Skip to contents

QC-Based Drift Correction Using LOESS

Source: R/drift-correction.R
step_measure_drift_qc_loess.Rd

step_measure_drift_qc_loess() creates a specification of a recipe step that corrects for signal drift across run order using QC (or reference) samples. This implements the QC-RLSC (robust LOESS signal correction) method.

Usage

step_measure_drift_qc_loess(
  recipe,
  ...,
  run_order_col = "run_order",
  sample_type_col = "sample_type",
  qc_type = "qc",
  apply_to = c("all", "unknown"),
  span = 0.75,
  degree = 2,
  robust = TRUE,
  min_qc = 5,
  role = NA,
  trained = FALSE,
  skip = FALSE,
  id = recipes::rand_id("measure_drift_qc_loess")
)

Arguments

recipe

A recipe object.

...

One or more selector functions to choose feature columns. For feature-level data, select the numeric response columns. For curve-level data with .measures, leave empty to apply to all locations.

run_order_col

Name of the column containing run order (injection sequence). Must be numeric/integer.

sample_type_col

Name of the column containing sample type.

qc_type

Value(s) in sample_type_col that identify QC samples to use for drift modeling. Default is "qc".

apply_to

Which samples to apply correction to:

  • "all" (default): Correct all samples

  • "unknown": Only correct unknown samples

span

LOESS span parameter controlling smoothness. Default is 0.75. Smaller values = more flexible fit.

degree

Polynomial degree for LOESS (1 or 2). Default is 2.

robust

Logical. Use robust LOESS fitting? Default is TRUE.

min_qc

Minimum number of QC samples required. Default is 5.

role

Not used by this step.

trained

Logical indicating if the step has been trained.

skip

Logical. Should the step be skipped when baking?

id

Unique step identifier.

Value

An updated recipe with the new step added.

Details

How It Works

  1. During prep(): A LOESS model is fit to QC sample responses vs run order for each feature/location.

  2. During bake(): Correction factors are calculated as: correction = median(QC_responses) / predicted_value

    Each sample's response is multiplied by the correction factor at its run order position.

Data Levels

This step supports both:

  • Feature-level data: Applies correction to each selected numeric column

  • Curve-level data: Applies correction to each location in the measure_list

Diagnostics

The trained step stores drift model information accessible via tidy():

  • LOESS model parameters

  • QC response trends

  • Correction factors applied

See also

measure_detect_drift() for drift detection before correction.

Other drift-correction: step_measure_drift_linear(), step_measure_drift_spline(), step_measure_qc_bracket()

Examples

library(recipes)

# Feature-level data with drift
data <- data.frame(
  sample_id = paste0("S", 1:20),
  sample_type = rep(c("qc", "unknown", "unknown", "unknown", "qc"), 4),
  run_order = 1:20,
  feature1 = 100 + (1:20) * 0.5 + rnorm(20, sd = 2),  # Upward drift
  feature2 = 50 - (1:20) * 0.3 + rnorm(20, sd = 1)    # Downward drift
)

rec <- recipe(~ ., data = data) |>
  update_role(sample_id, new_role = "id") |>
  step_measure_drift_qc_loess(feature1, feature2) |>
  prep()

corrected <- bake(rec, new_data = NULL)