In dmontemayor/Rcricvol: Data Processing Tools in R for cricvol project

data <- read.csv("../data/clustering_train_summary.csv")

What is the probability that duplicated samples are not found in the same cluster?

find_prob_cluster_dups <- function(method, clustering_summary) {
  df <- data.frame("pid" = clustering_summary$patients, "time" = as.integer(substring(clustering_summary$visit,4,5)), "cluster" = clustering_summary[grep(paste0(method, "_cluster"), names(clustering_summary))])
  mismatch <- c()
  match <- c()
  for (year in unique(df$time)) {
    df2 <- df[df$time == year,]
    for (pat in unique(df2$pid)) {
      df3 <- df2[df2$pid == pat,]
      if (length(unique(df3[,3])) > 1) {mismatch <- c(mismatch, pat)}
      else {match <- c(match, pat)}
    }
  }
  print(paste0(method, ": ", round(length(mismatch)/(length(mismatch) + length(match)), digits = 3)))
  return(mismatch)
}

pca_mismatch <- find_prob_cluster_dups("PCA", data)
tsne_mismatch <- find_prob_cluster_dups("TSNE", data)
som_mismatch <- find_prob_cluster_dups("SOM", data)

What is the probability that a patient stays in the cluster in which they started?

find_prob_of_staying_in_cluster <- function(method, clustering_summary, mismatched) {

  clustering_summary <- clustering_summary[!clustering_summary$patients %in% mismatched,]

  df <- data.frame("pid" = clustering_summary$patients, "time" = as.integer(substring(clustering_summary$visit,4,5)), "cluster" = clustering_summary[grep(paste0(method, "_cluster"), names(clustering_summary))])
  names(df)[3] <- "cluster"
  df <- df[!duplicated(df),]

  pids_with_baseline <- df[df$time == 0,]$pid
  df <- df[df$pid %in% pids_with_baseline,]

  res <- data.frame("patient" = unique(df$pid), "y1" = 0, "y2" = 0, "y3" = 0)

  for (pid in 1:nrow(res)) {
    pdf <- df[df$pid == res$patient[pid],]
    start <- pdf[pdf$time == 0,]$cluster
    for (x in c(1,2,3)) {
      if (x %in% pdf$time) {res[pid,(x+1)] <- (pdf[pdf$time == x,]$cluster == start)} 
      else {res[pid, (x+1)] <- NA}
    }
  }

  rownames(res) <- res[,1]
  res <- res[-1]

  print(paste0(method, " probability of staying in cluster: ", round(sum(res, na.rm = TRUE)/sum(!is.na(res)), digits = 3)))

  print("Number of years in starting cluster and Number of patients")
  kable(table(unlist(rowSums(res, na.rm = TRUE))))

  print(paste0("At year 1, ", sum(res$y1, na.rm = TRUE), " patients are in the same cluster as they started in (n = ", sum(!is.na(res$y1)), ")"))
  print(paste0("At year 2, ", sum(res$y2, na.rm = TRUE), " patients are in the same cluster as they started in (n = ", sum(!is.na(res$y2)), ")"))
  print(paste0("At year 3, ", sum(res$y3, na.rm = TRUE), " patients are in the same cluster as they started in (n = ", sum(!is.na(res$y3)), ")"))

  return(sum(res, na.rm = TRUE)/sum(!is.na(res)))
}


pca_stayincluster <- find_prob_of_staying_in_cluster(method = "PCA", clustering_summary = data, mismatched = pca_mismatch)
tsne_stayincluster <- find_prob_of_staying_in_cluster(method = "TSNE", clustering_summary = data, mismatched = tsne_mismatch)
som_stayincluster <- find_prob_of_staying_in_cluster(method = "SOM", clustering_summary = data, mismatched = som_mismatch)

Transition matrix normalized by dividing by row sums

make_transition_matrix <- function(method, clustering_summary, mismatched){

  clusters <- unlist(unique(clustering_summary[grep(paste0(method, "_cluster"), names(clustering_summary))]))
  clustering_summary <- clustering_summary[!clustering_summary$patients %in% mismatched,]

  df <- data.frame("pid" = clustering_summary$patients, "time" = as.integer(substring(clustering_summary$visit,4,5)), "cluster" = clustering_summary[grep(paste0(method, "_cluster"), names(clustering_summary))])
  names(df)[3] <- "cluster"
  df <- df[!duplicated(df),]


  tm <- matrix(nrow = length(clusters), ncol = length(clusters), 0)
  rownames(tm) <- sort(clusters)
  colnames(tm) <- sort(clusters)

  for (pid in unique(df$pid)) {
    pdf <- df[df$pid == pid,]
    pdf <- pdf[order(pdf$time),]
    for (row in 1:nrow(pdf)) {
      from <- pdf$cluster[row]
      to <- pdf$cluster[row + 1]
      tm[from, to] <- tm[from, to] + 1
    }
  }

  for (r in 1:nrow(tm)) {
    tm[r,] <- round((tm[r,] / sum(tm[r,])), digits = 3)
  }

  return(tm) 
}

pca_tm <- make_transition_matrix("PCA", data, pca_mismatch) 
tsne_tm <- make_transition_matrix("TSNE", data, tsne_mismatch)
som_tm <- make_transition_matrix("SOM", data, som_mismatch)

write.csv(pca_tm, "../data/pca_transition_matrix.csv", row.names = FALSE)
write.csv(tsne_tm, "../data/tsne_transition_matrix.csv", row.names = FALSE)
write.csv(som_tm, "../data/som_transition_matrix.csv", row.names = FALSE)

dmontemayor/Rcricvol documentation built on Sept. 9, 2021, 9:12 a.m.