---
title: "Statistical tests"
author: Abhijit Dasgupta
date: BIOF 339

---

```{r setup, include=FALSE, child = here::here('slides/templates/setup.Rmd')}
```


---
class: middle, center

# Comparing two groups

```{r, echo=F, results='hide'}
library(rio)
brca1 <- import('../data/clinical_data_breast_cancer_hw.csv')
brca2 <- import('../data/BreastCancer_Expression.csv')
brca <- left_join(brca1, brca2, by=c('Complete.TCGA.ID' = 'TCGA_ID')) %>% 
  mutate(Age.at.Initial.Pathologic.Diagnosis =
           as.numeric(Age.at.Initial.Pathologic.Diagnosis)) %>% 
  mutate(ER.Status = ifelse(ER.Status %in% c('Positive','Negative'),
                            ER.Status, NA))

```

---

## The t-test

The t-test compares whether the mean of a variable differs between two groups. 

It does assume the normal distribution for the data, but is robust to deviations from normality

Do **not** test for normality before doing the t-test. It isn't necessary and screws up your error rates

---

## The t-test

In R, there is a convenient function `t.test`

```{r 08-Summaries-24}
t.test(NP_958782 ~ ER.Status, data = brca)
```

Read the code as 

"Do a t-test to see if (the mean of) `NP_958782` differs by `ER.Status`, where
both are taken from the data set `brca`"

You can read the `~` as "by", as in "t-test of NP_958782 by ER.Status"

---

## The t-test

The packge `broom` provides a function `tidy` that makes the results of these
statistical tests tidy.

```{r 08-Summaries-25}
t.test(NP_958782 ~ ER.Status, data=brca) %>% 
  broom::tidy()
```

--

```{r 08-Summaries-26, echo=F}
t.test(NP_958782 ~ ER.Status, data=brca)
```

---

```{r through, include=FALSE, message=F, warning=F}
brca %>% 
  select(ER.Status, starts_with('NP')) %>% 
  pivot_longer(names_to = 'protein',
               values_to = 'expression', 
               cols = c(-ER.Status)) %>% 
  split(.$protein) %>% 
  map(~broom::tidy(t.test(expression ~ ER.Status, 
                          data=.))) %>% 
  bind_rows(.id = 'Protein') %>% 
  select(Protein, estimate, p.value, conf.low, conf.high)
```

`r chunk_reveal('through', title = '## Using <code>broom</code>
The fact that <code>broom::tidy</code> makes the results of tests into tibbles is in fact extremely useful in high-throughput work', widths = c(60,40))`


---
class: center, middle

# Back to testing

---

## Wilcoxon test, nonparametric t-test

```{r 08-Summaries-32}
wilcox.test(NP_958782 ~ ER.Status, data=brca) %>% 
  broom::tidy()
```

--

```{r 08-Summaries-33, echo=F}
wilcox.test(NP_958782 ~ ER.Status, data=brca)
```

---

## Wilcoxon test

.pull-left[
```{r test3, eval = F, echo = T}
brca %>% 
  select(ER.Status, starts_with('NP')) %>% 
  tidyr::gather(protein,expression, -ER.Status) %>% 
  split(.$protein) %>% 
  map(~broom::tidy(wilcox.test(expression ~ ER.Status, 
                          data=.))) %>% 
  bind_rows(.id='Protein') %>% 
  select(Protein, p.value)
```
]
.pull-right[
```{r 08-Summaries-34, eval=T, echo = F, ref.label="test3"}
```
]

---

## Using `tableone`


```{r table4, eval = F, echo = T}
CreateTableOne(
  data = brca %>% filter(!is.na(ER.Status)),
  vars = brca %>% 
    select(starts_with('NP')) %>% 
    names(),
  strata = 'ER.Status',
  test = T,
  testNormal = t.test
)
```
```{r 08-Summaries-35, eval=T, echo = F, ref.label="table4"}
```

--

This is not quite the same results as before

---

## Using `tableone`


```{r table4a, eval = F, echo = T}
CreateTableOne(
  data = brca %>% filter(!is.na(ER.Status)),
  vars = brca %>% 
    select(starts_with('NP')) %>% 
    names(),
  strata = 'ER.Status',
  test = T,
  testNormal = t.test,
  argsNormal = list(var.equal=F) #<<
)
```
```{r 08-Summaries-36, eval=T, echo = F, ref.label="table4a"}
```

---

## Tests for discrete data

Testing whether the distribution of a categorical variable differs by levels of 
another categorical variable can be done using either the Chi-square test (`chisq.test`) or the Fisher's test (`fisher.test`). Both require you to create a 2x2 table first.

```{r 08-Summaries-37}
fisher.test(table(brca$Tumor, brca$ER.Status))
```

---

## Tests for discrete data

Testing whether the distribution of a categorical variable differs by levels of 
another categorical variable can be done using either the Chi-square test (`chisq.test`) or the Fisher's test (`fisher.test`). Both require you to create a 2x2 table first.

```{r 08-Summaries-38}
chisq.test(table(brca$Tumor, brca$ER.Status))
```

---

## Tests for discrete data

We can use `broom::tidy` for either of these

```{r 08-Summaries-39}
chisq.test(table(brca$Tumor, brca$ER.Status)) %>% 
  broom::tidy()
```

---

## Using `tableone`

```{r 08-Summaries-40}
CreateCatTable(vars = c('Tumor','Node','Metastasis'),
               data = filter(brca, !is.na(ER.Status)),
               strata = 'ER.Status',
               test = T) # chisq.test
```

---

## Using `tableone`

```{r 08-Summaries-41}
c1 <- CreateCatTable(vars = c('Tumor','Node','Metastasis'),
               data = filter(brca, !is.na(ER.Status)),
               strata = 'ER.Status',
               test = T)
print(c1, exact = c('Tumor','Node','Metastasis')) # fisher.test
```