Over-representation analysis¶

Gene Ontology¶

We will start by analysing our gene lists with the Gene Ontology annotation which classifies genes into gene sets according to three main types of information:

Molecular Function (MF): protein activity of the gene product
Biological Process (BP): set of protein activities leading to a common task
Cellular Component (CC): location of the gene product

Each set of genes is called a gene set and is grouped according to Gene Ontology terms (referred to here as GO terms). The GO terms are classified in a tree structure with general gene sets that become more specific as we go along. The relationships between GO terms can reflect different cases:

is a : A GO term is a subtype of the GO term B. For example mitochondrion is a organelle.
part of : the term GO A is part of the term GO B, so if the term GO A is present then so is the term GO B. For example mitochondrion is a part of cytoplasm.
has part : the term GO A necessarily contains the term GO B, but if there is the term GO B there is not necessarily the term GO A. For example, the receptor tyrosine kinase activity has part ATP hydrolysis activity.
regulates : the term GO A necessarily impacts the term GO B, but the latter is not necessarily impacted by A.

In the ClusterProfiler package we will use the enrichGO function which calculates for each GO term the over-representation of the genes in the analysed cluster among those in the term.

To use the Gene Ontology database, we use an R package containing all the annotation of the desired organism. These packages are called OrgDb which can be found in the form org.[Genome Initials].eg.db (for the human annotation, the OrgDb is org.Hs.eg.db). They contain gene identifiers from different resources (NCBI, Ensembl, RefSeq, etc...) with annotation from different databases (GO, OMIM, PMID, Uniprot, etc...). The default gene identifiers are in the format entrez (which is a sequence of numbers). Since we only have the Ensembl identifiers or the gene name available to us, we will use the keyType parameter of the enrichGO function to use the gene names instead of the entrez identifiers and thus use the different information contained in the Org.Db.

## What's inside an organism database ?
ls("package:org.Hs.eg.db")

## You must see the help section of category you want to know about and don't hesite to test the examples to understand the architecture
?org.Hs.egENSEMBL    #Link between ensembl ID and entrez ID
?org.Hs.egSYMBOL2EG  #Link between entrez ID and gene name
?org.Hs.egSYMBOL     #Link between gene name and entrez ID
?org.Hs.egGENENAME   #Beware ! It concern gene description and not gene name as we know

## Different available mapping variable name
columns(org.Hs.eg.db)

We will therefore apply the enrichGO function for each clusters through a lapply. For each cluster :

We filter the marker result dataframe to retrieve only the rows concerning the genes overexpressed by the cluster cells.
We run the enrichGO function to obtain the GO terms (BP, CC and MF) enriched in the marker gene set.
Add the cluster name in a new column to the resulting dataframe
We visualize the results with the dotplot function of the enrichR package which allows to visualize the first 3 GO terms for each ontology type

The result of the lapply is a list where each element of the list is a resultant dataframe for each cluster. We then assemble the results with the do.call function which applies a function to the whole list. The rbind will concatenate the rows of all the elements in the list so that there is only one dataframe with the results for each cluster.

## GO enrichment for all clusters
enrich_go_list <- lapply(levels(pbmc_markers_signif$cluster), function(cluster_name){

  res_markers <- subset(pbmc_markers_signif, cluster == cluster_name & avg_log2FC > 0) #Filter markers to retrieve only positive markers for the specific cluster

  ego <- enrichGO(gene = res_markers$external_gene_name,                 #Vector of target genes
                  universe = unique(annotated_hg19$external_gene_name),  #Vector of reference genes (all genes from the differential analysis)
                  OrgDb = "org.Hs.eg.db",                                #Organisme database
                  keyType = "SYMBOL",                                    #Column name of the OrgDB that convert gene format in `gene`parameter to entrez ID
                  ont = "ALL")                                           #What category of GO you want to analyse (BP, CC, MF or ALL)

  ego@result$cluster <- cluster_name                                     #Add cluster name in a new column named "cluster"

  ## visualisation
  ### don't forget to add print when inside a function/loop/lapply
  print(dotplot(ego,                                                     #enrichResult object
                split = "ONTOLOGY",                                      #Do separated plot for each ontology type (only valable fo GO results)
                showCategory = 3,                                        #Only show first three categories
                title = paste("Cluster", cluster_name)) +                #Add title
    facet_grid(ONTOLOGY~., scales = "free_y") +                          #Create subplot according to type used with `split = "ONTOLOGY"`
    theme(axis.text.y = element_text(size = 5),
                legend.key.size = unit(0.2, 'cm')))                      #Reduce ontology labels names
  return(ego@result)
})

enrich_go <- do.call("rbind", enrich_go_list)                            #Bind all results together
enrich_go <- enrich_go %>%
  group_by(cluster, ONTOLOGY)                                            #Rearrange df according to cluster and ontology type


## show first results
kable(top_n(x= enrich_go, n = 3, wt = (Count))[,-9])               #Only remove list of genes for the visualisation

First Enriched GO terms for each cluster

ONTOLOGY	Description	GeneRatio	BgRatio	pvalue	p.adjust	qvalue	Count	cluster
BP	translational initiation	69/111	191/16175	0.0000000	0.0000000	0.0000000	69	0
BP	mRNA catabolic process	69/111	359/16175	0.0000000	0.0000000	0.0000000	69	0
BP	RNA catabolic process	69/111	395/16175	0.0000000	0.0000000	0.0000000	69	0
CC	cytosolic ribosome	66/113	104/16891	0.0000000	0.0000000	0.0000000	66	0
CC	ribosomal subunit	66/113	179/16891	0.0000000	0.0000000	0.0000000	66	0
CC	ribosome	66/113	216/16891	0.0000000	0.0000000	0.0000000	66	0
MF	structural constituent of ribosome	66/112	158/16533	0.0000000	0.0000000	0.0000000	66	0
MF	rRNA binding	15/112	58/16533	0.0000000	0.0000000	0.0000000	15	0
MF	mRNA binding	14/112	284/16533	0.0000000	0.0000005	0.0000005	14	0
BP	neutrophil degranulation	69/274	468/16175	0.0000000	0.0000000	0.0000000	69	1
BP	granulocyte activation	70/274	486/16175	0.0000000	0.0000000	0.0000000	70	1
BP	neutrophil activation involved in immune response	69/274	471/16175	0.0000000	0.0000000	0.0000000	69	1
BP	neutrophil activation	69/274	481/16175	0.0000000	0.0000000	0.0000000	69	1
BP	neutrophil mediated immunity	69/274	482/16175	0.0000000	0.0000000	0.0000000	69	1
CC	secretory granule lumen	37/285	314/16891	0.0000000	0.0000000	0.0000000	37	1
CC	cytoplasmic vesicle lumen	37/285	318/16891	0.0000000	0.0000000	0.0000000	37	1
CC	vesicle lumen	37/285	320/16891	0.0000000	0.0000000	0.0000000	37	1
MF	electron transfer activity	18/278	121/16533	0.0000000	0.0000000	0.0000000	18	1
MF	amide binding	17/278	363/16533	0.0001466	0.0041824	0.0035415	17	1
MF	enzyme inhibitor activity	16/278	362/16533	0.0004353	0.0090741	0.0076836	16	1
BP	T cell activation	17/67	458/16175	0.0000000	0.0000000	0.0000000	17	2
BP	mRNA catabolic process	15/67	359/16175	0.0000000	0.0000000	0.0000000	15	2
BP	RNA catabolic process	15/67	395/16175	0.0000000	0.0000000	0.0000000	15	2
BP	mononuclear cell differentiation	15/67	402/16175	0.0000000	0.0000000	0.0000000	15	2
CC	cytosolic ribosome	12/70	104/16891	0.0000000	0.0000000	0.0000000	12	2
CC	ribosomal subunit	12/70	179/16891	0.0000000	0.0000000	0.0000000	12	2
CC	ribosome	12/70	216/16891	0.0000000	0.0000000	0.0000000	12	2
MF	structural constituent of ribosome	12/69	158/16533	0.0000000	0.0000000	0.0000000	12	2
MF	protein-macromolecule adaptor activity	7/69	254/16533	0.0000891	0.0037620	0.0029300	7	2
MF	molecular adaptor activity	7/69	307/16533	0.0002863	0.0079818	0.0062167	7	2
MF	cell adhesion molecule binding	8/69	500/16533	0.0010961	0.0223564	0.0174124	8	2
BP	translational initiation	27/120	191/16175	0.0000000	0.0000000	0.0000000	27	3
BP	protein targeting to membrane	27/120	197/16175	0.0000000	0.0000000	0.0000000	27	3
BP	establishment of protein localization to membrane	27/120	328/16175	0.0000000	0.0000000	0.0000000	27	3
BP	RNA catabolic process	28/120	395/16175	0.0000000	0.0000000	0.0000000	28	3
BP	protein targeting	28/120	420/16175	0.0000000	0.0000000	0.0000000	28	3
BP	adaptive immune response	27/120	401/16175	0.0000000	0.0000000	0.0000000	27	3
CC	cytosolic ribosome	25/125	104/16891	0.0000000	0.0000000	0.0000000	25	3
CC	ribosome	27/125	216/16891	0.0000000	0.0000000	0.0000000	27	3
CC	ribosomal subunit	25/125	179/16891	0.0000000	0.0000000	0.0000000	25	3
MF	structural constituent of ribosome	26/123	158/16533	0.0000000	0.0000000	0.0000000	26	3
MF	rRNA binding	11/123	58/16533	0.0000000	0.0000000	0.0000000	11	3
MF	immune receptor activity	11/123	131/16533	0.0000000	0.0000001	0.0000001	11	3
BP	adaptive immune response	20/66	401/16175	0.0000000	0.0000000	0.0000000	20	4
BP	T cell activation	17/66	458/16175	0.0000000	0.0000000	0.0000000	17	4
BP	regulation of leukocyte activation	15/66	493/16175	0.0000000	0.0000001	0.0000001	15	4
CC	external side of plasma membrane	12/67	312/16891	0.0000000	0.0000002	0.0000001	12	4
CC	coated vesicle	10/67	283/16891	0.0000002	0.0000027	0.0000021	10	4
CC	endosome membrane	11/67	483/16891	0.0000028	0.0000330	0.0000251	11	4
MF	T cell receptor binding	5/67	11/16533	0.0000000	0.0000001	0.0000001	5	4
MF	antigen binding	7/67	52/16533	0.0000000	0.0000002	0.0000001	7	4
MF	peptide antigen binding	5/67	23/16533	0.0000000	0.0000021	0.0000017	5	4
MF	serine-type endopeptidase activity	5/67	156/16533	0.0004229	0.0113647	0.0089582	5	4
MF	serine-type peptidase activity	5/67	174/16533	0.0006950	0.0146216	0.0115254	5	4
MF	serine hydrolase activity	5/67	176/16533	0.0007318	0.0146216	0.0115254	5	4
BP	neutrophil mediated immunity	59/274	482/16175	0.0000000	0.0000000	0.0000000	59	5
BP	neutrophil degranulation	58/274	468/16175	0.0000000	0.0000000	0.0000000	58	5
BP	neutrophil activation involved in immune response	58/274	471/16175	0.0000000	0.0000000	0.0000000	58	5
BP	neutrophil activation	58/274	481/16175	0.0000000	0.0000000	0.0000000	58	5
BP	granulocyte activation	58/274	486/16175	0.0000000	0.0000000	0.0000000	58	5
CC	cytoplasmic vesicle lumen	33/285	318/16891	0.0000000	0.0000000	0.0000000	33	5
CC	vesicle lumen	33/285	320/16891	0.0000000	0.0000000	0.0000000	33	5
CC	secretory granule lumen	32/285	314/16891	0.0000000	0.0000000	0.0000000	32	5
MF	actin binding	26/277	424/16533	0.0000000	0.0000066	0.0000056	26	5
MF	ubiquitin-like protein ligase binding	16/277	300/16533	0.0000487	0.0041852	0.0035090	16	5
MF	cell adhesion molecule binding	20/277	500/16533	0.0003057	0.0162987	0.0136653	20	5
BP	immune response-regulating cell surface receptor signaling pathway	30/201	385/16175	0.0000000	0.0000000	0.0000000	30	6
BP	immune response-regulating signaling pathway	30/201	388/16175	0.0000000	0.0000000	0.0000000	30	6
BP	regulation of response to biotic stimulus	29/201	395/16175	0.0000000	0.0000000	0.0000000	29	6
BP	activation of immune response	29/201	437/16175	0.0000000	0.0000000	0.0000000	29	6
BP	positive regulation of response to external stimulus	29/201	479/16175	0.0000000	0.0000000	0.0000000	29	6
BP	granulocyte activation	29/201	486/16175	0.0000000	0.0000000	0.0000000	29	6
CC	focal adhesion	28/212	407/16891	0.0000000	0.0000000	0.0000000	28	6
CC	cell-substrate junction	28/212	413/16891	0.0000000	0.0000000	0.0000000	28	6
CC	actin cytoskeleton	22/212	478/16891	0.0000002	0.0000062	0.0000048	22	6
MF	actin binding	15/211	424/16533	0.0003694	0.0138507	0.0117285	15	6
MF	cell adhesion molecule binding	16/211	500/16533	0.0007065	0.0244203	0.0206787	16	6
MF	endopeptidase activity	14/211	411/16533	0.0008361	0.0250827	0.0212396	14	6
BP	immune response-regulating signaling pathway	26/112	388/16175	0.0000000	0.0000000	0.0000000	26	7
BP	immune response-regulating cell surface receptor signaling pathway	24/112	385/16175	0.0000000	0.0000000	0.0000000	24	7
BP	activation of immune response	25/112	437/16175	0.0000000	0.0000000	0.0000000	25	7
BP	neutrophil activation involved in immune response	24/112	471/16175	0.0000000	0.0000000	0.0000000	24	7
BP	neutrophil activation	24/112	481/16175	0.0000000	0.0000000	0.0000000	24	7
BP	neutrophil mediated immunity	24/112	482/16175	0.0000000	0.0000000	0.0000000	24	7
BP	granulocyte activation	24/112	486/16175	0.0000000	0.0000000	0.0000000	24	7
CC	lysosomal membrane	20/115	351/16891	0.0000000	0.0000000	0.0000000	20	7
CC	lytic vacuole membrane	20/115	351/16891	0.0000000	0.0000000	0.0000000	20	7
CC	vacuolar membrane	21/115	402/16891	0.0000000	0.0000000	0.0000000	21	7
MF	peptide binding	14/113	293/16533	0.0000000	0.0000013	0.0000010	14	7
MF	amide binding	15/113	363/16533	0.0000000	0.0000015	0.0000012	15	7
MF	enzyme inhibitor activity	11/113	362/16533	0.0000377	0.0012609	0.0010186	11	7
BP	wound healing	31/178	497/16175	0.0000000	0.0000000	0.0000000	31	8
BP	blood coagulation	23/178	321/16175	0.0000000	0.0000000	0.0000000	23	8
BP	hemostasis	23/178	325/16175	0.0000000	0.0000000	0.0000000	23	8
BP	coagulation	23/178	325/16175	0.0000000	0.0000000	0.0000000	23	8
BP	regulation of body fluid levels	23/178	477/16175	0.0000000	0.0000009	0.0000008	23	8
CC	actin cytoskeleton	25/181	478/16891	0.0000000	0.0000000	0.0000000	25	8
CC	cell-substrate junction	20/181	413/16891	0.0000000	0.0000012	0.0000010	20	8
CC	focal adhesion	19/181	407/16891	0.0000001	0.0000024	0.0000020	19	8
MF	actin binding	19/179	424/16533	0.0000002	0.0000735	0.0000652	19	8
MF	actin filament binding	10/179	201/16533	0.0000685	0.0090608	0.0080366	10	8
MF	GTPase activity	11/179	291/16533	0.0003372	0.0278493	0.0247013	11	8

The result is a dataframe where the GO terms have been considered as enriched:

ONTOLOGY : ontology type (BP, CC, or MF)
ID : Unique identifier of the GO term
Description : Description of the GO term
GeneRatio : Fraction representing the number of marker genes present in the GO term, \(GeneRatio = \frac{nbrMarkerGeneInKEGGcat}{nbrMarkerGene}\).
BgRatio : Fraction representing the number of reference genes present in the GO term, \(BgRatio = \frac{nbrTotalGeneInKEGGcat}{nbrTotalGene}\).
pvalue : p-value of the enrichment test
p.adjust : adjusted p-value of the Benjamini Hochberg test
qvalue : q-value after FDR (False Discovery Rate) check
geneID : String containing the list of marker genes present in the GO term (separated by /)
count : Number of marker genes present in the GO term
cluster : Column added before the do.call("rbind", list) to identify in which cluster the GO term has been considered as enriched.

A GO term has been considered as enriched if :

the p-value \(\lt\) 0.05
the adjusted p-value \(\lt\) 0.05
the q-value \(\lt\) 0.2

Kyoto Encyclopedia of Genes and Genomes (KEGG)¶

KEGG is a database that focuses on the molecular annotation of the different metabolic pathways. It allows the description of the biochemical reactions that compose the pathway. KEGG also allows the visualization of these pathways through hand-drawn maps representing the different reactions and the relationship between the genes. There are several main categories of KEGG pathways:

Metabolism
Genetic information processing
Environmental information processing
Cellular processes
Organ systems
Human diseases
Drug development

We will use the enrichKEGG function from the ClusterProfiler package. This function doesn't work exactly like enrichGO because it calls directly on the database and doesn't use an Orgdb package but it does require our genes to be annotated with an entrez id. We will have to find another way to convert our genes into the correct format.

To do this, we will use the Orgdb or org.Hs.egSYMBOL object to list each enter id as its "gene symbol" (gene name). When we convert this object to a dataframe we get a table with two columns (gene_id and symbol). We now have the possibility to switch from a gene name to an entrez id.

We will therefore apply the enrichKEGG function for each clusters through a lapply. For each cluster :

We filter the marker result dataframe to retrieve only the rows concerning the genes overexpressed by the cluster cells.
Run the enrichGO function to get the enriched KEGG terms in the marker gene set, filter our enter/symbol mapping table for the gene and universe parameters with the gene names contained in the marker and biomart annotation tables.
Add the cluster name in a new column to the resulting dataframe
We visualize the results with the dotplot function of the enrichR package which allows to visualize the first 5 KEGG terms

## Retrieve a corresponding table between entrez id and gene name (called gene symbol in org.db)
corresp_entrez <- as.data.frame(org.Hs.egSYMBOL)  #Change format to df

## Apply enrichKEGG for each cluster
enrich_kegg_list <- lapply(levels(pbmc_markers_signif$cluster), function(cluster_name){

  res_markers <- subset(pbmc_markers_signif, cluster == cluster_name & avg_log2FC > 0) #Filter markers dataframe based on cluster

  ## Perform enrichKEGG analysis
  ekegg <- enrichKEGG(gene = subset(corresp_entrez, symbol %in% res_markers$external_gene_name)$gene_id,                #Genes to analyse
                      universe = subset(corresp_entrez, symbol %in% unique(annotated_hg19$external_gene_name))$gene_id, #Background genes, here we take all genes from our expression matrix
                      organism = "hsa")

  ekegg@result$cluster <- cluster_name            #Add cluster name as column

  ## Add plot
  print(dotplot(ekegg,
                label_format = 30,
                font.size = 10,
                showCategory = 5,
                title = paste("Cluster", cluster_name)) +
    theme(axis.text.y = element_text(size = 10),
                legend.key.size = unit(0.2, 'cm')))

  return(ekegg@result)                            #Return dataframe result
})

## Concatenate all results in one dataframe
enrich_kegg <- do.call("rbind", enrich_kegg_list)

## Group result by cluster (easier to manipulate with dplyr)
enrich_kegg <- enrich_kegg %>%
  group_by(cluster)

## Visualise first 3 KEGG categories for each cluster (removing the vector of genes just for the visualisation)
kable(top_n(x= enrich_kegg, n = 3, wt = (Count))[,-8])

First Enriched KEGG categories for each cluster

ID	Description	GeneRatio	BgRatio	pvalue	p.adjust	qvalue	Count	cluster
hsa03010	Ribosome	66/94	134/7558	0.0000000	0.0000000	0.0000000	66	0
hsa05171	Coronavirus disease - COVID-19	66/94	228/7558	0.0000000	0.0000000	0.0000000	66	0
hsa04640	Hematopoietic cell lineage	6/94	94/7558	0.0010652	0.0242334	0.0241073	6	0
hsa05415	Diabetic cardiomyopathy	24/179	175/7558	0.0000000	0.0000000	0.0000000	24	1
hsa05020	Prion disease	23/179	246/7558	0.0000000	0.0000004	0.0000003	23	1
hsa05022	Pathways of neurodegeneration - multiple diseases	23/179	443/7558	0.0002945	0.0018710	0.0013802	23	1
hsa03010	Ribosome	12/44	134/7558	0.0000000	0.0000000	0.0000000	12	2
hsa05171	Coronavirus disease - COVID-19	14/44	228/7558	0.0000000	0.0000000	0.0000000	14	2
hsa05170	Human immunodeficiency virus 1 infection	9/44	204/7558	0.0000020	0.0000413	0.0000317	9	2
hsa03010	Ribosome	27/81	134/7558	0.0000000	0.0000000	0.0000000	27	3
hsa05171	Coronavirus disease - COVID-19	28/81	228/7558	0.0000000	0.0000000	0.0000000	28	3
hsa04640	Hematopoietic cell lineage	17/81	94/7558	0.0000000	0.0000000	0.0000000	17	3
hsa04650	Natural killer cell mediated cytotoxicity	10/47	124/7558	0.0000000	0.0000002	0.0000001	10	4
hsa04514	Cell adhesion molecules	10/47	147/7558	0.0000000	0.0000005	0.0000004	10	4
hsa05170	Human immunodeficiency virus 1 infection	11/47	204/7558	0.0000000	0.0000008	0.0000006	11	4
hsa05166	Human T-cell leukemia virus 1 infection	11/47	219/7558	0.0000001	0.0000011	0.0000008	11	4
hsa04145	Phagosome	21/180	146/7558	0.0000000	0.0000000	0.0000000	21	5
hsa05152	Tuberculosis	21/180	176/7558	0.0000000	0.0000001	0.0000001	21	5
hsa05022	Pathways of neurodegeneration - multiple diseases	21/180	443/7558	0.0018748	0.0133316	0.0100316	21	5
hsa04650	Natural killer cell mediated cytotoxicity	23/126	124/7558	0.0000000	0.0000000	0.0000000	23	6
hsa05163	Human cytomegalovirus infection	17/126	222/7558	0.0000001	0.0000043	0.0000033	17	6
hsa05170	Human immunodeficiency virus 1 infection	16/126	204/7558	0.0000002	0.0000064	0.0000049	16	6
hsa05152	Tuberculosis	17/72	176/7558	0.0000000	0.0000000	0.0000000	17	7
hsa04145	Phagosome	15/72	146/7558	0.0000000	0.0000000	0.0000000	15	7
hsa05164	Influenza A	15/72	166/7558	0.0000000	0.0000000	0.0000000	15	7
hsa05168	Herpes simplex virus 1 infection	15/72	484/7558	0.0000389	0.0002724	0.0002048	15	7
hsa04611	Platelet activation	9/97	122/7558	0.0000240	0.0043154	0.0039621	9	8
hsa04530	Tight junction	9/97	165/7558	0.0002499	0.0224924	0.0206509	9	8
hsa04510	Focal adhesion	9/97	198/7558	0.0009458	0.0340474	0.0312599	9	8
hsa05022	Pathways of neurodegeneration - multiple diseases	9/97	443/7558	0.1141696	0.5182323	0.4758039	9	8

The result is a dataframe where KEGG categories have been considered as enriched with the following columns :

ID : Unique identifier of the KEGG category
Description : Description of the KEGG category
GeneRatio : Fraction representing the number of marker genes present in the KEGG category, \(GeneRatio = \frac{nbrMarkerGeneInKEGGcat}{nbrMarkerGene}\)
BgRatio : Fraction representing the number of genes of the reference present in the KEGG category, \(BgRatio = \frac{nbrTotalGeneInKEGGcat}{nbrTotalGene}\)
pvalue : p-value of the enrichment test
p.adjust : adjusted p-value of the Benjamini Hochberg test
qvalue : q-value after FDR (False Discovery Rate) check
geneID : String containing the list of marker genes present in the KEGG category (separated by /)
Count : Number of marker genes present in the KEGG category
cluster : Column added before the do.call("rbind", list) in order to identify in which cluster the KEGG category has been considered as enriched.

A KEGG category has been considered enriched if :

the p-value \(\lt\) 0.05
the adjusted p-value \(\lt\) 0.05
q-value \(\lt\) 0.2

These over-representation methods are highly dependent on the database containing fairly generalized groups of genes. However, we can see that the results of enrichGO, enrichKEGG and Biomart intersect for some clusters where :

Cluster 6 would represent the Natural Killer cells as well as cluster 4: knowing that they are very close on the UMAP it reflects their proximity of the transcriptomes of the cells that compose these two clusters. The GO analysis would however lean more towards T cells for cluster 4.
Cluster 8 would be composed of platelet cells

Knowing when taking only the first 3 or 5 results (so very restricted), we are extremely stringent in identifying clusters.