rmarkdown.rmd

---
title: "Simple demo of rmarkdown"
author: "Soumya Banerjee"
date: "30 August 2022"
output:
  pdf_document:
    toc: yes
    number_sections: yes
    fig_caption: yes
  html_document:
    toc: yes
    df_print: paged
urlcolor: blue
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo=TRUE)
```

# Summary

This is a document that outlines a demo of an R markdown.

```{r, include = FALSE}

####################
# Load library
####################
library(knitr)
library(rmarkdown)
library(tinytex)
library(pROC)
library(precrec)
library(PRROC)
library(boot)
library(rlib)

# LIBRARY_PREFIX <- "https://egret.psychol.cam.ac.uk/rlib/"
# source(paste0(LIBRARY_PREFIX, "cris_common.R"))

# source("https://raw.githubusercontent.com/neelsoumya/rlib/master/cris_common.R")
source("cris_common.R")
source("miscstat.R")
# source("https://raw.githubusercontent.com/neelsoumya/rlib/master/miscstat.R")


#######################
# Get data
#######################
df_metagene_score_final = read.csv('metagene_score.csv', 
                                  sep = ',', header = TRUE, 
                                  stringsAsFactors=FALSE, na.strings="..")


```

# Visualize data

todo histogram





```{r, include=FALSE}

TRAIN = sample(c(TRUE,FALSE),
                  nrow(df_metagene_score_final),
                  replace = TRUE)

TEST = (!TRAIN)

df_metagene_score_final_TRAIN = df_metagene_score_final[TRAIN,]
df_metagene_score_final_TEST  = df_metagene_score_final[TEST,]

```


# Perform logistic regression





```{r, echo=FALSE}

mylogit <- glm(flag_yes_no ~ metagene_score, 
               data = df_metagene_score_final_TRAIN, 
               family = "binomial")
               
               ######################################
# Check linear model distributions
#######################################
# miscstat$check_distribution(model = mylogit)


#######################################
# Visualize parameter distributions
# holds for linear mixed effects models lmer()
#######################################
# cris$visualize_fixed_effects_from_lmer(lmer_result = glm_object_best)
# cris$fixed_effects_from_lmer(lmer_result = glm_object_best)


###############################
# predict on test set
###############################
prob = predict(mylogit, 
               type=c("response"),
               newdata = df_metagene_score_final_TEST)
df_metagene_score_final_TEST$prob = prob


###############################
# Additional code to do 
#   cross-validation NOT USED
# on TRAINING SET
###############################
cost <- function(r, pi=0) mean(abs(r-pi)>0.5)
cv_err <- cv.glm(data = df_metagene_score_final_TRAIN,
                 K = 4,
                 cost = cost,
                 glmfit = mylogit
)

# can use the following to perform model selection if necessary
cv_err$delta[1]


###############################
# Generate ROC curves
#   on TEST SET
###############################
g <- pROC::roc(flag_yes_no ~ prob, data=df_metagene_score_final_TEST)

plot(g)

cat("AUC is:", g$auc)


############################
# Precision recall curve
#     on test set
############################
mmdata_flag = mmdata(df_metagene_score_final_TEST$metagene_score,
                     df_metagene_score_final_TEST$flag_yes_no)
smcurves <- evalmod(mmdata_flag, raw_curves = TRUE)

# We also show the precision recall curve below 
# which is better suited for cases in which there are class imbalances.

plot(smcurves, raw_curves = FALSE)

############################
# even better AUC and AUPR 
# curves with areas reportet
#   on TEST SET
############################

fg <- prob[df_metagene_score_final_TEST$flag_yes_no == 1]
bg <- prob[df_metagene_score_final_TEST$flag_yes_no == 0]

# ROC Curve    
roc <- roc.curve(scores.class0 = fg, scores.class1 = bg, curve = TRUE)
plot(roc)

# PR Curve
pr <- pr.curve(scores.class0 = fg, scores.class1 = bg, curve = TRUE)
plot(pr)

cat("AUPR is:", pr$auc.integral)


##################################
# Better AUPR plot using ggplot
##################################
source("convert_aupr_to_ggplot.R")

i_y_line_threshold_signif_aupr = length(which(df_metagene_score_final$flag_yes_no == 1))/(length(which(df_metagene_score_final$flag_yes_no == 1)) + length(which(df_metagene_score_final$flag_yes_no == 0)))

convert_aupr_to_ggplot(i_y_line_threshold_signif=i_y_line_threshold_signif_aupr, 
                                   prroc_object=pr,
                                   str_filename_save="aupr_ggplot.pdf")

```



\newpage