Literature analysis

Sat, 01 Aug 2020 00:00:00 +0000

Here I discuss the various literature packages in R and show a use case on a topic I study. Available packages;

bibliometrix: has a variety of functions returning most useful metrics you could want
scholar: useful to evaluate journals and scholars (here I use it to retrieve impact factors of journals)
pubmed.mineR: allows to query PubMed in multiple ways (great option if your query results is very high, but it’s slow and you won’t get citation counts and pubtator data)
wosr: If you have an API username and password, this might be the best option to query a large amount of papers
other packages: pubmedR, RISmed and RISmed

After spending some time exploring the different options, I found that MeSH terms were often missing, and Keywords or Keywords plus were not very informative. I ended up finding the results from pubtator. This is a NCBI Deep learning algorithm that identified and categorised 5 class of words in publications (Genes, Diseases, Mutations, Chemicals, Species) from the PMID. For use cases where you have a large number of results, I provide a script at the end of the document.

Aims:

Determin the Keywords associated to the 14q32 locus evolution in time
Determin journals yielding most citations on the subject

The following data was retreived from Web of science database with the following search therms: TOPIC: ((14q32) OR (DLK1-DIO3)) AND TOPIC: (heart OR cardi) The results were exported as plain text files.

pacman::p_load(bibliometrix,
               pubmed.mineR,
               rio,
               ttutils,
               tidyverse,
               ggplot2,
               ggrepel,
               scholar,
               reshape2)

Overview of the results

f <- c("data_litterature_14q32/14q32_OR_DLK1-DIO3_AND_heartORcardio.txt")
M <- convert2df(file = f, dbsource = "wos", format = "plaintext")

## 
## Converting your wos collection into a bibliographic dataframe
## 
## Done!
## 
## 
## Generating affiliation field tag AU_UN from C1:  Done!

results <- biblioAnalysis(M, sep = ";")
plot(x = results, k = 10, pause = FALSE)

Co-occurence of Author Keywords

# Create keyword co-occurrences network
NetMatrix <- biblioNetwork(M, analysis = "co-occurrences", network = "author_keywords", sep = ";")

# Plot the network
net=networkPlot(NetMatrix, normalize="association", weighted=T, n = 30, Title = "Keyword Co-occurrences", type = "auto", size=T,edgesize = 5,labelsize=0.7)

Genes and diseases analysis

Keywords are quite uninforamtive. I find that a better option is to use gene names and diseases identified by the NCBI Deep learning algorythm pubtator.

Pubtator query

Pubator returns all identifiers merged from a PMID list. Here we want this information for each articles so we need to do this iteratively.

# pubtator_o <- list()
# for (i in 1:length(M$PM)) {
#   pubtator_o[[i]] <- pubtator_function(M$PM[i])
#   if (isInteger(i/100)) {
#     print(i)
#   }
# }
# export(pubtator_o, "data_litterature_14q32/14q32_OR_DLK1-DIO3_AND_heartORcardio_pubtator.rds")

pubtator_o <- import("data_litterature_14q32/14q32_OR_DLK1-DIO3_AND_heartORcardio_pubtator.rds")

Network analysis of genes

Text needs to be reformated to be processed with the bibliometrix package

do <- as.data.frame(do.call(rbind, lapply(pubtator_o, as.character)), stringsAsFactors = F)
colnames(do) <- names(pubtator_o[[1]])
do <- do %>% replace(.=="NULL", NA)
do <- do[do$PMID != " No Data ",]

do <- apply(do,2,function(x) gsub("c[(]", "", x))
do <- apply(do,2,function(x) gsub("[)]", "", x))
do <- as.data.frame(apply(do, 2,function(x) gsub("\"", "", x)), stringsAsFactors = F)
do <- as.data.frame(apply(do, 2,function(x) gsub(">[0-9]+","", x)), stringsAsFactors = F)
do <- as.data.frame(apply(do, 2,function(x) gsub(">MESH:[A-Z][0-9]+","", x)), stringsAsFactors = F)
do <- as.data.frame(apply(do, 2,function(x) gsub(">-","", x)), stringsAsFactors = F)
do <- as.data.frame(apply(do, 2,function(x) gsub(",",";", x)), stringsAsFactors = F)

Now we can replace Keywords with the columns for Genes and/or Diseases from Pubtator

M2 <- M # new df to replace ID
df <- merge(M, do, by.x="PM", by.y = "PMID", all.x = T)


change_ID <- function(new_ID, DF) {
  # remove duplicated gene names (some are capitalized and others not)
  DF$ID <- new_ID
  DF$ID <- gsub("NA","",DF$ID)
  t <- split(toupper(DF$ID), seq(length(DF$ID)))
  t <- lapply(t, function(x) unique(str_trim(strsplit(x, ";")[[1]])))
  t <- lapply(t, function(x) paste(x, collapse = ";"))
  DF$ID <- unlist(t)
  DF$ID <- gsub("NULL","",DF$ID)
  return(DF)
}

new_ID <- paste0(df$Genes) # replacing ID with genes from Pubtator
M2 <- change_ID(new_ID,M2)

##### These other functions may also be helpful.
# summary(results2)
# CS <- conceptualStructure(M2,field="ID", method="CA", minDegree=4, clust=5, stemming=FALSE, labelsize=10, documents=10)

# rio::export(M2, "data_litterature_14q32/14q32_lit.xlsx") # to import in shiny
# biblioshiny()

NetMatrix <- biblioNetwork(M2, analysis = "co-occurrences", network = "keywords", sep = ";")

net=networkPlot(NetMatrix, normalize="association", weighted=T, n = 30, Title = "Keyword Cooccurrences", type = "auto", size=T,edgesize = 5)

Variation of investigated genes over time

ID_evol_p <- function(DF, top_n_ID) {
  KWG_df <- KeywordGrowth(DF, Tag = "ID", sep = ";", top = top_n_ID, cdf = TRUE)
  m <- melt(KWG_df,  measure.vars = c(colnames(KWG_df)[2:(top_n_ID+1)]))
  m <- m %>% mutate(label = if_else(Year == max(Year), as.character(variable), NA_character_))
  
  labelInfo <-
    split(m, m$variable) %>%
    lapply(function(t) {
      data.frame(
        predAtMax = loess(value ~ Year, data = t) %>%
          predict(newdata = data.frame(Year = max(t$Year)))
        , max = max(t$Year)
      )}) %>%
    bind_rows
  
  labelInfo$label = levels(factor(m$variable))
  
  p <- ggplot(m, aes(x=Year, y=value, colour=variable))+
    geom_smooth(se=F) +
    geom_label_repel(data = labelInfo, 
                     aes(x = max, y = predAtMax, 
                         label = label, 
                         color = label),
                     size = 3,
                     nudge_x = 5) +
    theme_minimal() + 
    theme(legend.position = "none")+
    xlim(min(m$Year), 10+max(m$Year))+
    ylab("Cumulative frequency")
  return(p)
}

ID_evol_p(M2,30)

Network analysis of diseases

new_ID <- paste0(df$Diseases) # replacing ID with genes from Pubtator
M2 <- change_ID(new_ID,M2)

NetMatrix <- biblioNetwork(M2, analysis = "co-occurrences", network = "keywords", sep = ";")

net=networkPlot(NetMatrix, normalize="association", weighted=T, n = 50, Title = "Keyword Cooccurrences", type = "auto", size=T,edgesize = 5,labelsize=0.7)

Variation of investigated diseases over time

ID_evol_p(M2,30)

Combined genes and diseases network

new_ID <- paste0(df$Genes, ";",df$Diseases)
M2 <- change_ID(new_ID,M2)
results2 <- biblioAnalysis(M2, sep = ";")
# summary(results2)
# rio::export(M2, "data_litterature_14q32/14q32_lit.xlsx")

NetMatrix <- biblioNetwork(M2, analysis = "co-occurrences", network = "keywords", sep = ";")

net=networkPlot(NetMatrix, normalize="association", weighted=T, n = 50, Title = "Keyword Cooccurrences", type = "auto", size=T,edgesize = 5,labelsize=0.7)

library("wordcloud")
set.seed(1234)
d <- as.data.frame(results2$ID)
wordcloud(words = d$Tab, freq = d$Freq, min.freq = 2,
          max.words=200, random.order=FALSE, rot.per=0.35, 
          colors=brewer.pal(8, "Dark2"))

Best journal in topic

number of citation per paper per year
compare to their impact factor

sources <- data.frame(sources = M2$SO , TCperYear = results2$TCperYear)
sources <- sources %>% group_by(sources) %>% summarise(TCpY = sum(TCperYear))
sources$n_aticles <- M2 %>% group_by(SO) %>% count() %>% pull(n)
sources$TCpYpA <- sources$TCpY/sources$n_aticles


impact <- get_impactfactor(journals=sources$sources, max.distance = 0.1)
sources <- merge(sources, impact, by.x="sources", by.y = "Journal", all.x = T)
sources <- sources[!duplicated(sources$sources),]
top_J <- head(sources[order(sources$TCpYpA, decreasing = T),], 15)

ggplot(top_J, aes(y=reorder(sources,TCpYpA),x=TCpYpA))+
  geom_bar(stat = "identity") +
  theme_minimal()+
  theme(legend.position = "none")+
  xlab("Citations/(Years*Articles)")+
  ylab("Journal")

ggplot(top_J, aes(x = ImpactFactor, y=TCpYpA, label=sources, colour=sources))+
  geom_point() +
  geom_label_repel(size = 2.3)+
  scale_y_log10() +
  scale_x_log10() +
  theme_minimal() +
  theme(legend.position = "none") +
  ylab("Citations/(Years*Articles)")

Large queries analyses

##### Function to get citation count from PMID 
##### (modified from RISmed package to avoid errors from multiple queries)
f <- function(id){
  
  base <- "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/elink.fcgi?retmode=xml&dbfrom=pubmed&id=ID&cmd=neighbor"
  base <- sub("ID", id, base)
  
  the_url <- url(description = base)
  on.exit(close(the_url))
  
  lines <- readLines(the_url)
  
  citedstart <- grep("pubmed_pubmed_citedin", lines)
  
  if(length(citedstart) == 0){
    return(0)
  }
  else{
    citedend <- grep("pubmed_pubmed", lines)
    citedend <- min(citedend[citedend > citedstart])
    
    tags <- grep("<Id>", lines)
    tags <- tags[tags > citedstart & tags < citedend]
    
    if(any(tags)){
      hits <- sub("(.*<ID>)([0-9]+)(.*)","\\2",lines[tags])
      hits <- unique(hits)
      length(hits[hits != id])
    }
    else{
      0
    }   
  }
}

# get 
library(pubmedR)
query <- "atrial fibrillation[Title/Abstract]"
res <- pmQueryTotalCount(query = query, api_key = NULL)
D <- pmApiRequest(query = query, limit = 100000, api_key = NULL)
M <- pmApi2df(D)


library(ttutils)
v <- c()
for (i in 1:length(M$PMID)) {
  v <- c(v,f(M$PMID[i])) # get citations 1 by one with sleep to avoid errors
  Sys.sleep(0.3)
  if (isInteger(i/100)) { # follow the long process
    print(i)
  }
}

# get Genes, Diseases, Mutations, Chemicals, Species from pubtator
library(pubmed.mineR)
pubtator_o <- list()
for (i in 1:length(M$PMID[1:100])) {
  pubtator_o[[i]] <- pubtator_function(M$PMID[i])
  if (isInteger(i/100)) {
    print(i)
  }
}

sessionInfo()

## R version 3.6.1 (2019-07-05)
## Platform: x86_64-w64-mingw32/x64 (64-bit)
## Running under: Windows 7 x64 (build 7601) Service Pack 1
## 
## Matrix products: default
## 
## locale:
## [1] LC_COLLATE=French_Canada.1252  LC_CTYPE=French_Canada.1252   
## [3] LC_MONETARY=French_Canada.1252 LC_NUMERIC=C                  
## [5] LC_TIME=French_Canada.1252    
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
##  [1] wordcloud_2.6       RColorBrewer_1.1-2  reshape2_1.4.4     
##  [4] scholar_0.1.7       ggrepel_0.8.2       forcats_0.5.0      
##  [7] stringr_1.4.0       dplyr_1.0.0         purrr_0.3.4        
## [10] readr_1.3.1         tidyr_1.0.2         tibble_3.0.0       
## [13] ggplot2_3.3.1       tidyverse_1.3.0     ttutils_1.0-1      
## [16] rio_0.5.16          pubmed.mineR_1.0.16 bibliometrix_3.0.2 
## 
## loaded via a namespace (and not attached):
##   [1] colorspace_1.4-1     ellipsis_0.3.1       R2HTML_2.3.2        
##   [4] fs_1.4.1             rstudioapi_0.11      dimensionsR_0.0.1   
##   [7] farver_2.0.3         graphlayouts_0.7.0   rscopus_0.6.6       
##  [10] SnowballC_0.7.0      DT_0.13              fansi_0.4.1         
##  [13] lubridate_1.7.8      xml2_1.3.0           splines_3.6.1       
##  [16] codetools_0.2-16     R.methodsS3_1.8.0    leaps_3.1           
##  [19] knitr_1.28           shinythemes_1.1.2    polyclip_1.10-0     
##  [22] jsonlite_1.6.1       broom_0.5.6          cluster_2.1.0       
##  [25] dbplyr_1.4.4         R.oo_1.23.0          ggforce_0.3.1       
##  [28] shiny_1.4.0.2        rentrez_1.2.2        compiler_3.6.1      
##  [31] httr_1.4.1           backports_1.1.6      assertthat_0.2.1    
##  [34] Matrix_1.2-18        fastmap_1.0.1        cli_2.0.2           
##  [37] later_1.0.0          tweenr_1.0.1         htmltools_0.4.0     
##  [40] tools_3.6.1          igraph_1.2.5         gtable_0.3.0        
##  [43] glue_1.4.1           FactoMineR_2.3       Rcpp_1.0.4.6        
##  [46] cellranger_1.1.0     vctrs_0.3.0          nlme_3.1-147        
##  [49] blogdown_0.20        ggraph_2.0.2         xfun_0.14           
##  [52] networkD3_0.4        openxlsx_4.1.4       rvest_0.3.5         
##  [55] mime_0.9             lifecycle_0.2.0      pacman_0.5.1        
##  [58] shinycssloaders_0.3  XML_3.99-0.3         stringdist_0.9.6    
##  [61] factoextra_1.0.7     MASS_7.3-51.5        scales_1.1.1        
##  [64] tidygraph_1.1.2      hms_0.5.3            promises_1.1.0      
##  [67] parallel_3.6.1       yaml_2.2.1           curl_4.3            
##  [70] gridExtra_2.3        stringi_1.4.6        boot_1.3-25         
##  [73] zip_2.0.4            rlang_0.4.6          pkgconfig_2.0.3     
##  [76] bitops_1.0-6         evaluate_0.14        lattice_0.20-40     
##  [79] labeling_0.3         htmlwidgets_1.5.1    tidyselect_1.1.0    
##  [82] plyr_1.8.6           magrittr_1.5         bookdown_0.19       
##  [85] R6_2.4.1             generics_0.0.2       DBI_1.1.0           
##  [88] mgcv_1.8-31          withr_2.2.0          pillar_1.4.4        
##  [91] haven_2.2.0          foreign_0.8-76       scatterplot3d_0.3-41
##  [94] RCurl_1.98-1.1       modelr_0.1.8         crayon_1.3.4        
##  [97] rmarkdown_2.2        viridis_0.5.1        grid_3.6.1          
## [100] readxl_1.3.1         data.table_1.12.8    blob_1.2.1          
## [103] reprex_0.3.0         digest_0.6.25        flashClust_1.01-2   
## [106] R.cache_0.14.0       xtable_1.8-4         httpuv_1.5.2        
## [109] R.utils_2.9.2        munsell_0.5.0        viridisLite_0.3.0   
## [112] pubmedR_0.0.3

R | F. Leblanc