Fit robust-regression hemolysis nuisance model from training controls

Fits per-feature Huber M-estimation robust regressions (MASS::rlm) of log-expression on a hemolysis proxy score (and optionally a platelet proxy). The resulting intercept and slope coefficients are frozen at fit time. At deployment, the predicted nuisance contribution is subtracted from test-sample log-expressions via apply_hemolysis_rr.

Failed fits (convergence failures, NA coefficients) fall back to a zero-slope model (intercept = feature mean) and are flagged in the stable vector.

Usage

fit_hemolysis_rr(
  X_train,
  hemo_score,
  platelet_score = NULL,
  method = c("huber", "rlm"),
  clip_max = 2
)

Arguments

X_train: Numeric matrix (samples \(\times\) features). Ideally a mix of clean and mildly contaminated training controls.
hemo_score: Numeric vector of length nrow(X_train). Hemolysis proxy score per training sample.
platelet_score: Optional numeric vector of length nrow(X_train). Platelet contamination proxy. Default NULL (no platelet correction).
method: "huber" (default, MASS::rlm with psi.huber) or "rlm" (alias).
clip_max: Maximum absolute correction per feature in log-units. Default 2. Prevents over-correction from unstable slope estimates.

Value

Object of class hemolysis_rr_fit with components: betas (features \(\times\) coefficients matrix), stable, has_platelet, clip_max, train_hemo_range, train_platelet_range, method, n_train.

References

Huber PJ. (1964) Robust Estimation of a Location Parameter. The Annals of Mathematical Statistics 35(1): 73–101.

Examples

if (FALSE) { # \dontrun{
set.seed(42)
X <- matrix(rlnorm(100 * 20, 5, 0.8), nrow = 100,
            dimnames = list(NULL, paste0("mir-", seq_len(20))))
hemo <- runif(100)
fit <- fit_hemolysis_rr(X, hemo)
fit$n_train
} # }