distance_from_origin.Rmd

---
title: "test"
output: html_document
---

```{r setup, include=FALSE}
# all usefull libraries
library(magrittr)
library(dplyr)
library(ggplot2)
library(foreach)
library(doMC)
library(stringr)
library(tidyverse)
```

```{r import data with feature selection, message=FALSE, eval=FALSE}
filename <- "Pf_Gustafsdottir.rds"
profiles <- readRDS(file.path("..", "..", "input", "BBBC022_2013", "old", filename))
profiles <- profiles$data

profiles %<>% rename(Metadata_plate = Plate, Metadata_well = Well, Metadata_BROAD_sample = Image_Metadata_BROAD_ID,  Metadata_SOURCE_COMPOUND_NAME = Image_Metadata_SOURCE_COMPOUND_NAME)
variables <-
  names(profiles) %>% str_subset("^Cells_|^Cytoplasm_|^Nuclei_") # nfeat

metadata <-
  names(profiles) %>% str_subset("^Metadata_") # metadata

# add column wheather DMSO or treatment
profiles %<>% mutate(Metadata_broad_sample_type = ifelse(Metadata_BROAD_sample == "", "control", "trt"))
metadata <-
  names(profiles) %>% str_subset("^Metadata_") # metadata
```


```{r compound check}
zero_vector <- rep(0, length(variables))
zero_vector.2 <- profiles %>% select(one_of(variables)) %>% summarise_each(funs(mean)) %>% unlist
zero_vector.3 <- profiles %>% select(one_of(variables)) %>% summarise_each(funs(median)) %>% unlist

dist <- apply(profiles %>% select(one_of(variables)), 1, function(x) sqrt(sum((x-zero_vector)^2)))
dist.2 <- apply(profiles %>% select(one_of(variables)), 1, function(x) sqrt(sum((x-zero_vector.2)^2)))
dist.3 <- apply(profiles %>% select(one_of(variables)), 1, function(x) sqrt(sum((x-zero_vector.3)^2)))

dat <- data.frame(distance = dist, type = profiles$Metadata_broad_sample_type)
dat2 <- data.frame(distance = dist.2, type = profiles$Metadata_broad_sample_type)
dat3 <- data.frame(distance = dist.3, type = profiles$Metadata_broad_sample_type)

library(plyr)
cdat <- ddply(dat, "type", summarise, rating.mean=mean(distance))
cdat

# Overlaid histograms with means
ggplot(dat, aes(x=distance, fill=type, y=..density../sum(..density..))) +
  geom_histogram(binwidth=5, alpha=.4, position="identity")	+ 
  xlim(0, 500) +
  geom_vline(data=cdat, aes(xintercept=rating.mean,  colour=type),
               linetype="dashed", size=0.5) +
  ggtitle("Euclidean distance to center") +
  ylab("density")

library(plyr)
cdat2 <- ddply(dat2, "type", summarise, rating.mean=mean(distance))
cdat2

# Overlaid histograms with means
ggplot(dat2, aes(x=distance, fill=type, y=..density../sum(..density..))) +
  geom_histogram(binwidth=5, alpha=.4, position="identity")	+ 
  xlim(0, 500) +
  geom_vline(data=cdat2, aes(xintercept=rating.mean,  colour=type),
               linetype="dashed", size=0.5) +
  ggtitle("Euclidean distance to center") +
  ylab("density")

library(plyr)
cdat3 <- ddply(dat3, "type", summarise, rating.mean=mean(distance))
cdat3

# Overlaid histograms with means
ggplot(dat3, aes(x=distance, fill=type, y=..density../sum(..density..))) +
  geom_histogram(binwidth=5, alpha=.4, position="identity")	+ 
  xlim(0, 500) +
  geom_vline(data=cdat, aes(xintercept=rating.mean,  colour=type),
               linetype="dashed", size=0.5) +
  ggtitle("Euclidean distance to center") +
  ylab("density")

thres <- quantile(dat %>% filter(type == "control") %>% select(one_of(c("distance"))) %>% unlist, .95)
print(thres)

tmp <- dat %>% filter(type == "trt") %>% select(one_of(c("distance"))) %>% unlist
inds <- which(tmp > thres)

percentage_selected <- length(inds)/length(tmp) # arount 22%

```