Feature_Importance.Rmd

---
title: "Mechanistic interpretation of feature importance"
author: Saw Simeon, Nuttapat Anuwongcharoen, Watshara Shoombuatong, Aijaz Ahmad Malik,
  Virapong Prachayasittikul, Jarl E. S. Wikberg and Chanin Nantasenamat
date: "June 9, 2016"
output: pdf_document
---

### Feature importance analysis from Random forest using Gini index (i.e., variance of the responses).

Creating a R function to perform feature importance
```{r}
randomForest_feature_importance <- function(x) {
  library(doSNOW)
  library(foreach)
  library(parallel)
  cl <- makeCluster(8)
  registerDoSNOW(cl)
  
  results <- list(100)
  results <- foreach (i = 1:100) %dopar% {
    x <- na.omit(x)
    set.seed(i)
    para <- dplyr::sample_n(x, size = 2570, replace = TRUE)
    set.seed(i)
    in_train_para <- sample(nrow(para),
                            size = as.integer(nrow(para) * 0.8),
                            replace = FALSE)
    Train <- para[in_train_para, ]
    Test <- para[-in_train_para, ]
    rm(in_train_para)
    rm(Test)
    set.seed(i)
    model <- ranger::ranger(pIC50~., data = Train, importance = 'impurity',
                                  write.forest = TRUE, save.memory = TRUE)
    rm(Train)
    importance <- model$variable.importance
    results[[i]] <- importance
  }
  return(results)
  stopCluster(cl)
}

```

Generating the plot of feature importance

```{r, fig.width=10, fig.height=20, warning=FALSE, message=FALSE, fig.align='center'}
input <- readRDS(file = "data.Rds")
SubStructure_fingerPrintCount <- input$Substructure_fingerPrintCount
set.seed(10)
results_feature_importance_RF <- randomForest_feature_importance(SubStructure_fingerPrintCount)
data1 <- data.frame(results_feature_importance_RF)
data1 <- cbind(features = rownames(data1), data1)
library(reshape2)
data_melt <- suppressWarnings(melt(data1, id.vars = "features"))
#data_melt <- melt(data1, id.vars = "features")
data_melt$features <- factor(data_melt$features)
library(ggplot2)
set.seed(1)
plot_feature <- ggplot(data_melt, aes(x = reorder(features, value, FUN = median), y = value)) +
  geom_boxplot(fill = "#F8766D", colour = "black", alpha = 0.5) +
  theme_bw() + xlab("") + ylab("Gini Index") + coord_flip() + theme(
    axis.text.y = element_text(size = 20, colour = "black"),
    axis.text.x = element_text(size = 20, colour = "black"),
    #axis.title.x = element_blank(),
    plot.margin = unit(c(1, 1, 1, 1), "cm"),
    panel.border = element_rect(linetype = "solid", colour = "black", fill = NA, size = 1),
    axis.title = element_text(size = 25, face = "bold", colour = "black")
  )

plot_feature
```