cdev-real-data.Rmd

---
title: "cdev range in real data sets"
output: html_notebook
---

```{r setup, include = FALSE}
library(magrittr)
library(ggplot2)
library(parallel)
library(Biobase)
options(stringsAsFactors = FALSE)
devtools::load_all('../')
burd = colorRampPalette(colors = c("blue", "white", "red"))(n = 255)
rdbu = colorRampPalette(colors = c("red", "white", "blue"))(n = 255)
blues = colorRampPalette(colors = c('white', 'blue'))(n = 255)
breakList = seq(-1,1,by = 0.01)

is.good.refs = function(cnt_matrix, min.count = 10) {
    and(grepl('ERCC-', rownames(cnt_matrix)), apply(cnt_matrix, 1, min) >= min.count)
}
DATAPATH = '../data/'
```

### Summary

```{r}
load(file.path(DATAPATH,'data_container.Rdata'))
DATASET_PREFIXES = c('rbm1', 'rbm2', 'rnor1', 'rnor2',
                     'dmelAV', 'dmelAC', 
                     'mpreg1a', 'mpreg1b', 'mpreg2a', 'mpreg2b',
                     'sarcoma', 'lymphoma', 'yfv',
                     'xtro1m', 'xtro2'
                     )
```


```{r}
lapply(DATASET_PREFIXES, function(data_prefix) {
    x.cnt = get(paste0(data_prefix, '.cnt'), envir = data.container)
    x.refs.idx = get(paste0(data_prefix, '.refs.idx'), envir = data.container)
    data.frame('label' = data_prefix,
               'n_genes' = dim(x.cnt)[1],
               'n_samples' = dim(x.cnt)[2],
               'n_references' = length(x.refs.idx))
}) %>%
    do.call(rbind, .)
```


## cdev(random, ground)

```{r}
env.random_vs_ground = readRDS(file.path(DATAPATH,'cdev_random_vs_ground.RDS'))
df.random_vs_ground =
    DATASET_PREFIXES %>%
    # ls(env.random_vs_random) %>%
    lapply(function(data_prefix) {
       `$<-`(env.random_vs_ground[[data_prefix]], 'data', data_prefix)
    }) %>%
    do.call(rbind, .)

ggplot(df.random_vs_ground) +
    geom_boxplot(aes(x=data,y=cdev,group=data)) +
    scale_y_log10() +
    theme(axis.text.x = element_text(angle = 30,hjust = 1)) +
    ggtitle('cdev(X, Y),\nX normalized by true references,\nY by random reference set of the same size')
```

```{r,fig.width=10,fig.height=10}
ggplot(df.random_vs_ground) +
    facet_wrap('data', scales = 'free') +
    geom_point(aes(x=refs_cv_max,y=cdev))
```


## cdev(subset_ground, ground)

Must be close to 1

```{r}
env.ground_vs_ground = readRDS(file.path(DATAPATH,'cdev_ground_vs_ground.RDS'))
df.ground_vs_ground =
    DATASET_PREFIXES %>%
    # ls(env.random_vs_random) %>%
    lapply(function(data_prefix) {
       `$<-`(env.ground_vs_ground[[data_prefix]], 'data', data_prefix)
    }) %>%
    do.call(rbind, .)

ggplot(df.ground_vs_ground) +
    geom_boxplot(aes(x=data,y=cdev,group=data)) +
    theme(axis.text.x = element_text(angle = 30,hjust = 1)) +
    ggtitle('cdev(X_i, X),\nX normalized by true references S,\nX_i by random subsets of S')
```

```{r,fig.width=10,fig.height=10}
ggplot(df.ground_vs_ground) +
    facet_wrap('data', scales = 'free') +
    geom_point(aes(x=refs_cv_median,y=cdev,group=data))
```



## cdev(subset_random, random)

What if the "ground-truth" is wrong/random? Can $cdev(X_i, X)$ help distinguish a wrong reference normalization?

```{r,fig.width=10,fig.height=10}
env.random_vs_random = readRDS(file.path(DATAPATH,'cdev_random_vs_random.RDS'))
df.random_vs_random =
    DATASET_PREFIXES %>%
    # ls(env.random_vs_random) %>%
    lapply(function(data_prefix) {
       `$<-`(env.random_vs_random[[data_prefix]], 'data', data_prefix)
    }) %>%
    do.call(rbind, .)

ggplot(df.random_vs_random) +
    facet_wrap('data', scales = 'free_y') +
    geom_boxplot(aes(x=random_fullset_index,y=cdev,group=random_fullset_index)) +
    scale_y_log10()
    
```
```{r}
ggplot(df.random_vs_random) +
    geom_boxplot(aes(x=data,y=cdev)) +
    scale_y_log10() + 
    theme(axis.text.x = element_text(angle = 30,hjust = 1)) +
    ggtitle('cdev(Xi, X),\nX normalized by random references S,\nXi by random subset of S')
```

```{r}
df.random_vs_random %>%
    ggplot() +
    geom_point(aes(x=subset_avgCV,y=cdev))
```

## cdev behavior by data sets

```{r, fig.width=12,fig.height=10}
# df.ground_vs_ground$cdev <- df.ground_vs_ground$cdev_i_ref
# df.random_vs_ground$cdev <- df.random_vs_ground$cdev_i_ref
# df.random_vs_random$cdev <- df.random_vs_random$cdev_i

df.bydataset = lapply(c('random_vs_random', 'random_vs_ground', 'ground_vs_ground'), function(comparison) {
    the_df = get(paste0('df.', comparison))
    data.frame('cdev' = the_df$cdev,
               'data' = the_df$data, 
               'comparison' = comparison)
}) %>%
    do.call(rbind, .)

ggplot(df.bydataset) +
    facet_wrap('data', scales = 'free_y') +
    geom_boxplot(aes(x=comparison,y=cdev,color=comparison)) +
    theme(axis.text.x = element_text(angle=60,hjust=1))
```
```{r,fig.width=12,fig.height=10}
ggplot(df.bydataset) +
    facet_wrap('data', scales = 'free_y') +
    geom_violin(aes(x=comparison,y=cdev,color=comparison)) +
    theme(axis.text.x = element_text(angle=60,hjust=1))
```

```{r,fig.width=10,fig.height=9}
ggplot(df.bydataset) +
    facet_wrap('data') +
    geom_boxplot(aes(x=comparison,y=cdev,color=comparison)) +
    theme_bw(base_size = 14) +
    # theme_replace(axis.text.x = element_text(angle=60,hjust=1)) +
    theme(axis.text.x = element_text(angle=90,hjust=1,vjust=0)) +
    scale_y_log10(limits=c(1,100))
ggsave('cdev-by-dataset.png',width=10,height = 9)
```