Skip to contents

BiocRevSumm: LLM-based summaries for Bioconductor package submissions

Vincent J. Carey, stvjc at channing.harvard.edu

December 04, 2025

Source: vignettes/BiocRevSumm.Rmd
BiocRevSumm.Rmd

Introduction

This package surveys issues in the Bioconductor contributions repositories with the aim of producing a structured summary of package objectives and capabilities.

Enumerating packages in review 

library(BiocRevSumm)
stopifnot(nchar(Sys.getenv("GITHUB_TOKEN"))>0)
iss = get_repo_issues("bioconductor", "contributions", Sys.getenv("GITHUB_TOKEN"))
urls = get_issue_repo_urls(iss)
head(urls)
##   number                              title
## 1   4047                           EMTscore
## 2   4041 RFGeneRank: New Package Submission
## 3   4039                          panoramic
## 4   4038                            DenoIST
## 5   4036                        PostChicago
## 6   4033                               fraq
##                                           repo_url
## 1                https://github.com/wenmm/EMTscore
## 2 https://github.com/Abdulaziz-Albeshri/RFGeneRank
## 3       https://github.com/plevritis-lab/panoramic
## 4              https://github.com/aaronkwc/DenoIST
## 5   https://github.com/FeldmannLabDKFZ/PostChicago
## 6                 https://github.com/traversc/fraq

Finding the vignettes for high-resolution categorization 

u1 = urls$repo_url[1]
lk1 = strsplit(u1, "\\/")[[1]]
ghuser = lk1[4]
ghrepo = lk1[5]
lkvigs = list_github_dir(user = ghuser, repo = ghrepo, path="vignettes") 
vigs = grep("Rmd", lkvigs, value=TRUE)
# arbitrarily choose first
build_vignette_path = function(ghuser, ghrepo, branch, vname) {
  sprintf(
    "https://raw.githubusercontent.com/%s/%s/refs/heads/%s/vignettes/%s",
     ghuser, ghrepo, branch, vname
     )
   }
url = build_vignette_path(ghuser, ghrepo, branch="main", vname=vigs[1])
url
## [1] "https://raw.githubusercontent.com/wenmm/EMTscore/refs/heads/main/vignettes/EMTscore.Rmd"

Summarize a selected vignette 

ans = vig2data(url)
writeLines(strwrap(ans$focused, 75))

The article “EMTscore Analysis and Plots” details the utilization of various computational methods within the R package “EMTscore” to assess epithelial-mesenchymal transition (EMT) at the genetic level. EMT is characterized by the transformation of epithelial cells into mesenchymal cells, a process pivotal in cancer metastasis and tissue regeneration. By computing and visualizing EMT scores, researchers can better understand this biological transition both in bulk RNA-sequencing and single-cell RNA-sequencing data. The article explains key aspects such as:

  1. Objective of EMTscore: The package allows for the computation of scores for epithelial (E) and mesenchymal (M) gene sets as well as the option for user-defined gene sets. It provides multiple algorithms for score calculation, including nonnegative PCA (nnPCA), AUCell, ssGSEA, JASMINE, and SCSE.

  2. Analytical Techniques: By leveraging nnPCA, the article demonstrates how multiple mesenchymal phenotype scores can be identified and correlated with EMT scores and other gene sets. Additional functionalities include: - Loading and annotating cell/sample data. - Utilizing existing datasets and gene signatures, such as the Panchy_et_al signatures. - Computing gene set scores using bulk RNA-seq example walk-throughs.

  3. Visualizations and Plot Construction: - Plotting E-M score correlations to observe phenotypic variations across cells. - Dimension-specific scatter plots for mesenchymal (M) scores, highlighting potential gene divergence or cellular heterogeneity. - Creation of combined plots for comprehensive visualization of between-score relationships. - Analysis of score distributions across cell types using histograms. - Heatmaps of gene expression data.

  4. Single-Cell RNA-seq Analysis: The package extends functionality to single-cell data, indicating steps for importing, EMT scoring, and plotting dynamic gene expression changes along pseudotime. Gaussian Mixture Models (GMM) can be used for clustering EMT states and correlating these states with biological processes.

  5. Pathway and Process Integration: In assessing pathways associated with EMT, the article illustrates how pathway activity scores can be calculated and correlated with EMT scores to identify significant biological pathways. - The significant positive and negative correlations among pathways are visualized using bar charts.

  6. Additional Biological Scores: Calculation of stemness and senescence scores provide insights into cellular states like pluripotency or aging. The article explores relationships between these scores with EMT scores in single-cell data.

Overall, the article offers a systematic guide for researchers seeking to quantify and visualize EMT processes at the genomic level, with the flexibility to incorporate custom datasets and gene sets within their EMTscore analyses.

Collect summaries, and categorize them

Get paths to vignette

The first vignette found in the listing of vignettes folder with Rmd or Rnw suffix will be selected.

rs = urls$repo_url
issusers = sapply(strsplit(rs, "\\/"), "[", 4)
issrepos = sapply(strsplit(rs, "\\/"), "[", 5)
nrep = length(issusers)

vpaths = sapply(head(seq_len(nrep)), function(i) {
   br = find_existing_branch(issusers[i], issrepos[i], path="vignettes", branches = c("main", "master", "devel",
     "origin"))
   if (is.na(br)) return(NA)
   Sys.sleep(5)  # watch out for rate limit
   vd = try(list_github_dir(user = issusers[i], repo = issrepos[i], branch=br, path="vignettes"), silent=FALSE)
   if (inherits(vd, "try-error")) return(NA)
   vn = grep("Rmd|Rnw", vd, value=TRUE)[1]
   build_vignette_path(issusers[i], issrepos[i], branch=br, vname=vn)
   })
head(vpaths)
## [1] "https://raw.githubusercontent.com/wenmm/EMTscore/refs/heads/main/vignettes/EMTscore.Rmd"                       
## [2] "https://raw.githubusercontent.com/Abdulaziz-Albeshri/RFGeneRank/refs/heads/main/vignettes/RFGeneRank-intro.Rmd"
## [3] "https://raw.githubusercontent.com/plevritis-lab/panoramic/refs/heads/main/vignettes/tutorial.Rmd"              
## [4] "https://raw.githubusercontent.com/aaronkwc/DenoIST/refs/heads/master/vignettes/denoist_spe.Rmd"                
## [5] "https://raw.githubusercontent.com/FeldmannLabDKFZ/PostChicago/refs/heads/main/vignettes/PostChicago.Rmd"       
## [6] "https://raw.githubusercontent.com/traversc/fraq/refs/heads/main/vignettes/fraq_getting_started.Rmd"

Use vig2data to summarize vignette (and thus package) 

z1 = vig2data(vpaths[6])
z1$focused

Code that produced summaries for packages in review on 22 Nov 2025 

library(BiocRevSumm) 
data("viglist_23_nov_2025", package="BiocRevSumm")
url2heading = function(x) {
 stopifnot(length(x)==1)
 z = strsplit(x, "\\/")[[1]]
 sprintf('\n\n## Package %s from repo %s ([repo link](https://github.com/%s/%s))\n\n', z[5], z[4], z[4], z[5])
 }
for (i in seq_len(length(viglist_23_nov_2025))) {
 curvig = viglist_23_nov_2025[[i]]
 cat(url2heading(curvig), "\n")
 Sys.sleep(15)
 ans = vig2data( curvig )
 z = strsplit(curvig, "\\/")[[1]]
 fn = sprintf("%s_%s.rds", z[4], z[5])
 saveRDS(ans, file=fn)
 foc = ans$focused
 cat(foc)
}