In this analysis we seek to find Gene Ontology terms which may be overrepresented in a "target set" of abstracts, such as the results of a PubMed query.

We first fetch all the results of a specific query from Pubmed using the RISmed package and store their abstracts in a data.frame.

```{R, eval = F} library(RISmed)

Store the input string for reuse

search_topic <- "anaphylaxis genetics" search_query <- EUtilsSummary(search_topic, mindate=2014, maxdate=2015)

summary(search_query)

pull <- EUtilsGet(search_query)

data <- data.frame('Abstracts' = AbstractText(pull))

Get rid of first entry for some reason, seems to always be blank

data[,1] <- as.character(data[,1]) data <- data[-1,]

head(data)

We want to compare the terms found here to something, so we grab all abstracts from 2014 to 2015 which match a similar field, i.e. immunology genetics.  Note that only 1000 records are taken by default.

```{R, eval = F}
# Store the input string for reuse
search_topic <- "immunology genetics"
search_query <- EUtilsSummary(search_topic, # Find all articles matching the string
                              mindate=2014, # From 2014
                              maxdate=2015, # to 2015
                              retmax = 1000)  # This is the default but explicit

summary(search_query)

pull_control <- EUtilsGet(search_query)

control <- data.frame('Abstracts' = AbstractText(pull_control))

# Get rid of first entry for some reason, seems to always be blank
control[,1] <- as.character(control[,1])
control <- control[-1,]

head(control)

We now run goldi on each of the entries in both the target group and the control group.

```{R, eval = F} data(package = "goldi", "TDM.go.df") TDM.go.df <- TDM.go.df[, !duplicated(colnames(TDM.go.df))]

lims <- c(1,2,2,3,4,5,6,7,7,8,10)

results <- list()

for(i in 1:length(data)){

if(!data[i] == ""){
  results[[i]] <- goldi(doc = data[i],
        terms = terms,
        lims = lims,
        syn = F,
        object = T,
        log = "/dev/null",
        reader = "local",
        output = "/dev/null",
        term_tdm = TDM.go.df)
}

}

results <- do.call("rbind", results)

control_results <- list()

for(i in 1:length(control)){ if(!control[i] == ""){ control_results[[i]] <- goldi(doc = control[i], terms = terms, lims = lims, syn = F, object = T, log = "/dev/null", reader = "local", output = "/dev/null", term_tdm = TDM.go.df) } } control_results <- do.call("rbind", control_results)

This gives us two objects holding the results, `results` and `control_results`.  We summarize the results and take all the terms in the result set with more than two occurances to minimize spurious hits. We use the method employed by GOrilla to calculate the enrichment of terms in the target set, and limit it to those which have been identified more than five times. We calculate $P$ values using the hypergeometric distribution.

```{R, eval = F}
goldi::enrichment(target = results,
                  control = control_results
                  threshold = 5)

         Term               Enrichment      P

protein_C_(activated)_activity 66.55 1.257e-20

CD27_receptor_activity          66      2.614e-21

CD40_receptor_activity          66      2.614e-21

receptor_activator_activity 66 2.614e-21

  receptor_activity             66      2.614e-21

     IgE_binding              65.34     8.639e-20

kinase_activator_activity 65.34 8.639e-20

   kinase_activity            65.34     8.639e-20

B_cell_receptor_activity 62.23 5.11e-14

T_cell_receptor_activity 62.23 5.11e-14

This replicates the analysis presented in our prepublication. The terms above can be easily changed around, and any bunch of strings may be used for comparrison.



Chris1221/GOldilocks documentation built on July 6, 2017, 6:31 a.m.