Conditional Logic

Turning numbers into labels is a constant task: bin an expression value into low, medium, or high; flag a gene as housekeeping or a target; classify a result as up, down, or unchanged. This is multi-branch conditional logic. dplyr writes it as case_when, Polars as chained when().then() clauses ending in otherwise. Both test conditions top to bottom and take the first that matches.

Every code block runs in a container in the companion repository, and the two languages produce the same labels and the same group sizes.

Bin a value into categories

Classify each measurement by its count: below 300 is low, below 1000 is medium, the rest high. The order matters, since the first matching branch wins.

R

expr <- expr |>
  mutate(level = case_when(
    counts < 300  ~ "low",
    counts < 1000 ~ "medium",
    TRUE          ~ "high"
  ))
head(expr)
count(expr, level)

  sample_id condition  gene counts  level
1       S01   control  TP53    393 medium
2       S01   control  EGFR    294    low
3       S01   control   MYC    247    low
4       S01   control BRCA1    195    low
5       S01   control GAPDH   3939   high
6       S01   control  ACTB   3117   high

   level  n
1   high 16
2    low 15
3 medium 17

Python

expr = expr.with_columns(
    pl.when(pl.col("counts") < 300).then(
        pl.lit("low")).when(pl.col("counts") < 1000).then(
            pl.lit("medium")).otherwise(pl.lit("high")).alias("level"))
expr.head()
expr.group_by("level").agg(pl.len().alias("n")).sort("level")

shape: (5, 5)
┌───────────┬───────────┬───────┬────────┬────────┐
│ sample_id ┆ condition ┆ gene  ┆ counts ┆ level  │
│ ---       ┆ ---       ┆ ---   ┆ ---    ┆ ---    │
│ str       ┆ str       ┆ str   ┆ i64    ┆ str    │
╞═══════════╪═══════════╪═══════╪════════╪════════╡
│ S01       ┆ control   ┆ TP53  ┆ 393    ┆ medium │
│ S01       ┆ control   ┆ EGFR  ┆ 294    ┆ low    │
│ S01       ┆ control   ┆ MYC   ┆ 247    ┆ low    │
│ S01       ┆ control   ┆ BRCA1 ┆ 195    ┆ low    │
│ S01       ┆ control   ┆ GAPDH ┆ 3939   ┆ high   │
└───────────┴───────────┴───────┴────────┴────────┘

shape: (3, 2)
┌────────┬─────┐
│ level  ┆ n   │
│ ---    ┆ --- │
│ str    ┆ u32 │
╞════════╪═════╡
│ high   ┆ 16  │
│ low    ┆ 15  │
│ medium ┆ 17  │
└────────┴─────┘

The TRUE ~ branch in dplyr and .otherwise(...) in Polars are the catch-all: anything not matched above falls through to it. Forgetting the catch-all leaves unmatched rows as missing, a common source of stray NAs.

Flag by set membership

The condition can be any test, including “is this value in a set”. Here, mark the two housekeeping genes and call everything else a target.

R

expr <- expr |>
  mutate(role = case_when(
    gene %in% c("GAPDH", "ACTB") ~ "housekeeping",
    TRUE                         ~ "target"
  ))
count(expr, role)

          role  n
1 housekeeping 16
2       target 32

Python

expr = expr.with_columns(
    pl.when(pl.col("gene").is_in(["GAPDH", "ACTB"])).then(
        pl.lit("housekeeping")).otherwise(pl.lit("target")).alias("role"))
expr.group_by("role").agg(pl.len().alias("n")).sort("role")

shape: (2, 2)
┌──────────────┬─────┐
│ role         ┆ n   │
│ ---          ┆ --- │
│ str          ┆ u32 │
╞══════════════╪═════╡
│ housekeeping ┆ 16  │
│ target       ┆ 32  │
└──────────────┴─────┘

Sixteen housekeeping rows, two genes across eight samples, and thirty-two targets. Both languages agree. Next, Missing Values handles the gaps these tests silently create.