OMICS workshop I Exercise

Advanced Exercise: Integrative Tidy Omics data Wrangling

Goal

In this exercise you will combine core tidy tools from the tutorial in one end-to-end analysis:

• left_join() and anti_join()
• pivot_longer() and pivot_wider()
• mutate() + case_when()
• group_by() + summarise()
• across()

You will work on a synthetic RNA-like count dataset and produce a compact analysis report.

Setup

library(tidyverse)
set.seed(12345) # for reproducibility

Data generation

# 1) Sample metadata
sample_meta <- tibble(
  sample_id = str_c("S", str_pad(1:30, width = 3, pad = "0")),
  tissue = sample(c("tumor", "normal"), 30, replace = TRUE, prob = c(0.55, 0.45)),
  cohort = sample(c("A", "B", "C"), 30, replace = TRUE),
  batch = sample(c("B1", "B2", "B3"), 30, replace = TRUE)
)

# 2) Clinical table (with intentional unmatched rows for join debugging)
clinical <- tibble(
  sample_id = c(sample_meta$sample_id[1:26], "S900", "S901"),
  patient_age = sample(35:85, 28, replace = TRUE),
  response = sample(c("CR", "PR", "SD", "PD"), 28, replace = TRUE)
)

# 3) Wide expression/count matrix (genes x samples)
genes <- str_c("GENE_", str_pad(1:120, 3, pad = "0"))
count_mat <- matrix(
  rnbinom(length(genes) * nrow(sample_meta), mu = 120, size = 1.3),
  nrow = length(genes),
  ncol = nrow(sample_meta),
  dimnames = list(genes, sample_meta$sample_id)
)

counts_wide <- as_tibble(count_mat, rownames = "gene_id")

sample_meta
clinical
counts_wide

Exercise tasks

Part 1 — Join diagnostics and clean analysis table

1. Use anti_join() to identify:
   • sample IDs in sample_meta that do not exist in clinical
   • sample IDs in clinical that do not exist in sample_meta
2. Create clinical_clean by keeping only rows that can match sample_meta.
3. Build sample_annot using left_join(sample_meta, clinical_clean, by = "sample_id").
4. Add a new age group variable with case_when():
   • < 50 = "young"
   • 50-69 = "middle"
   • >= 70 = "older"

# TODO: write your Part 1 solution here

Part 2 — Reshape wide counts and annotate samples

1. Convert counts_wide to long format as counts_long with columns: gene_id, sample_id, count.
2. Join counts_long with sample_annot to create counts_annot.
3. Create a log-transformed value: log_count = log2(count + 1).

# TODO: write your Part 2 solution here

Part 3 — Multi-level summaries using group_by(), summarise(), and across()

1. Compute per-gene/per-tissue summary as gene_tissue_summary:
   • mean_count
   • sd_count
   • mean_log_count
   • n_samples
2. Compute a second table cohort_summary with one row per cohort and tissue:
   • mean and median for all numeric columns in counts_annot using across(where(is.numeric), ...)
   • include n() as n_rows

# TODO: write your Part 3 solution here

Part 4 — Wide result table and differential signal

1. Pivot gene_tissue_summary wider to get one row per gene with separate tumor/normal means:
   • columns like mean_count_tumor, mean_count_normal
2. Create:
   • delta = mean_count_tumor - mean_count_normal
   • abs_delta = abs(delta)
3. Return top 15 genes by abs_delta.

# TODO: write your Part 4 solution here

Deliverables

Create and print these objects:

• sample_annot
• counts_annot
• gene_tissue_summary
• cohort_summary
• top_genes_delta

# TODO: print your final deliverables

Self-check questions

• Do your unmatched IDs from anti_join() make sense given the generated data?
• Does counts_annot have exactly n_genes * n_samples rows for matched samples?
• Are delta values positive for genes higher in tumor and negative for genes higher in normal?
• Can you explain why log2(count + 1) is often used before summarizing count-like data?

Stretch challenge (optional)

Create a compact report table with one row per cohort containing:

• top 3 genes by absolute tumor-normal delta within that cohort
• mean age of included samples
• fraction of response == "CR"

Tips: use grouped summaries, ranking, and string aggregation (str_c() with collapse = ", ").

# TODO: stretch solution