Introduction to R

.title[
# Introduction to R
]
.subtitle[
## R Foundations for Data Analysis
]
.author[
### Guilherme Dias
]

---

---
name: content
class: spaced

# Contents

1. [The problem: life in spreadsheets](#whyR) 🧨
2. [What R is (and isn't)](#about) 🧭
3. [How R compares](#compare) ⚖️
4. [Thinking like a programmer](#programming_language) 🧠
5. [Packages: standing on giants' shoulders](#packages) 📦
6. [Coding in the age of AI](#ai) 🤖

---
.pull-left-50[
Have you ever...

* lost track of which Excel file made a figure?
* redone an analysis by hand because data changed?
* had a gene name silently turned into a date?
* asked ChatGPT/Claude/Copilot to write code for you?
]

#### ...this course is for you.
]

---
name: whyR

# The GUI trap 🧨

* Manual copy-pasting across Excel files
* "Which version of `data_final_v3_REAL.xlsx` was this graph made from?"
* No record of *what you clicked*
* Silent gene name conversions (`SEPT9` `$\rightarrow$` `Sept-09`, `MARCH1` `$\rightarrow$` `1-Mar`)

--
 
Gene-name autocorrection errors were found in roughly **1 in 5** genomics papers with Excel supplements (Ziemann *et al.*, *Genome Biology*, 2016). The HGNC eventually **renamed the genes** to stop it.

---
name: about

# Briefly about R 🧭

* a programming language **built for data & statistics**
* free, open-source, and runs everywhere
* driven by a core team **and** a huge community
* the home of cutting-edge statistics and bioinformatics
* a powerful computational tool in general
]

* the fastest language
* the most elegant language
* the best tool for *every* job (e.g. building games or phone apps)

]

---
name: compare

# How does R compare? ⚖️

---
# Cost & licensing 💸

| Tool | Academic / Student | Commercial | Licensing Model |
| :--- | :---: | :---: | :--- |
| **R / Python** | **$0** | **$0** | **Open-source** |
| **MS Excel** | ~$100 / yr (M365) | ~$150 / yr | Subscription |
| **GraphPad Prism** | ~$140 – $260 / yr | ~$520 / yr | Per seat |
| **Stata** | ~$125 – $250 / yr | ~$1,000 – $3,700 / yr | Per user / core |
| **SPSS** | ~$100 – $300 / yr | ~$1,200 – $3,800 / yr | Per user / module |
| **SAS** | Free (OnDemand, limited) | $10,000s+ / yr | Enterprise |

* Your code keeps working when you **change jobs, institutions or countries**. 🔓
* Reviewers and collaborators can **re-run your analysis** without buying anything.

---
# R vs. Python 🐍

> **Python** is a general-purpose language that *can* do data analysis.  
> **R** was built *specifically* for data and statistics.

--
### Where R shines

* **Data wrangling:** the `tidyverse` (`dplyr`, `tidyr`) reads almost like English.
* **Plots:** `ggplot2` publication-quality figures with sensible defaults.
* **Statistics & biology:** new methods often appear in R/Bioconductor first.
* **Reporting:** **R Markdown**, **Quarto** and **Shiny** turn your code into reports, slides (like these!) and web apps.

.content-box-green[
**It's not a war.** The concepts you learn here (variables, functions, data frames, loops) transfer directly to Python.
]

---
name: demo
class: middle

# Three lines of R 🪄

``` r
ggplot(iris, aes(x = Petal.Length, y = Petal.Width, colour = Species)) +
  geom_point(size = 2, alpha = 0.8) +
  geom_smooth(method = "lm", se = FALSE)
```

New data? Change **one word** (`iris`) and press **Run**. Try that in Excel.

---

# Why programming matters

Point-and-click is easy for **today**.  
`coding` is a superpower for **tomorrow**.

---
# Why programming matters

- **Self-documenting:** code records every step — no guessing how you got your results.
- **Instant re-runs:** new data? Press **Run** instead of remaking charts by hand.
- **Fewer silent errors:** no hidden formulas, shifted cells or forgotten menu clicks.
- **Scales effortlessly:** the same script handles 10 rows or 10 million.
- **Shareable:** a script is a recipe anyone can follow.

---
name: programming_language

# Thinking like a programmer 🧠

Every programming language has three ingredients:

.pull-left-50[
1. **Data types** — *what* are we working with?
2. **Syntax** — *how* do we write it?
3. **Semantics** — what does it *mean*?
]

* words (nouns, numbers...)
* grammar
* meaning
]

---
# 1- Data types in R 🧱

Missing values: `NA` (fits any type)
]

---
# 2 — Syntax: how do we write it?

The rules of the language's *grammar*.

The same calculation in two languages:

```
2 * 1 + 1        # R, Python, most languages
(+ (* 2 1) 1)    # Lisp
```

Break the grammar and the computer **refuses** to run:

``` r
2 * (1 + 1
```

```
## Error in parse(text = input): <text>:2:0: unexpected end of input
## 1: 2 * (1 + 1
## ^
```

**Good news:** syntax errors are loud and easy to fix.

---
# 3 — Semantics: what does it *mean*?

Grammatically correct ≠ meaningful:

* *Colorful yellow sleeps on a crazy wave.* — no meaning
* *There is $500 on his empty bank account.* — internal contradiction

The same is true for code:

``` r
height <- 180 # cm
weight <- 70 # kg
bmi <- weight / height # runs fine... but is it right?
bmi
```

```
## [1] 0.3888889
```

.content-box-yellow[
BMI = weight (kg) / height (m)². R happily computed **nonsense**.  
**Semantic errors are silent.** Only *you* can catch them. Remember this when using AI!
]

---
name: topic2

# Where to start?
.pull-left-50[

``` r
knitr::include_graphics("images/breakit.jpeg")
```

<img src="images/breakit.jpeg" width="395" style="display: block; margin: auto auto auto 0;" />
]
--

*Decomposition* — divide and rule.

**Top-down approach:** define the big problem and split it into smaller ones.

This is exactly how to work well with an AI assistant: break the problem into small, verifiable pieces.

---
name: packages

# Packages 📦

.pull-left-50[
* ready-made functions, data & documentation
* developed by the community
* cover very diverse areas of science
* uniformly structured and documented
* organised in repositories:
    + [CRAN](https://cran.r-project.org) — general purpose, 20,000+ packages
    + [Bioconductor](https://www.bioconductor.org) — bioinformatics, 2,000+ packages
    + [GitHub](https://github.com) — development versions
    + [R-universe](https://r-universe.dev) — browse it all
]

.pull-right-50[
<img src="data/slide_intro/packages.jpg" width="220pt" style="display: block; margin: auto;" />

<img src="images/tidyverse-icons.png" width="220pt" style="display: block; margin: auto;" />
]

---
name: CRAN

# Working with packages — a CRAN page

**Tip:** the *Reference manual* and *Vignettes* are your best friends (and great context to give an AI assistant!).

---
name: pkg_cran_inst

# Installing packages

Only a few packages come pre-installed:

``` r
library(modelr)
```

Install **once** per computer:

``` r
install.packages("ggplot2")
```

Load **every** session:

``` r
library(ggplot2)   # stops with an error if the package is missing
require(ggplot2)   # only warns, returns FALSE — handy inside functions
```

---
name: work_pkg_details
exclude: true

# Installing packages — details

Choose a specific mirror (e.g. if your default one is down):

``` r
install.packages("ggplot2", repos = "https://cloud.r-project.org")
```

Sometimes no pre-built version exists for your platform, so you install from **source code**:

``` r
install.packages("path/to/ggplot2_3.5.1.tar.gz",
                 repos = NULL,
                 type  = "source")
```

---
name: pkg_github

# Packages — GitHub

More and more developers share packages via GitHub. Use `remotes` (or `devtools`, which includes it):

``` r
install.packages("remotes")
remotes::install_github("talgalili/installr")  # "username/repo"
```

.pull-center-50[
<img src="images/installr.png" width="650pt" style="display: block; margin: auto;" />
]

.footnote[⚠️ GitHub packages are not reviewed like CRAN/Bioconductor ones. Check who wrote it and whether it's maintained.]

---
name: pkg_bioconductor

# Packages — Bioconductor

First install the Bioconductor manager (once):

``` r
if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
```

Then install Bioconductor packages:

``` r
BiocManager::install("GenomicRanges")
```

For more info, visit the [Bioconductor website](https://www.bioconductor.org/install/).

---

# One tool to install them all: `pak` 🚀

`pak` installs from **CRAN, Bioconductor and GitHub** with one function — and it's fast.

``` r
install.packages("pak")

pak::pak("ggplot2")              # CRAN
pak::pak("GenomicRanges")        # Bioconductor
pak::pak("talgalili/installr")   # GitHub
```

---

# Coding in the age of AI 🤖

### "If AI can write code, why should I learn R?"

---

# AI is a great assistant...

AI tools (ChatGPT, Claude, GitHub Copilot, Positron Assistant...) can:

* write a first draft of a script in seconds
* explain cryptic error messages
* translate code (Excel formula → R, Python → R)
* suggest packages and functions you didn't know about
* comment and document your code

There are even R packages to call language models **from R**, e.g. `ellmer` (from the Posit / tidyverse team).

---

# ...but you are still the scientist 🔬

* produce code that **runs**
* sound **very** confident
* write a lot, fast
]

* invent functions or packages that **don't exist**
* use the **wrong statistical test**
* silently drop `NA`s, rows or samples
* be wrong in ways that *look* right
]

---
exclude: true
# Spot the bug 🕵️ (an "AI-generated" snippet)

> *Prompt: "Compare mean expression between treated and control samples."*

``` r
expr <- data.frame(
 sample = paste0("S", 1:6),
 group = c("control", "control", "control", "treated", "treated", "treated"),
 value = c(5.1, 4.8, NA, 7.9, 8.4, 8.1)
)

tapply(expr$value, expr$group, mean)
```

```
##  control  treated 
##       NA 8.133333
```

* Why is `control` `NA`? What should we do about the missing value?
* Is dropping it OK? Should we *report* it?
* With n = 3 per group, which test is appropriate — and why?

These are **scientific** decisions. AI can't make them for you.

---

# Working well with AI: a checklist ✅

1. **Understand first, then prompt.** Know roughly what the answer should look like.
2. **Ask for small pieces**, not a whole analysis (*divide et impera* again!).
3. **Read every line.** If you can't explain it, don't run it on real data.
4. **Test on tiny data** where you already know the answer.
5. **Check that functions & packages actually exist** (`?function_name`).
6. **Save the code, not the chat.** A script is reproducible; a conversation is not.
7. **Never paste sensitive data** (e.g. patient data) into public AI tools.

.content-box-green[
**AI writes the draft. You own the analysis.** That's why this course matters more now, not less.
]

---
# A suggestion:

- Try **not** to use AI agents
- Instead, use AI chatbots to:
  - fix syntax errors
  - learn more about concepts

---
name: recap
# Recap
* Why R?  
* About R.   
* R is a programming language for statistical computing and data analysis.  
* Packages and repositories.   
* Installing packages from CRAN, Bioconductor, GitHub. 
* AI and coding

---
name: end_slide
class: end-slide, middle
count: false

# Thank you! Questions?

.end-text[

Graphics from <img src="./assets/freepik.jpg" style="max-height:20px; vertical-align:middle;"> 
Created: 21-Oct-2026 • <a href="https://www.scilifelab.se/">SciLifeLab</a> • <a href="https://nbis.se/">NBIS</a> 

]