dmontemayor/Rcricvol
Data Processing Tools in R for cricvol project

Package index
Search the dmontemayor/Rcricvol package
Vignettes
Functions
2
Source code
10
Man pages
1
Home
/
GitHub
/
dmontemayor/Rcricvol
/

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

Visualize differences between clusters with forest plots and box plots comparing MSEs of predictive models

packages

library(ggforestplot)
library(dplyr)
library(Rtsne)
library(ggplot2)
library(ggpubr)
library(randomForest)
library(knitr)
knitr::opts_chunk$set(fig.width=12, fig.height=8)

functions

plot_cluster=function(data, var_cluster, point_size) {
  var_cluster <- factor(var_cluster)
  ggplot(data, aes_string(x="V1", y="V2", color=var_cluster)) +
  geom_point(size=point_size) +
  guides(colour=guide_legend(override.aes=list(size=6))) +
  xlab("") + ylab("") +
  ggtitle("") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank(),
        legend.direction = "horizontal", 
        legend.position = "bottom",
        legend.box = "horizontal")
}


plot_forest <- function(forest_data, title, xlab) {
  forestplot(
    df = forest_data,
    estimate = beta,
    se = se,
    pvalue = pval,
    logodds = FALSE,
    colour = trait,
    title = title,
    xlab = xlab,
  )
}

format data to make a forest plot in: out: input for plot_forest

make_forest_data <- function(c_sum, ions, method) {

  data <- cbind(c_sum, ions)

  test <- data.frame("cluster" = data[grep(paste0(method, "_cluster"), names(data))], data[grep("V", names(data))])
  names(test)[1] <- "cluster"

  # need a data frame of name, trait, beta, se, pval
  my_df_linear <- data.frame("name" = 0, "trait" = 0, "beta" = 0, "se" = 0, "pval" = 0)

  for (ion in names(test[,grep("V", names(test))])) {
    for (cluster in unique(test$cluster)) {
      df <- test
      df[1] <- as.numeric(df$cluster==cluster)
      ion_data <- test[,grep(paste0("^", ion, "$"), names(test))]
      model <- glm(df$cluster ~ ion_data)

      entry <- data.frame("name" = ion, "trait" = paste0("c", cluster), 
                          "beta" = coef(summary(model))[2, "Estimate"], 
                          "se" = coef(summary(model))[2, "Std. Error"], 
                          "pval" = signif(coef(summary(model)))[2, "Pr(>|t|)"])
      my_df_linear <- rbind(my_df_linear, entry) 

    }
  }

  return(my_df_linear[my_df_linear$name != 0,])

}

plot_compare_models <- function(linear_clin, linear_other, rf_clin, rf_other, description_other, predicting){

  combined_data <- data.frame("model" = c(rep("Clin: All Patients", length(linear_clin)),
                                    rep(paste0("Clin: ", description_other), length(linear_other)),
                                    rep("Clin and Coords: All Patients", length(rf_clin)),
                                    rep(paste0("Clin and Coords: ", description_other), length(rf_other))),
                        "error" = c(linear_clin, linear_other, rf_clin, rf_other)
  )

  combined_data$model <- factor(combined_data$model, levels=c("Clin: All Patients", paste0("Clin: ", description_other), "Clin and Coords: All Patients", paste0("Clin and Coords: ", description_other)))
  compare_means(error~model, data = combined_data)
  my_comparisons <- list(c("Clin: All Patients", paste0("Clin: ", description_other)), c("Clin and Coords: All Patients", paste0("Clin and Coords: ", description_other)))

  ggplot(combined_data, aes(x = model, y = error))+ 
    geom_boxplot(outlier.shape = NA) +
    stat_compare_means(comparisons = my_comparisons, method = "t.test", label.y = 0.02, tip.length = 0.001) + 
    ylim(0, 0.025)+
    ggtitle(paste0("MSE for predicting ", predicting))

}

load data

# use untargeted, transformed data  
data <- read.csv("../data/zscore_untargeted_annotatedions.csv")

train <- data[data$group <=2,]
pats <- train$patientid
train <- train[c(grep("V", names(train)))]
val <- data[data$group == 3,]
val_pats <- val$patientid
val <- val[c(grep("V", names(val)))]


# clustering data
cluster_train_sum <- read.csv("../data/clustering_train_summary.csv")
cluster_train_sum_filtered <- cluster_train_sum[cluster_train_sum$visit == "V3Y0",]
cluster_train_sum_filtered <- cluster_train_sum_filtered[!duplicated(cluster_train_sum_filtered$patients),]
cluster_val_sum <- read.csv("../data/clustering_val_summary.csv")
cluster_val_sum_filtered <- cluster_val_sum[cluster_val_sum$visit == "V3Y0",]
cluster_val_sum_filtered <- cluster_val_sum_filtered[!duplicated(cluster_val_sum_filtered$patients),]

# clusters
pca_train_clusters <- read.csv("../data/pca_train_clusters.csv")
pca_val_clusters <- read.csv("../data/pca_val_clusters.csv")
tsne_train_clusters <- read.csv("../data/tsne_train_clusters.csv")
tsne_val_clusters <- read.csv("../data/tsne_val_clusters.csv")
som_train_clusters <- read.csv("../data/som_train_clusters.csv")

try plotting outcomes onto metabolically determined clusters; no trends

PCA

rate_on_pca <- merge(pca_train_clusters, cluster_train_sum, by = "patients")

plot_cluster(rate_on_pca[c(2,3)], rate_on_pca$rawdata.r_group, 5)
plot_cluster(rate_on_pca[c(2,3)], rate_on_pca$rawdata.v_group, 5)

TSNE

rate_on_tsne <- merge(tsne_train_clusters, cluster_train_sum, by = "patients")

plot_cluster(rate_on_tsne[c(2,3)], rate_on_tsne$rawdata.r_group, 5)
plot_cluster(rate_on_tsne[c(2,3)], rate_on_tsne$rawdata.v_group, 5)

SOM

rate_on_som <- merge(som_clusters, cluster_sum, by = "patients")

plot_cluster(rate_on_som[c(2,3)], rate_on_som$rawdata.r_group, 5)
plot_cluster(rate_on_som[c(2,3)], rate_on_som$rawdata.v_group, 5)

Forest plot of metabolite associations with clusters for all years

forest_pca_train <- make_forest_data(c_sum = cluster_train_sum, ions = train, method = "PCA")
forest_tsne_train <- make_forest_data(c_sum = cluster_train_sum, ions = train, method = "TSNE")
forest_som_train <- make_forest_data(c_sum = cluster_train_sum, ions = train, method = "SOM")

plot_forest(forest_data = forest_pca_train, 
            title = "PCA cluster associations with annotated metabolites", xlab = "1-SD increments in metabolite concentration")

plot_forest(forest_data = forest_tsne_train, 
            title = "TSNE cluster associations with annotated metabolites", xlab = "1-SD increments in metabolite concentration")

plot_forest(forest_data = forest_som_train, 
            title = "SOM cluster associations with annotated metabolites", xlab = "1-SD increments in metabolite concentration")

Forest plot of baseline clinical variable associations with groups (will also save summary statistics for each cluster: mean, sd, n)

methods <- c("PCA", "TSNE", "SOM")

for (method in methods) {

  # create data frame of name, trait, beta, se, pval
  forest <- data.frame("name" = 0, "trait" = 0, "beta" = 0, "se" = 0, "pval" = 0)
  # create data frame of clin_var, mean, sd, n
  stats <- data.frame("clin_var" = 0, "cluster" = 0, "mean" = 0, "sd" = 0, "n" = 0)

  clust_vec <- cluster_train_sum_filtered[grep(paste0(method, "_cluster"), names(cluster_train_sum_filtered))]
  clusters <- unlist(unique(clust_vec))
  clinicals <- names(cluster_train_sum_filtered)[grep("v3y0", names(cluster_train_sum_filtered))]

  for (cluster in clusters) {

    for (clin_var in clinicals) {

      clin_vec <- unlist(cluster_train_sum_filtered[grep(paste0("^", clin_var, "$"), names(cluster_train_sum_filtered))])

      model <- glm(as.numeric(clust_vec == cluster) ~ clin_vec)
      forest_entry <- data.frame("name" = clin_var, "trait" = paste0("c", cluster), 
                                 "beta" = coef(summary(model))[2, "Estimate"], 
                                 "se" = coef(summary(model))[2, "Std. Error"], 
                                 "pval" = signif(coef(summary(model)))[2, "Pr(>|t|)"])
      stats_entry <- data.frame("clin_var" = clin_var, "cluster" = paste0("c", cluster), "mean" = mean(clin_vec), "sd" = sd(clin_vec), "n" = sum(as.numeric(clust_vec == cluster)))
      forest <- rbind(forest, forest_entry)
      stats <- rbind(stats, stats_entry)

    }
  }

  # save name, trait, beta, se, pval df
  write.csv(forest[-1,], paste0("../data/forest_", method, "_train.csv"), row.names = FALSE)
  # save clin_var, mean, sd, n df
  write.csv(stats[-1,], paste0("../data/stats_", method, "_train.csv"), row.names = FALSE)

}


pca_forest <- read.csv("../data/forest_PCA_train.csv")
plot_forest(pca_forest, "PCA Clinical Variable associations", "1-SD increments in clinical variable concentration")

tsne_forest <- read.csv("../data/forest_TSNE_train.csv")
plot_forest(tsne_forest, "TSNE Clinical Variable associations", "1-SD increments in clinical variable concentration")

som_forest <- read.csv("../data/forest_SOM_train.csv")
plot_forest(som_forest, "SOM Clinical Variable associations", "1-SD increments in clinical variable concentration")

Predicting EGFR rate using all clinical variables

PCA

pca_df <- data.frame(cluster_train_sum_filtered[grep("PCA", names(cluster_train_sum_filtered))], 
                      "egfr_rate" = cluster_train_sum_filtered$rawdata.mu_egfr, 
                      cluster_train_sum_filtered[grep("v3y0", names(cluster_train_sum_filtered))])

low_pca <- pca_df[pca_df$PCA_cluster %in% c(6),]
high_pca <- pca_df[pca_df$PCA_cluster %in% c(8),]
med_pca <- pca_df[!pca_df$PCA_cluster %in% c(6,8),]

pca_val_df <- data.frame(cluster_val_sum_filtered[grep("PCA", names(cluster_val_sum_filtered))], 
                      "egfr_rate" = cluster_val_sum_filtered$rawdata.mu_egfr, 
                      cluster_val_sum_filtered[grep("v3y0", names(cluster_val_sum_filtered))])

low_pca_val <- pca_val_df[pca_val_df$PCA_cluster %in% c(6),]
high_pca_val <- pca_val_df[pca_val_df$PCA_cluster %in% c(8),]
med_pca_val <- pca_val_df[!pca_val_df$PCA_cluster %in% c(6,8),]

LINEAR

############# ONLY CLIN

# all clin, all pats
model <- lm(egfr_rate ~ . - PCA_cluster - PCA_X - PCA_Y, data = pca_df)
lm_pca_clin_all <- (model$fitted.values - pca_df$egfr_rate)^2

# all clin, all pats (VALIDATE)
val_model <- predict(model, pca_val_df)
lm_pca_val_clin_all <- (val_model - pca_val_df$egfr_rate)^2

# all clin, low pats
model <- lm(egfr_rate ~ . - PCA_cluster - PCA_X - PCA_Y, data = low_pca)
lm_pca_clin_low <- (model$fitted.values - low_pca$egfr_rate)^2

# all clin, med pats
model <- lm(egfr_rate ~ . - PCA_cluster - PCA_X - PCA_Y, data = med_pca)
lm_pca_clin_med <- (model$fitted.values - med_pca$egfr_rate)^2

# all clin, high pats
model <- lm(egfr_rate ~ . - PCA_cluster - PCA_X - PCA_Y, data = high_pca)
lm_pca_clin_high <- (model$fitted.values - high_pca$egfr_rate)^2

# all clin, high pats (VALIDATE)
val_model <- predict(model, high_pca_val) 
lm_pca_val_clin_high <- (val_model - high_pca_val$egfr_rate)^2

############## INCLUDE COORDS

# all clin + coords, all pats
model <- lm(egfr_rate ~ . - PCA_cluster, data = pca_df)
lm_pca_coords_all <- (model$fitted.values - pca_df$egfr_rate)^2

# all clin + coords, all pats (VALIDATE)
val_model <- predict(model, pca_val_df)
lm_pca_val_coords_all <- (val_model - pca_val_df$egfr_rate)^2

# all clin + coords, low pats
model <- lm(egfr_rate ~ . - PCA_cluster, data = low_pca)
lm_pca_coords_low <- (model$fitted.values - low_pca$egfr_rate)^2

# all clin + coords, med pats
model <- lm(egfr_rate ~ . - PCA_cluster, data = med_pca)
lm_pca_coords_med <- (model$fitted.values - med_pca$egfr_rate)^2

# all clin + coords, high pats
model <- lm(egfr_rate ~ . - PCA_cluster, data = high_pca)
lm_pca_coords_high <- (model$fitted.values - high_pca$egfr_rate)^2

# clin + coords, high pats (VALIDATE)
val_model <- predict(model, high_pca_val)
lm_pca_val_coords_high <- (val_model - high_pca_val$egfr_rate)^2

################ PLOT

plot_compare_models(linear_clin = lm_pca_clin_all, linear_other = lm_pca_clin_low, rf_clin = lm_pca_coords_all, rf_other = lm_pca_coords_low, description_other = "LOW group", predicting = paste0("EGFR Rate with Linear Models (LOW PCA group, n = ", length(lm_pca_clin_low), ")"))

plot_compare_models(linear_clin = lm_pca_clin_all, linear_other = lm_pca_clin_med, rf_clin = lm_pca_coords_all, rf_other = lm_pca_coords_med, description_other = "MED group", predicting = paste0("EGFR Rate with Linear Models (MED PCA group, n = ", length(lm_pca_clin_med), ")"))

plot_compare_models(linear_clin = lm_pca_clin_all, linear_other = lm_pca_clin_high, rf_clin = lm_pca_coords_all, rf_other = lm_pca_coords_high, description_other = "HIGH group", predicting = paste0("EGFR Rate with Linear Models (HIGH PCA group, n = ", length(lm_pca_clin_high), ")"))

plot_compare_models(linear_clin = lm_pca_val_clin_all, linear_other = lm_pca_val_clin_high, rf_clin = lm_pca_val_coords_all, rf_other = lm_pca_val_coords_high, description_other = "HIGH group", predicting = paste0("EGFR Rate w/LM using VALIDATION set (HIGH PCA group, n = ", length(lm_pca_val_clin_high), ")"))

RF

# ONLY CLIN

# all clin, all pats
model <- randomForest(egfr_rate ~ . - PCA_cluster - PCA_X - PCA_Y, data = pca_df)
rf_pca_clin_all <- (model$predicted - pca_df$egfr_rate)^2

# all clin, low pats
model <- randomForest(egfr_rate ~ . - PCA_cluster - PCA_X - PCA_Y, data = low_pca)
rf_pca_clin_low <- (model$predicted - low_pca$egfr_rate)^2

# all clin, med pats
model <- randomForest(egfr_rate ~ . - PCA_cluster - PCA_X - PCA_Y, data = med_pca)
rf_pca_clin_med <- (model$predicted - med_pca$egfr_rate)^2

# all clin, high pats
model <- randomForest(egfr_rate ~ . - PCA_cluster - PCA_X - PCA_Y, data = high_pca)
rf_pca_clin_high <- (model$predicted - high_pca$egfr_rate)^2

# INCLUDE COORDS

# all clin + coords, all pats
model <- randomForest(egfr_rate ~ . - PCA_cluster, data = pca_df)
rf_pca_coords_all <- (model$predicted - pca_df$egfr_rate)^2

# all clin + coords, low pats
model <- randomForest(egfr_rate ~ . - PCA_cluster, data = low_pca)
rf_pca_coords_low <- (model$predicted - low_pca$egfr_rate)^2

# all clin + coords, med pats
model <- randomForest(egfr_rate ~ . - PCA_cluster, data = med_pca)
rf_pca_coords_med <- (model$predicted - med_pca$egfr_rate)^2

# all clin + coords, high pats
model <- randomForest(egfr_rate ~ . - PCA_cluster, data = high_pca)
rf_pca_coords_high <- (model$predicted - high_pca$egfr_rate)^2

# PLOT

plot_compare_models(linear_clin = rf_pca_clin_all, linear_other = rf_pca_clin_low, rf_clin = rf_pca_coords_all, rf_other = rf_pca_coords_low, description_other = "LOW group", predicting = paste0("EGFR Rate with RF Models (LOW PCA group, n = ", length(rf_pca_clin_low), ")"))

plot_compare_models(linear_clin = rf_pca_clin_all, linear_other = rf_pca_clin_med, rf_clin = rf_pca_coords_all, rf_other = rf_pca_coords_med, description_other = "MED group", predicting = paste0("EGFR Rate with RF Models (MED PCA group, n = ", length(rf_pca_clin_med), ")"))

plot_compare_models(linear_clin = rf_pca_clin_all, linear_other = rf_pca_clin_high, rf_clin = rf_pca_coords_all, rf_other = rf_pca_coords_high, description_other = "HIGH group", predicting = paste0("EGFR Rate with RF Models (HIGH PCA group, n = ", length(rf_pca_clin_high), ")"))

TSNE

tsne_df <- data.frame(cluster_train_sum_filtered[grep("TSNE", names(cluster_train_sum_filtered))], 
                      "egfr_rate" = cluster_train_sum_filtered$rawdata.mu_egfr, 
                      cluster_train_sum_filtered[grep("v3y0", names(cluster_train_sum_filtered))])

low_tsne <- tsne_df[tsne_df$TSNE_cluster %in% c(2),]
high_tsne <- tsne_df[tsne_df$TSNE_cluster %in% c(3),]
med_tsne <- tsne_df[!tsne_df$TSNE_cluster %in% c(2,3),]

tsne_val_df <- data.frame(cluster_val_sum_filtered[grep("TSNE", names(cluster_val_sum_filtered))], 
                      "egfr_rate" = cluster_val_sum_filtered$rawdata.mu_egfr, 
                      cluster_val_sum_filtered[grep("v3y0", names(cluster_val_sum_filtered))])

low_tsne_val <- tsne_val_df[tsne_val_df$TSNE_cluster %in% c(6),]
high_tsne_val <- tsne_val_df[tsne_val_df$TSNE_cluster %in% c(8),]
med_tsne_val <- tsne_val_df[!tsne_val_df$TSNE_cluster %in% c(6,8),]

LINEAR

# ONLY CLIN

# all clin, all pats
model <- lm(egfr_rate ~ . - TSNE_cluster - TSNE_X - TSNE_Y, data = tsne_df)
lm_tsne_clin_all <- (model$fitted.values - tsne_df$egfr_rate)^2

# all clin, all pats (VALIDATE)
val_model <- predict(model, tsne_val_df)
lm_tsne_val_clin_all <- (val_model - tsne_val_df$egfr_rate)^2

# all clin, low pats
model <- lm(egfr_rate ~ . - TSNE_cluster - TSNE_X - TSNE_Y, data = low_tsne)
lm_tsne_clin_low <- (model$fitted.values - low_tsne$egfr_rate)^2

# all clin, med pats
model <- lm(egfr_rate ~ . - TSNE_cluster - TSNE_X - TSNE_Y, data = med_tsne)
lm_tsne_clin_med <- (model$fitted.values - med_tsne$egfr_rate)^2

# all clin, high pats
model <- lm(egfr_rate ~ . - TSNE_cluster - TSNE_X - TSNE_Y, data = high_tsne)
lm_tsne_clin_high <- (model$fitted.values - high_tsne$egfr_rate)^2

# all clin, high pats (VALIDATE)
val_model <- predict(model, high_tsne_val) 
lm_tsne_val_clin_high <- (val_model - high_tsne_val$egfr_rate)^2

# INCLUDE COORDS

# all clin + coords, all pats
model <- lm(egfr_rate ~ . - TSNE_cluster, data = tsne_df)
lm_tsne_coords_all <- (model$fitted.values - tsne_df$egfr_rate)^2

# all clin + coords, all pats (VALIDATE)
val_model <- predict(model, tsne_val_df)
lm_tsne_val_coords_all <- (val_model - tsne_val_df$egfr_rate)^2

# all clin + coords, low pats
model <- lm(egfr_rate ~ . - TSNE_cluster, data = low_tsne)
lm_tsne_coords_low <- (model$fitted.values - low_tsne$egfr_rate)^2

# all clin + coords, med pats
model <- lm(egfr_rate ~ . - TSNE_cluster, data = med_tsne)
lm_tsne_coords_med <- (model$fitted.values - med_tsne$egfr_rate)^2

# all clin + coords, high pats
model <- lm(egfr_rate ~ . - TSNE_cluster, data = high_tsne)
lm_tsne_coords_high <- (model$fitted.values - high_tsne$egfr_rate)^2

# clin + coords, high pats (VALIDATE)
val_model <- predict(model, high_tsne_val)
lm_tsne_val_coords_high <- (val_model - high_tsne_val$egfr_rate)^2

# PLOT

plot_compare_models(linear_clin = lm_tsne_clin_all, linear_other = lm_tsne_clin_low, rf_clin = lm_tsne_coords_all, rf_other = lm_tsne_coords_low, description_other = "LOW group", predicting = paste0("EGFR Rate with Linear Models (LOW TSNE group, n = ", length(lm_tsne_clin_low), ")"))

plot_compare_models(linear_clin = lm_tsne_clin_all, linear_other = lm_tsne_clin_med, rf_clin = lm_tsne_coords_all, rf_other = lm_tsne_coords_med, description_other = "MED group", predicting = paste0("EGFR Rate with Linear Models (MED TSNE group, n = ", length(lm_tsne_clin_med), ")"))

plot_compare_models(linear_clin = lm_tsne_clin_all, linear_other = lm_tsne_clin_high, rf_clin = lm_tsne_coords_all, rf_other = lm_tsne_coords_high, description_other = "HIGH group", predicting = paste0("EGFR Rate with Linear Models (HIGH TSNE group, n = ", length(lm_tsne_clin_high), ")"))

plot_compare_models(linear_clin = lm_tsne_val_clin_all, linear_other = lm_tsne_val_clin_high, rf_clin = lm_tsne_val_coords_all, rf_other = lm_tsne_val_coords_high, description_other = "HIGH group", predicting = paste0("EGFR Rate w/LM using VALIDATION set (HIGH TSNE group, n = ", length(lm_tsne_val_clin_high), ")"))

RF

# ONLY CLIN

# all clin, all pats
model <- randomForest(egfr_rate ~ . - TSNE_cluster - TSNE_X - TSNE_Y, data = tsne_df)
rf_tsne_clin_all <- (model$predicted - tsne_df$egfr_rate)^2

# all clin, low pats
model <- randomForest(egfr_rate ~ . - TSNE_cluster - TSNE_X - TSNE_Y, data = low_tsne)
rf_tsne_clin_low <- (model$predicted - low_tsne$egfr_rate)^2

# all clin, med pats
model <- randomForest(egfr_rate ~ . - TSNE_cluster - TSNE_X - TSNE_Y, data = med_tsne)
rf_tsne_clin_med <- (model$predicted - med_tsne$egfr_rate)^2

# all clin, high pats
model <- randomForest(egfr_rate ~ . - TSNE_cluster - TSNE_X - TSNE_Y, data = high_tsne)
rf_tsne_clin_high <- (model$predicted - high_tsne$egfr_rate)^2

# INCLUDE COORDS

# all clin + coords, all pats
model <- randomForest(egfr_rate ~ . - TSNE_cluster, data = tsne_df)
rf_tsne_coords_all <- (model$predicted - tsne_df$egfr_rate)^2

# all clin + coords, low pats
model <- randomForest(egfr_rate ~ . - TSNE_cluster, data = low_tsne)
rf_tsne_coords_low <- (model$predicted - low_tsne$egfr_rate)^2

# all clin + coords, med pats
model <- randomForest(egfr_rate ~ . - TSNE_cluster, data = med_tsne)
rf_tsne_coords_med <- (model$predicted - med_tsne$egfr_rate)^2

# all clin + coords, high pats
model <- randomForest(egfr_rate ~ . - TSNE_cluster, data = high_tsne)
rf_tsne_coords_high <- (model$predicted - high_tsne$egfr_rate)^2

# PLOT

plot_compare_models(linear_clin = rf_tsne_clin_all, linear_other = rf_tsne_clin_low, rf_clin = rf_tsne_coords_all, rf_other = rf_tsne_coords_low, description_other = "LOW group", predicting = paste0("EGFR Rate with RF Models (LOW TSNE group, n = ", length(rf_tsne_clin_low), ")"))

plot_compare_models(linear_clin = rf_tsne_clin_all, linear_other = rf_tsne_clin_med, rf_clin = rf_tsne_coords_all, rf_other = rf_tsne_coords_med, description_other = "MED group", predicting = paste0("EGFR Rate with RF Models (MED TSNE group, n = ", length(rf_tsne_clin_med), ")"))

plot_compare_models(linear_clin = rf_tsne_clin_all, linear_other = rf_tsne_clin_high, rf_clin = rf_tsne_coords_all, rf_other = rf_tsne_coords_high, description_other = "HIGH group", predicting = paste0("EGFR Rate with RF Models (HIGH TSNE group, n = ", length(rf_tsne_clin_high), ")"))

SOM

som_df <- data.frame(cluster_train_sum_filtered[grep("SOM", names(cluster_train_sum_filtered))], 
                      "egfr_rate" = cluster_train_sum_filtered$rawdata.mu_egfr, 
                      cluster_train_sum_filtered[grep("v3y0", names(cluster_train_sum_filtered))])

low_som <- som_df[som_df$SOM_cluster %in% c(5),]
high_som <- som_df[som_df$SOM_cluster %in% c(1),]
med_som <- som_df[!som_df$SOM_cluster %in% c(1,5),]

LINEAR

# ONLY CLIN

# all clin, all pats
model <- lm(egfr_rate ~ . - SOM_cluster - SOM_X - SOM_Y, data = som_df)
lm_som_clin_all <- (model$fitted.values - som_df$egfr_rate)^2

# all clin, low pats
model <- lm(egfr_rate ~ . - SOM_cluster - SOM_X - SOM_Y, data = low_som)
lm_som_clin_low <- (model$fitted.values - low_som$egfr_rate)^2

# all clin, med pats
model <- lm(egfr_rate ~ . - SOM_cluster - SOM_X - SOM_Y, data = med_som)
lm_som_clin_med <- (model$fitted.values - med_som$egfr_rate)^2

# all clin, high pats
model <- lm(egfr_rate ~ . - SOM_cluster - SOM_X - SOM_Y, data = high_som)
lm_som_clin_high <- (model$fitted.values - high_som$egfr_rate)^2

# INCLUDE COORDS

# all clin + coords, all pats
model <- lm(egfr_rate ~ . - SOM_cluster, data = som_df)
lm_som_coords_all <- (model$fitted.values - som_df$egfr_rate)^2

# all clin + coords, low pats
model <- lm(egfr_rate ~ . - SOM_cluster, data = low_som)
lm_som_coords_low <- (model$fitted.values - low_som$egfr_rate)^2

# all clin + coords, med pats
model <- lm(egfr_rate ~ . - SOM_cluster, data = med_som)
lm_som_coords_med <- (model$fitted.values - med_som$egfr_rate)^2

# all clin + coords, high pats
model <- lm(egfr_rate ~ . - SOM_cluster, data = high_som)
lm_som_coords_high <- (model$fitted.values - high_som$egfr_rate)^2

# PLOT

plot_compare_models(linear_clin = lm_som_clin_all, linear_other = lm_som_clin_low, rf_clin = lm_som_coords_all, rf_other = lm_som_coords_low, description_other = "LOW group", predicting = paste0("EGFR Rate with Linear Models (LOW SOM group, n = ", length(lm_som_clin_low), ")"))

plot_compare_models(linear_clin = lm_som_clin_all, linear_other = lm_som_clin_med, rf_clin = lm_som_coords_all, rf_other = lm_som_coords_med, description_other = "MED group", predicting = paste0("EGFR Rate with Linear Models (MED SOM group, n = ", length(lm_som_clin_med), ")"))

plot_compare_models(linear_clin = lm_som_clin_all, linear_other = lm_som_clin_high, rf_clin = lm_som_coords_all, rf_other = lm_som_coords_high, description_other = "HIGH group", predicting = paste0("EGFR Rate with Linear Models (HIGH SOM group, n = ", length(lm_som_clin_high), ")"))

RF

# ONLY CLIN

# all clin, all pats
model <- randomForest(egfr_rate ~ . - SOM_cluster - SOM_X - SOM_Y, data = som_df)
rf_som_clin_all <- (model$predicted - som_df$egfr_rate)^2

# all clin, low pats
model <- randomForest(egfr_rate ~ . - SOM_cluster - SOM_X - SOM_Y, data = low_som)
rf_som_clin_low <- (model$predicted - low_som$egfr_rate)^2

# all clin, med pats
model <- randomForest(egfr_rate ~ . - SOM_cluster - SOM_X - SOM_Y, data = med_som)
rf_som_clin_med <- (model$predicted - med_som$egfr_rate)^2

# all clin, high pats
model <- randomForest(egfr_rate ~ . - SOM_cluster - SOM_X - SOM_Y, data = high_som)
rf_som_clin_high <- (model$predicted - high_som$egfr_rate)^2

# INCLUDE COORDS

# all clin + coords, all pats
model <- randomForest(egfr_rate ~ . - SOM_cluster, data = som_df)
rf_som_coords_all <- (model$predicted - som_df$egfr_rate)^2

# all clin + coords, low pats
model <- randomForest(egfr_rate ~ . - SOM_cluster, data = low_som)
rf_som_coords_low <- (model$predicted - low_som$egfr_rate)^2

# all clin + coords, med pats
model <- randomForest(egfr_rate ~ . - SOM_cluster, data = med_som)
rf_som_coords_med <- (model$predicted - med_som$egfr_rate)^2

# all clin + coords, high pats
model <- randomForest(egfr_rate ~ . - SOM_cluster, data = high_som)
rf_som_coords_high <- (model$predicted - high_som$egfr_rate)^2

# PLOT

plot_compare_models(linear_clin = rf_som_clin_all, linear_other = rf_som_clin_low, rf_clin = rf_som_coords_all, rf_other = rf_som_coords_low, description_other = "LOW group", predicting = paste0("EGFR Rate with RF Models (LOW SOM group, n = ", length(rf_som_clin_low), ")"))

plot_compare_models(linear_clin = rf_som_clin_all, linear_other = rf_som_clin_med, rf_clin = rf_som_coords_all, rf_other = rf_som_coords_med, description_other = "MED group", predicting = paste0("EGFR Rate with RF Models (MED SOM group, n = ", length(rf_som_clin_med), ")"))

plot_compare_models(linear_clin = rf_som_clin_all, linear_other = rf_som_clin_high, rf_clin = rf_som_coords_all, rf_other = rf_som_coords_high, description_other = "HIGH group", predicting = paste0("EGFR Rate with RF Models (HIGH SOM group, n = ", length(rf_som_clin_high), ")"))


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

R Package Documentation

rdrr.io home R language documentation Run R code online

Browse R Packages

CRAN packages Bioconductor packages R-Forge packages GitHub packages

We want your feedback!

Note that we can't provide technical support on individual packages. You should contact the package authors for that.
GitHub issue tracker
ian@mutexlabs.com
Personal blog

 
Embedding an R snippet on your website

Add the following code to your website.

For more information on customizing the embed code, read Embedding Snippets.

Close