Literature results
If we explore the full table in lit_df_filter, we can see lots of links between the two traits, pinned on specific overlapping terms. For example:
Aspirin:INHIBITS:Oral anticoagulants from Sleep duration (s1) and anticoagulants:INHIBITS:P2RY12 from CHD (s2).
We can summarise the SemMedDB semantic type and number of overlapping terms:
lit_counts <- lit_df_filter %>%
count(`st.type`, `st.name`) %>%
arrange(`st.type`, desc(`n`))
lit_counts %>% print(n = 30)
#> # A tibble: 21 x 3
#> st.type st.name n
#> <chr> <chr> <int>
#> 1 aapp Leptin 6
#> 2 aapp Somatotropin 5
#> 3 aapp Adrenergic Receptor 4
#> 4 aapp Glutathione 2
#> 5 aapp apolipoprotein E-4 2
#> 6 aapp Monoamine Oxidase 1
#> 7 gngm Cytochrome P450 10
#> 8 gngm Glycoproteins 2
#> 9 horm Melatonin 28
#> 10 horm Hormones 3
#> 11 orch Ethanol 672
#> 12 orch Oral anticoagulants 10
#> 13 orch Metoprolol 8
#> 14 orch Propofol 2
#> 15 orch Acetaldehyde 1
#> 16 orch Benzodiazepines 1
#> 17 orch Luteolin 1
#> 18 orch Morphine 1
#> 19 orch Neostigmine 1
#> 20 orch Thiopental 1
#> 21 orch gamma hydroxybutyrate 1Note, the SemMedDB semantic types have been pre-filtered to only include a subset of possibilities.
Further examples of these term IDs and descriptions can be found here - https://mmtx.nlm.nih.gov/MMTx/semanticTypes.shtml
Plotting the overlapping term frequencies
We can also visualise the above table as a bar chart. In this case we will remove Ethanol as it is an outlier.
lit_counts %>%
filter(n < 100) %>%
{
ggplot(.) +
aes(x = `st.name`, y = `n`, fill = `st.type`) +
geom_col() +
geom_text(
aes(label = `n`),
position = position_dodge(0.9),
hjust = 0
) +
coord_flip()
}
Look in detail at one overlapping term
Here we look at cases where Leptin is the central overlapping term.
focus_term <- "Leptin"
lit_detail <- lit_df_filter %>% filter(`st.name` == focus_term)
lit_detail %>% head()
#> # A tibble: 6 x 18
#> gwas.trait gwas.id gs1.localCount gs1.pval s1.subject_name s1.predicate s1.id
#> <chr> <chr> <int> <dbl> <chr> <chr> <chr>
#> 1 Sleep dur… ieu-a-… 5 1.77e-12 ghrelin INHIBITS ghre…
#> 2 Sleep dur… ieu-a-… 5 1.77e-12 ghrelin INHIBITS ghre…
#> 3 Sleep dur… ieu-a-… 5 1.77e-12 ghrelin INHIBITS ghre…
#> 4 Sleep dur… ieu-a-… 5 1.77e-12 ghrelin INHIBITS ghre…
#> 5 Sleep dur… ieu-a-… 5 1.77e-12 ghrelin INHIBITS ghre…
#> 6 Sleep dur… ieu-a-… 5 1.77e-12 ghrelin INHIBITS ghre…
#> # … with 11 more variables: s1.object_name <chr>, st.name <chr>, st.type <chr>,
#> # s2.subject_name <chr>, s2.predicate <chr>, s2.id <chr>,
#> # s2.object_name <chr>, gs2.localCount <int>, gs2.pval <dbl>,
#> # assoc_gwas.trait <chr>, assoc_gwas.id <chr>We can create a network diagram to visualise these relationships.
lit_detail <- lit_detail %>%
mutate_at(vars(`gwas.trait`, `assoc_gwas.trait`), stringr::str_to_upper)
# add node types: 1 - selected GWAS, 2 - traits from literature, 3 - current focus term connecting 1 and 2
nodes <- bind_rows(
lit_detail %>% select(node = `gwas.trait`) %>% distinct() %>% mutate(node_type = 1),
lit_detail %>% select(node = `assoc_gwas.trait`) %>% distinct() %>% mutate(node_type = 1),
lit_detail %>% select(node = `s1.subject_name`) %>% distinct() %>% mutate(node_type = 2),
lit_detail %>% select(node = `s2.subject_name`) %>% distinct() %>% mutate(node_type = 2),
lit_detail %>% select(node = `s1.object_name`) %>% distinct() %>% mutate(node_type = 2),
lit_detail %>% select(node = `s2.object_name`) %>% distinct() %>% mutate(node_type = 2)
) %>%
mutate(node_type = ifelse(node == focus_term, 3, node_type)) %>%
distinct()
nodes
#> # A tibble: 10 x 2
#> node node_type
#> <chr> <dbl>
#> 1 SLEEP DURATION 1
#> 2 CORONARY HEART DISEASE 1
#> 3 ghrelin 2
#> 4 Leptin 3
#> 5 Coronary Arteriosclerosis 2
#> 6 Coronary heart disease 2
#> 7 Proteome 2
#> 8 Sleep Apnea, Obstructive 2
#> 9 Adiponectin 2
#> 10 Insulin 2edges <- bind_rows(
# exposure -> s1 subject
lit_detail %>%
select(node = `gwas.trait`, assoc_node = `s1.subject_name`) %>%
distinct(),
# s2 object -> outcome
lit_detail %>%
select(node = `s2.object_name`, assoc_node = `assoc_gwas.trait`) %>%
distinct(),
# s1 subject - s1 predicate -> s1 object
lit_detail %>%
select(
node = `s1.subject_name`, assoc_node = `s1.object_name`,
label = `s1.predicate`
) %>%
distinct(),
# s2 subject - s2 predicate -> s2 object
lit_detail %>%
select(
node = `s2.subject_name`, assoc_node = `s2.object_name`,
label = `s2.predicate`
) %>%
distinct()
) %>%
distinct()
edges
#> # A tibble: 14 x 3
#> node assoc_node label
#> <chr> <chr> <chr>
#> 1 SLEEP DURATION ghrelin <NA>
#> 2 Coronary Arteriosclerosis CORONARY HEART DISEASE <NA>
#> 3 Coronary heart disease CORONARY HEART DISEASE <NA>
#> 4 Proteome CORONARY HEART DISEASE <NA>
#> 5 Sleep Apnea, Obstructive CORONARY HEART DISEASE <NA>
#> 6 Adiponectin CORONARY HEART DISEASE <NA>
#> 7 Insulin CORONARY HEART DISEASE <NA>
#> 8 ghrelin Leptin INHIBITS
#> 9 Leptin Coronary Arteriosclerosis TREATS
#> 10 Leptin Coronary heart disease PREDISPOSES
#> 11 Leptin Proteome INTERACTS_WITH
#> 12 Leptin Sleep Apnea, Obstructive TREATS
#> 13 Leptin Adiponectin STIMULATES
#> 14 Leptin Insulin INTERACTS_WITHplot_network <- function(edges, nodes, show_edge_labels = FALSE) {
# default is to not display edge labels
if (!show_edge_labels) {
edges <- select(edges, -label)
}
graph <- graph_from_data_frame(edges, directed = TRUE, vertices = nodes)
graph$layout <- layout_with_kk
# generate colors based on node type
colrs <- c("tomato", "steelblue", "gold")
V(graph)$color <- colrs[V(graph)$node_type]
# Configure canvas
default_mar <- par("mar")
new_mar <- c(0, 0, 0, 0)
par(mar = new_mar)
plot.igraph(graph,
vertex.size = 13,
vertex.label.color = "black",
vertex.label.family = "Helvetica",
vertex.label.cex = 0.8,
edge.arrow.size = 0.4,
edge.label.color = "black",
edge.label.family = "Helvetica",
edge.label.cex = 0.5
)
par(mar = default_mar)
}plot_network(edges, nodes, show_edge_labels = TRUE)