data-raw/convert/0.13.0.R

In hms-dbmi/drugseqr: GUI to Explore Single-Cell and Bulk RNA-Seq from Fastq to Pathways and Perturbations

app_dir <- "/srv/dseqr/example"
app_dirs <- list.files('/srv/dseqr')
app_dirs <- setdiff(app_dirs,
                    c('indices', 'node_modules', 'pert_signature_dir',
                      'pert_query_dir', 'example_data.tar.gz'))

convert <- function(app_dir) {
  cat('==== Working on', app_dir, '====\n')

  sc_dir <- file.path(app_dir, 'single-cell')

  # make sure we are all qs
  rds_files <- list.files(app_dir, '.rds$', full.names = TRUE, recursive = TRUE)
  for (rds_file in rds_files) {
    val <- readRDS(rds_file)
    new_path <- gsub('.rds$', '.qs', rds_file)
    qs::qsave(val, new_path)
    unlink(rds_file)
  }

  integrated <- dseqr:::qread.safe(file.path(sc_dir, 'integrated.qs'))

  if (length(integrated)) {

    dataset_dirs <- file.path(sc_dir, integrated)
    deleted <- c()

    for (i in seq_along(dataset_dirs)) {
      dataset_name <- integrated[i]
      dataset_dir <- dataset_dirs[i]
      cat('working on', integrated[i], '...\n')
      scseq_path <- file.path(dataset_dir, 'scseq.qs')

      if (!file.exists(scseq_path)) {
        deleted <- c(deleted, i)
        next()
      }

      # update test/ctrl from args json
      args <- dseqr:::load_args(sc_dir, dataset_name)
      if (is.null(args$dataset_names)) {
        args$dataset_names <- c(args$test, args$ctrl)
        args$test <- args$ctrl <- NULL
        dseqr:::save_scseq_args(args, dataset_name, sc_dir)
      }

      combined <- qs::qread(scseq_path)
      dataset_names <- unique(combined$batch)

      # add ambient info for each sample
      scseqs <- dseqr:::load_scseq_subsets(dataset_names, sc_dir)
      combined <- dseqr:::add_combined_ambience(combined, scseqs)

      # remove 'test' and 'ctrl' ambient
      SummarizedExperiment::rowData(combined)$ctrl_ambient <- NULL
      SummarizedExperiment::rowData(combined)$test_ambient <- NULL

      # create meta & prev_groups from test/ctrl orig.ident
      if (all(levels(combined$orig.ident) %in% c('test', 'ctrl'))) {
        samples <- unique(combined$batch)
        test <- unique(combined$batch[combined$orig.ident == 'test'])
        ctrl <- unique(combined$batch[combined$orig.ident == 'ctrl'])

        group <- rep(NA, length(samples))
        group[samples %in% test] <- 'test'
        group[samples %in% ctrl] <- 'ctrl'

        meta <- data.frame(group,
                           pair = NA,
                           row.names = samples,
                           check.names = FALSE, stringsAsFactors = FALSE)

        prev_groups <- c('test', 'ctrl')
        qs::qsave(meta, file.path(dataset_dir, 'meta.qs'))
        qs::qsave(prev_groups, file.path(dataset_dir, 'prev_groups.qs'))
      }

      # use sample for orig.ident
      combined$orig.ident <- factor(combined$batch)


      # overwrite loom/qs objects
      qs::qsave(combined, scseq_path, preset = 'fast')

      # remove unnecessary
      unlink(file.path(dataset_dir, c('ambient.qs', 'has_replicates.qs')))
      snn_dirs <- list.files(dataset_dir, 'snn\\d', full.names = TRUE)
      unlink(file.path(snn_dirs, 'tests'), recursive = TRUE)
      unlink(file.path(snn_dirs, 'plots'), recursive = TRUE)
      unlink(file.path(snn_dirs, 'lm_fit_0svs.qs'), recursive = TRUE)
      unlink(file.path(snn_dirs, 'has_replicates.qs'), recursive = TRUE)
      unlink(file.path(snn_dirs, 'top_tables.qs'), recursive = TRUE)
      unlink(file.path(snn_dirs, 'top_markers.qs'), recursive = TRUE)
      unlink(file.path(snn_dirs, 'cluster_stats.qs'), recursive = TRUE)
    }

    # remove deleted
    if(length(deleted)) {
      integrated <- integrated[-deleted]
      qs::qsave(integrated, file.path(sc_dir, 'integrated.qs'))
    }

  }


  # change to presto markers
  # and replace looms with fast qs files
  dataset_names <- list.files(sc_dir)
  dataset_names <- dataset_names[!grepl('.qs$', dataset_names)]


  for (dataset_name in dataset_names) {
    cat('working on', dataset_name, '...\n')
    dataset_dir <- file.path(sc_dir, dataset_name)
    scseq_path <- file.path(dataset_dir, 'scseq.qs')
    if (!file.exists(scseq_path)) next()

    scseq <- qs::qread(scseq_path)
    qs::qsave(scseq, scseq_path, preset = 'fast')
    unlink(file.path(dataset_dir, 'scle.loom'))

    snn_names <- list.files(dataset_dir, '^snn\\d')

    marker_files <- list.files(dataset_dir, '^markers_\\d+.qs$', recursive = TRUE, full.names = TRUE)
    unlink(marker_files)

    for (snn_name in snn_names) {
      snn_dir <- file.path(dataset_dir, snn_name)

      # remove uneeded
      unlink(file.path(snn_dir, 'tests'), recursive = TRUE)
      unlink(list.files(snn_dir, 'l1000_.+?.qs', full.names = TRUE))
      unlink(list.files(snn_dir, 'cmap_.+?.qs', full.names = TRUE))
      unlink(list.files(snn_dir, 'kegg_.+?.qs', full.names = TRUE))
      unlink(list.files(snn_dir, 'kegga_.+?.qs', full.names = TRUE))
      unlink(list.files(snn_dir, 'go_.+?.qs', full.names = TRUE))
      unlink(list.files(snn_dir, 'goana_.+?.qs', full.names = TRUE))
      unlink(file.path(snn_dir, 'applied.qs'))
      unlink(file.path(snn_dir, 'top_markers.qs'))
      unlink(file.path(snn_dir, 'pbulk_esets.qs'))
    }
  }
}