated principal curve can then be visualized on the UMAP space.


```r
## Slingshot trajectory plot
library(RColorBrewer)
colors <- colorRampPalette(brewer.pal(11, 'Spectral')[-6])(100)
plotcol <- colors[cut(slc$slingPseudotime_1, breaks = 100)]
plot(reducedDims(slc)$UMAP, col = plotcol, pch=16, asp = 1)
lines(SlingshotDataSet(slc), lwd = 2, col = 'black')
```
-->

## Session Info {-}

<button class="rebook-collapse">View session info</button>
<div class="rebook-content">
```
R version 4.3.1 (2023-06-16)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 22.04.3 LTS

Matrix products: default
BLAS:   /home/biocbuild/bbs-3.18-bioc/R/lib/libRblas.so 
LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.10.0

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=en_GB              LC_COLLATE=C              
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

time zone: America/New_York
tzcode source: system (glibc)

attached base packages:
[1] stats4    stats     graphics  grDevices utils     datasets  methods  
[8] base     

other attached packages:
 [1] SingleR_2.4.0               celldex_1.11.1             
 [3] pheatmap_1.0.12             bluster_1.12.0             
 [5] BiocSingular_1.18.0         BiocNeighbors_1.20.0       
 [7] batchelor_1.18.0            scran_1.30.0               
 [9] DelayedArray_0.28.0         BiocParallel_1.36.0        
[11] scater_1.30.0               ggplot2_3.4.4              
[13] scuttle_1.12.0              EnsDb.Hsapiens.v86_2.99.0  
[15] ensembldb_2.26.0            AnnotationFilter_1.26.0    
[17] GenomicFeatures_1.54.0      AnnotationDbi_1.64.0       
[19] HDF5Array_1.30.0            rhdf5_2.46.0               
[21] HCAData_1.17.0              SingleCellExperiment_1.24.0
[23] SummarizedExperiment_1.32.0 Biobase_2.62.0             
[25] GenomicRanges_1.54.0        GenomeInfoDb_1.38.0        
[27] IRanges_2.36.0              S4Vectors_0.40.0           
[29] BiocGenerics_0.48.0         MatrixGenerics_1.14.0      
[31] matrixStats_1.0.0           BiocStyle_2.30.0           
[33] rebook_1.12.0              

loaded via a namespace (and not attached):
  [1] later_1.3.1                   BiocIO_1.12.0                
  [3] bitops_1.0-7                  filelock_1.0.2               
  [5] tibble_3.2.1                  CodeDepends_0.6.5            
  [7] graph_1.80.0                  XML_3.99-0.14                
  [9] lifecycle_1.0.3               edgeR_4.0.0                  
 [11] lattice_0.22-5                magrittr_2.0.3               
 [13] limma_3.58.0                  sass_0.4.7                   
 [15] rmarkdown_2.25                jquerylib_0.1.4              
 [17] yaml_2.3.7                    metapod_1.10.0               
 [19] httpuv_1.6.12                 RColorBrewer_1.1-3           
 [21] cowplot_1.1.1                 DBI_1.1.3                    
 [23] ResidualMatrix_1.12.0         abind_1.4-5                  
 [25] zlibbioc_1.48.0               purrr_1.0.2                  
 [27] RCurl_1.98-1.12               rappdirs_0.3.3               
 [29] GenomeInfoDbData_1.2.11       ggrepel_0.9.4                
 [31] irlba_2.3.5.1                 dqrng_0.3.1                  
 [33] DelayedMatrixStats_1.24.0     codetools_0.2-19             
 [35] xml2_1.3.5                    tidyselect_1.2.0             
 [37] farver_2.1.1                  ScaledMatrix_1.10.0          
 [39] viridis_0.6.4                 BiocFileCache_2.10.0         
 [41] GenomicAlignments_1.38.0      jsonlite_1.8.7               
 [43] ellipsis_0.3.2                tools_4.3.1                  
 [45] progress_1.2.2                Rcpp_1.0.11                  
 [47] glue_1.6.2                    gridExtra_2.3                
 [49] SparseArray_1.2.0             xfun_0.40                    
 [51] dplyr_1.1.3                   withr_2.5.1                  
 [53] BiocManager_1.30.22           fastmap_1.1.1                
 [55] rhdf5filters_1.14.0           fansi_1.0.5                  
 [57] digest_0.6.33                 rsvd_1.0.5                   
 [59] R6_2.5.1                      mime_0.12                    
 [61] colorspace_2.1-0              biomaRt_2.58.0               
 [63] RSQLite_2.3.1                 utf8_1.2.4                   
 [65] generics_0.1.3                rtracklayer_1.62.0           
 [67] prettyunits_1.2.0             httr_1.4.7                   
 [69] S4Arrays_1.2.0                uwot_0.1.16                  
 [71] pkgconfig_2.0.3               gtable_0.3.4                 
 [73] blob_1.2.4                    XVector_0.42.0               
 [75] htmltools_0.5.6.1             bookdown_0.36                
 [77] ProtGenerics_1.34.0           scales_1.2.1                 
 [79] png_0.1-8                     knitr_1.44                   
 [81] rstudioapi_0.15.0             rjson_0.2.21                 
 [83] curl_5.1.0                    cachem_1.0.8                 
 [85] stringr_1.5.0                 BiocVersion_3.18.0           
 [87] parallel_4.3.1                vipor_0.4.5                  
 [89] restfulr_0.0.15               pillar_1.9.0                 
 [91] grid_4.3.1                    vctrs_0.6.4                  
 [93] promises_1.2.1                dbplyr_2.3.4                 
 [95] beachmat_2.18.0               xtable_1.8-4                 
 [97] cluster_2.1.4                 beeswarm_0.4.0               
 [99] evaluate_0.22                 cli_3.6.1                    
[101] locfit_1.5-9.8                compiler_4.3.1               
[103] Rsamtools_2.18.0              rlang_1.1.1                  
[105] crayon_1.5.2                  labeling_0.4.3               
[107] ggbeeswarm_0.7.2              stringi_1.7.12               
[109] viridisLite_0.4.2             munsell_0.5.0                
[111] Biostrings_2.70.0             lazyeval_0.2.2               
[113] Matrix_1.6-1.1                dir.expiry_1.10.0            
[115] ExperimentHub_2.10.0          hms_1.1.3                    
[117] sparseMatrixStats_1.14.0      bit64_4.0.5                  
[119] Rhdf5lib_1.24.0               KEGGREST_1.42.0              
[121] statmod_1.5.0                 shiny_1.7.5.1                
[123] interactiveDisplayBase_1.40.0 AnnotationHub_3.10.0         
[125] igraph_1.5.1                  memoise_2.0.1                
[127] bslib_0.5.1                   bit_4.0.5                    
```
</div>