Basic Usage

goreparent clusters GO terms into groups of directly related terms with a common parent. The extent of the search for common parents between GO terms can be customised by the variables max_parents (default=2) - which defines how many generations up the search will be performed from initial ‘child’ GO terms - and max_from_top (default=3) - which defines how many levels down from the top level GO terms (“biological_process”, “molecular_function”, “cellular_component”) is considered outside of the search frame and should enable more specific clusters to be identified.

library(goreparent)

enrich_file = system.file("extdata","go_enrich_results.txt", package = "goreparent")
go_input = read.table(enrich_file, sep="\t", header=TRUE)

Group GO terms by common parents

go_parents = add_go_groups(go_input, descriptive_parent=FALSE)
## grouping 75 GO terms by common parents
## defined 13 parental groups
# plot with terms collapsed into groups
plot_go_parents(go_parents)

# plot with expanded GO terms for the top 2 groups 
plot_go_parents(go_parents, collapse=FALSE, n_top=2)

Advanced Usage

More refined and descriptive GO groups

Sometimes, if max_from_top is set too low, or too many terms are grouping under a few too broadly descriptive parents, you may want to further refine or describe these groups. This can be achieved one of three ways: altering max_from_top, adding terms to ignore_terms, or adding group descriptions with descriptive_parent=TRUE.

go_parents_l2 = add_go_groups(go_input, max_from_top = 2, descriptive_parent=FALSE)
## grouping 75 GO terms by common parents
## defined 13 parental groups
table(go_parents_l2$parent_description) %>% 
  as.data.frame() %>% 
  rename(parent_description=Var1) %>% 
  arrange(desc(Freq)) %>% 
  knitr::kable()
parent_description Freq
regulation of developmental process 15
cellular component organization 13
cell differentiation 11
cell-cell signaling 6
nervous system process 6
neuron death 6
cell growth 5
anatomical structure development 4
regulation of biological quality 3
regulation of cellular process 3
intrinsic apoptotic signaling pathway 1
locomotory behavior 1
multicellular organismal signaling 1

As we see from before, using max_from_top=3 [default], we found terms associated with neural processes, whereas here using max_from_top=2 we get broader categories such as "cell differentiation" and "cell-cell signaling". If we look at the GO terms contained within these clusters, we find that they are clearly related to neural processes.

go_parents_l2 %>% 
  filter(parent_description %in% c("cell differentiation", "cell-cell signaling")) %>% 
  select(parent_description, Description) %>% arrange(parent_description) %>% 
  knitr::kable()
parent_description Description
cell differentiation axonogenesis
cell differentiation axon development
cell differentiation dendrite morphogenesis
cell differentiation dendrite development
cell differentiation regulation of neuron projection development
cell differentiation dendritic spine morphogenesis
cell differentiation dendritic spine development
cell differentiation positive regulation of neuron projection development
cell differentiation neuron migration
cell differentiation neuron projection guidance
cell differentiation axon guidance
cell-cell signaling modulation of synaptic transmission
cell-cell signaling regulation of synaptic plasticity
cell-cell signaling long-term synaptic potentiation
cell-cell signaling Wnt signaling pathway
cell-cell signaling cell-cell signaling by wnt
cell-cell signaling canonical Wnt signaling pathway

1. Change max_from_top to a higher value

go_parents_l2 %>%
  select(ID, Description, FDR) %>%
  add_go_groups(max_from_top = 3, descriptive_parent = FALSE) %>%
  left_join(go_parents_l2[,c("Description", "parent_description")], by = "Description", suffix=c(".level3", ".level2")) %>%
  arrange(parent_description.level2, parent_description.level3)  %>%  select("Description", starts_with("parent_description")) %>% 
  filter(parent_description.level2 %in% c("cell differentiation","cell-cell signaling")) %>% 
  knitr::kable()
## grouping 75 GO terms by common parents
## defined 13 parental groups
Description parent_description.level3 parent_description.level2
axonogenesis nervous system development cell differentiation
axon development nervous system development cell differentiation
dendrite morphogenesis nervous system development cell differentiation
dendrite development nervous system development cell differentiation
regulation of neuron projection development nervous system development cell differentiation
dendritic spine morphogenesis nervous system development cell differentiation
dendritic spine development nervous system development cell differentiation
positive regulation of neuron projection development nervous system development cell differentiation
neuron migration nervous system development cell differentiation
neuron projection guidance nervous system development cell differentiation
axon guidance nervous system development cell differentiation
Wnt signaling pathway cell-cell signaling by wnt cell-cell signaling
cell-cell signaling by wnt cell-cell signaling by wnt cell-cell signaling
canonical Wnt signaling pathway cell-cell signaling by wnt cell-cell signaling
modulation of synaptic transmission chemical synaptic transmission cell-cell signaling
regulation of synaptic plasticity chemical synaptic transmission cell-cell signaling
long-term synaptic potentiation chemical synaptic transmission cell-cell signaling

2. Remove parent GO term manually

This can be helpful if all other groups are descriptive of their children terms, but a few are not. cell-cell signalling is too broad, but neuron death is not. This requires specifying add_go_groups(ignore_terms = "GO:0007267")

# GO description for cell-cell signaling
id2term("GO:0007267")
## [1] "cell-cell signaling"
go_parents_l2 %>%
  filter(parent_description %in% c("cell-cell signaling", "neuron death")) %>%
  select(ID, Description, FDR) %>%
  add_go_groups(descriptive_parent = FALSE, ignore_terms = "GO:0007267") %>%
  left_join(go_parents_l2[,c("Description", "parent_description")], by = "Description", suffix=c(".ignore_term", ".level2")) %>%
  arrange(parent_description.level2, parent_description.ignore_term)  %>%  select("Description", starts_with("parent_description")) %>% 
  knitr::kable()
## grouping 12 GO terms by common parents
## defined 3 parental groups
Description parent_description.ignore_term parent_description.level2
Wnt signaling pathway cell-cell signaling by wnt cell-cell signaling
cell-cell signaling by wnt cell-cell signaling by wnt cell-cell signaling
canonical Wnt signaling pathway cell-cell signaling by wnt cell-cell signaling
modulation of synaptic transmission modulation of chemical synaptic transmission cell-cell signaling
regulation of synaptic plasticity modulation of chemical synaptic transmission cell-cell signaling
long-term synaptic potentiation modulation of chemical synaptic transmission cell-cell signaling
neuron death neuron death neuron death
neuron apoptotic process neuron death neuron death
regulation of neuron death neuron death neuron death
regulation of neuron apoptotic process neuron death neuron death
negative regulation of neuron death neuron death neuron death
negative regulation of neuron apoptotic process neuron death neuron death

3. Remove GO terms with too many children

This is a generalised way to remove parent terms that may encompass too many child terms to be descriptive by themselves.

go_parents_l2 %>%
  select(ID, Description, FDR) %>%
  add_go_groups(max_from_top = 2, descriptive_parent = FALSE, max_children = 100) %>%
  left_join(go_parents_l2[,c("Description", "parent_description")], by = "Description", suffix=c(".max_child", ".level2")) %>%
  arrange(parent_description.level2, parent_description.max_child)  %>%  select("Description", starts_with("parent_description")) %>% 
  filter(parent_description.level2 %in% c("cell differentiation","cell-cell signaling")) %>% 
  knitr::kable()
## grouping 75 GO terms by common parents
## defined 32 parental groups
Description parent_description.max_child parent_description.level2
axonogenesis axon development cell differentiation
axon development axon development cell differentiation
dendrite morphogenesis dendrite development cell differentiation
dendrite development dendrite development cell differentiation
dendritic spine morphogenesis dendrite development cell differentiation
dendritic spine development dendrite development cell differentiation
neuron migration neuron migration cell differentiation
neuron projection guidance neuron projection guidance cell differentiation
axon guidance neuron projection guidance cell differentiation
positive regulation of neuron projection development positive regulation of cell projection organization cell differentiation
regulation of neuron projection development regulation of neuron projection development cell differentiation
Wnt signaling pathway cell-cell signaling by wnt cell-cell signaling
cell-cell signaling by wnt cell-cell signaling by wnt cell-cell signaling
canonical Wnt signaling pathway cell-cell signaling by wnt cell-cell signaling
modulation of synaptic transmission modulation of chemical synaptic transmission cell-cell signaling
regulation of synaptic plasticity modulation of chemical synaptic transmission cell-cell signaling
long-term synaptic potentiation modulation of chemical synaptic transmission cell-cell signaling

4. Adding additional “descriptive” parents

This option works on clusters of GO terms that may not be closely related, and thus works on GO clustering output from other clustering algorithms. It is, however, currently very slow, and therefore we only recommend running it on the top clusters of significance by setting n_top to an integer.

The descriptive parent option uses make_descriptive_parent() - a function which relies on finding GO terms which are semi-related to all GO terms in a cluster. This function works to find parent terms which are step-parents/uncles to children terms. This allows 1 degree of separation in defining parent terms, which once terms are ranked, chooses the parent term with the highest relatedness. If all child GO terms are unable to be related to the one parent even with this relaxation in relationships, further terms are tried until all child terms are covered by one parent.

Using normal goreparent parental clustering + descriptive_parent=TRUE

go_parents_l2.descriptive = add_go_groups(go_input, descriptive_parent=TRUE, max_from_top=2, verbose=FALSE, n_top = 10)
plot_go_parents(go_parents_l2.descriptive)

using make_descriptive_parent() separately post-add_go_group This creates a table with new_parent_description in a separate column which can be used to add to/replace the original ‘parent term’ in a format of your choosing.

go_parents_l2.add_desc = go_parents_l2 %>%
  filter(parent_description %in% c("cellular component organization")) %>%
  select(ID, Description, parent_description, FDR) %>%
  make_descriptive_parent(replace_group_col = FALSE) 

go_parents_l2.add_desc %>% head(2) %>% select(-FDR) %>% knitr::kable()
ID Description parent_description new_parent_description
GO:0050808 synapse organization cellular component organization synapse organization
GO:0007416 synapse assembly cellular component organization synapse organization 
library(ggplot2)
go_parents_l2.add_desc %>% 
  ggplot( aes(x=-log10(FDR), y=Description, col=-log10(FDR))) + 
  geom_point() + 
  facet_wrap(~parent_description+new_parent_description) +
  theme_minimal()+ 
  theme(panel.border = element_rect(fill=NA, size=0.1)) +
  theme(strip.text.x = element_text(face="bold", hjust=0))

Usage with simplifyEnrichment

We are aware of simplifyEnrichment and how it can cluster GO terms, however we found that the clusters - while semantically related - did not have an easily string-convertible description of what terms the cluster actually contains, and therefore still requires displaying all the child terms and naming in a non-informative manner.

There is a built-in option to use simplifyEnrichment semantic similarity clustering within add_go_groups(), using the option method="cluster". For this method we HIGHLY recommend running with descriptive_parent=TRUE.

go_clusters = add_go_groups(go_input, method = "cluster", descriptive_parent=TRUE)
plot_go_parents(go_clusters)