Initial setup

To show compatibility with tidy workflows we will use magrittr pipe operator and the dplyr verbs.

library("BaseSet")
library("dplyr")

Human gene ontology

We will explore the genes with assigned gene ontology terms. These terms describe what is the process and role of the genes. The links are annotated with different evidence codes to indicate how such annotation is supported.

# We load some libraries
library("org.Hs.eg.db")
library("GO.db")
library("ggplot2")
# Prepare the data 
h2GO_TS <- tidySet(org.Hs.egGO)
h2GO <- as.data.frame(org.Hs.egGO)

We can now explore if there are differences in evidence usage for each ontology in gene ontology:

library("forcats")
h2GO %>% 
  group_by(Evidence, Ontology) %>% 
  count(name = "Freq") %>% 
  ungroup() %>% 
  mutate(Evidence = fct_reorder2(Evidence, Ontology, -Freq),
         Ontology = case_when(Ontology == "CC" ~ "Cellular Component",
                              Ontology == "MF" ~ "Molecular Function",
                              Ontology == "BP" ~ "Biological Process",
                              TRUE ~ NA_character_)) %>% 
  ggplot() +
  geom_col(aes(Evidence, Freq)) +
  facet_grid(~Ontology) + 
  theme_minimal() +
  coord_flip() +
  labs(x = element_blank(), y = element_blank(),
       title = "Evidence codes for each ontology")

We can see that biological process are more likely to be defined by IMP evidence code that means inferred from mutant phenotype. While inferred from physical interaction (IPI) is almost exclusively used to assign molecular functions.

This graph doesn’t consider that some relationships are better annotated than other:

h2GO_TS %>% 
  relations() %>% 
  group_by(elements, sets) %>% 
  count(sort = TRUE, name = "Annotations") %>% 
  ungroup() %>% 
  count(Annotations, sort = TRUE) %>% 
  ggplot() +
  geom_col(aes(Annotations, n)) +
  theme_minimal() +
  labs(x = "Evidence codes", y = "Annotations", 
       title = "Evidence codes for each annotation",
       subtitle = "in human") +
  scale_x_continuous(breaks = 1:7)

We can see that mostly all the annotations are done with a single evidence code. So far we have explored the code that it is related to a gene but how many genes don’t have any annotation?

# Add all the genes and GO terms
h2GO_TS <- add_elements(h2GO_TS, keys(org.Hs.eg.db)) %>% 
  add_sets(grep("^GO:", keys(GO.db), value = TRUE))

sizes_element <- element_size(h2GO_TS) %>% 
    arrange(desc(size))
sum(sizes_element$size == 0)
#> [1] 40855
sum(sizes_element$size != 0)
#> [1] 20692

sizes_set <- set_size(h2GO_TS) %>% 
    arrange(desc(size))
sum(sizes_set$size == 0)
#> [1] 25924
sum(sizes_set$size != 0)
#> [1] 18348

So we can see that both there are more genes without annotation and more gene ontology terms without a (direct) gene annotated.

sizes_element %>% 
    filter(size != 0) %>% 
    ggplot() +
    geom_histogram(aes(size), binwidth = 1) +
    theme_minimal() +
    labs(x = "# sets per element", y = "Count")


sizes_set %>% 
    filter(size != 0) %>% 
    ggplot() +
    geom_histogram(aes(size), binwidth = 1) +
    theme_minimal() +
    labs(x = "# elements per set", y = "Count")

As you can see on the second plot we have very large values but that are on associated on many genes:

head(sizes_set, 10)
#>          sets  size probability Ontology
#> 1  GO:0005515 12067           1       MF
#> 2  GO:0005634  5183           1       CC
#> 3  GO:0005829  5151           1       CC
#> 4  GO:0005737  4742           1       CC
#> 5  GO:0005886  4642           1       CC
#> 6  GO:0005654  3682           1       CC
#> 7  GO:0016021  3661           1       CC
#> 8  GO:0046872  2323           1       MF
#> 9  GO:0070062  2167           1       CC
#> 10 GO:0016020  2114           1       CC

Using fuzzy values

This could radically change if we used fuzzy values. We could assign a fuzzy value to each evidence code given the lowest fuzzy value for the IEA (Inferred from Electronic Annotation) evidence. The highest values would be for evidence codes coming from experiments or alike.

nr <- h2GO_TS %>% 
  relations() %>% 
  dplyr::select(sets, Evidence) %>% 
  distinct() %>% 
  mutate(fuzzy = case_when(
    Evidence == "EXP" ~ 0.9,
    Evidence == "IDA" ~ 0.8,
    Evidence == "IPI" ~ 0.8,
    Evidence == "IMP" ~ 0.75,
    Evidence == "IGI" ~ 0.7,
    Evidence == "IEP" ~ 0.65,
    Evidence == "HEP" ~ 0.6,
    Evidence == "HDA" ~ 0.6,
    Evidence == "HMP" ~ 0.5,
    Evidence == "IBA" ~ 0.45,
    Evidence == "ISS" ~ 0.4,
    Evidence == "ISO" ~ 0.32,
    Evidence == "ISA" ~ 0.32,
    Evidence == "ISM" ~ 0.3,
    Evidence == "RCA" ~ 0.2,
    Evidence == "TAS" ~ 0.15,
    Evidence == "NAS" ~ 0.1,
    Evidence == "IC" ~ 0.02,
    Evidence == "ND" ~ 0.02,
    Evidence == "IEA" ~ 0.01,
    TRUE ~ 0.01)) %>% 
  dplyr::select(sets = "sets", elements = "Evidence", fuzzy = fuzzy)

We have several evidence codes for the same ontology, this would result on different fuzzy values for each relation. Instead, we extract this and add them as new sets and elements and add an extra column to classify what are those elements:

ts <- h2GO_TS %>% 
  relations() %>% 
  dplyr::select(-Evidence) %>% 
  rbind(nr) %>% 
  tidySet() %>% 
  mutate_element(Type = ifelse(grepl("^[0-9]+$", elements), "gene", "evidence"))

Now we can see which gene ontologies are more supported by the evidence:

ts %>% 
  dplyr::filter(Type != "Gene") %>% 
  cardinality() %>% 
  arrange(desc(cardinality)) %>% 
  head()
#>         sets cardinality
#> 1 GO:0005515    12069.61
#> 2 GO:0005634     5188.10
#> 3 GO:0005829     5155.08
#> 4 GO:0005737     4747.10
#> 5 GO:0005886     4646.78
#> 6 GO:0005654     3685.26

Surprisingly the most supported terms are protein binding, nucleus and cytosol. I would expect them to be the top three terms for cellular component, biological function and molecular function.

Calculating set sizes would be interesting but it requires computing a big number of combinations that make it last long and require many memory available.

ts %>% 
  filter(sets %in% c("GO:0008152", "GO:0003674", "GO:0005575"),
         Type != "gene") %>% 
  set_size()
#>         sets size probability
#> 1 GO:0003674    0        0.98
#> 2 GO:0003674    1        0.02
#> 3 GO:0005575    0        0.98
#> 4 GO:0005575    1        0.02
#> 5 GO:0008152    0        0.99
#> 6 GO:0008152    1        0.01

Unexpectedly there is few evidence for the main terms:

ts %>% 
  filter(sets %in% c("GO:0008152", "GO:0003674", "GO:0005575")) %>% 
  filter(Type != "gene") 
#>   elements       sets fuzzy     Type
#> 1      IEA GO:0008152  0.01 evidence
#> 2       ND GO:0005575  0.02 evidence
#> 3       ND GO:0003674  0.02 evidence

In fact those terms are arbitrarily decided or inferred from electronic analysis.

Human pathways

Now we will repeat the same analysis with pathways:

# We load some libraries
library("reactome.db")

# Prepare the data (is easier, there isn't any ontoogy or evidence column)
h2p <- as.data.frame(reactomeEXTID2PATHID)
colnames(h2p) <- c("sets", "elements")
# Filter only for human pathways
h2p <- h2p[grepl("^R-HSA-", h2p$sets), ]

# There are duplicate relations with different evidence codes!!: 
summary(duplicated(h2p[, c("elements", "sets")]))
#>    Mode   FALSE    TRUE 
#> logical  118909   13179
h2p <- unique(h2p)
# Create a tidySet and 
h2p_TS <- tidySet(h2p) %>% 
    # Add all the genes 
    add_elements(keys(org.Hs.eg.db))

Now that we have everything ready we can start measuring some things…

sizes_element <- element_size(h2p_TS) %>% 
    arrange(desc(size))
sum(sizes_element$size == 0)
#> [1] 50843
sum(sizes_element$size != 0)
#> [1] 10997

sizes_set <- set_size(h2p_TS) %>% 
    arrange(desc(size))

We can see there are more genes without pathways than genes with pathways.

sizes_element %>% 
    filter(size != 0) %>% 
    ggplot() +
    geom_histogram(aes(size), binwidth = 1) +
    scale_y_log10() +
    theme_minimal() +
    labs(x = "# sets per element", y = "Count")
#> Warning: Transformation introduced infinite values in continuous y-axis
#> Warning: Removed 264 rows containing missing values (geom_bar).


sizes_set %>% 
    ggplot() +
    geom_histogram(aes(size), binwidth = 1) +
    scale_y_log10() +
    theme_minimal() +
    labs(x = "# elements per set", y = "Count")
#> Warning: Transformation introduced infinite values in continuous y-axis
#> Warning: Removed 2550 rows containing missing values (geom_bar).

As you can see on the second plot we have very large values but that are on associated on many genes:

head(sizes_set, 10)
#>             sets size probability
#> 1   R-HSA-162582 2804           1
#> 2   R-HSA-168256 2161           1
#> 3  R-HSA-1430728 2114           1
#> 4   R-HSA-392499 2036           1
#> 5  R-HSA-1643685 1859           1
#> 6    R-HSA-74160 1504           1
#> 7   R-HSA-597592 1431           1
#> 8    R-HSA-73857 1364           1
#> 9   R-HSA-212436 1241           1
#> 10  R-HSA-372790 1185           1

Session info

#> R version 4.0.1 (2020-06-06)
#> Platform: x86_64-pc-linux-gnu (64-bit)
#> Running under: Ubuntu 20.04.2 LTS
#> 
#> Matrix products: default
#> BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.9.0
#> LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.9.0
#> 
#> locale:
#>  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
#>  [3] LC_TIME=es_ES.UTF-8        LC_COLLATE=C              
#>  [5] LC_MONETARY=es_ES.UTF-8    LC_MESSAGES=en_US.UTF-8   
#>  [7] LC_PAPER=es_ES.UTF-8       LC_NAME=C                 
#>  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
#> [11] LC_MEASUREMENT=es_ES.UTF-8 LC_IDENTIFICATION=C       
#> 
#> attached base packages:
#> [1] parallel  stats4    stats     graphics  grDevices utils     datasets 
#> [8] methods   base     
#> 
#> other attached packages:
#>  [1] reactome.db_1.74.0   forcats_0.5.1        ggplot2_3.3.3       
#>  [4] GO.db_3.12.1         org.Hs.eg.db_3.12.0  AnnotationDbi_1.52.0
#>  [7] IRanges_2.24.1       S4Vectors_0.28.1     Biobase_2.50.0      
#> [10] BiocGenerics_0.36.0  dplyr_1.0.5          BaseSet_0.0.16      
#> 
#> loaded via a namespace (and not attached):
#>  [1] tidyselect_1.1.0  xfun_0.22         bslib_0.2.4       purrr_0.3.4      
#>  [5] colorspace_2.0-0  vctrs_0.3.7       generics_0.1.0    htmltools_0.5.1.1
#>  [9] yaml_2.2.1        utf8_1.2.1        blob_1.2.1        XML_3.99-0.6     
#> [13] rlang_0.4.10      jquerylib_0.1.3   pillar_1.6.0      withr_2.4.2      
#> [17] glue_1.4.2        DBI_1.1.1         bit64_4.0.5       lifecycle_1.0.0  
#> [21] stringr_1.4.0     munsell_0.5.0     gtable_0.3.0      memoise_2.0.0    
#> [25] evaluate_0.14     labeling_0.4.2    knitr_1.32        fastmap_1.1.0    
#> [29] fansi_0.4.2       highr_0.9         GSEABase_1.52.1   Rcpp_1.0.6       
#> [33] xtable_1.8-4      scales_1.1.1      cachem_1.0.4      graph_1.68.0     
#> [37] annotate_1.68.0   jsonlite_1.7.2    farver_2.1.0      bit_4.0.4        
#> [41] digest_0.6.27     stringi_1.5.3     grid_4.0.1        tools_4.0.1      
#> [45] magrittr_2.0.1    sass_0.3.1        RSQLite_2.2.6     tibble_3.1.1     
#> [49] crayon_1.4.1      pkgconfig_2.0.3   ellipsis_0.3.1    assertthat_0.2.1 
#> [53] rmarkdown_2.7     httr_1.4.2        R6_2.5.0          compiler_4.0.1