Fit sliced random-projection Gaussian LRT discriminator — fit_ot

Learns a frozen two-class discriminator from training data by projecting each specimen's within-sample robust centered log-ratio (rCLR) vector onto deterministic random one-dimensional slices. Along every slice, the case and control training anchors are summarized by frozen Gaussian moments, and scoring uses the average Gaussian log-likelihood ratio. Larger scores are more case-like.

The rCLR transform is computed separately within each specimen: z[pos] = log(v[pos]) - mean(log(v[pos])) for positive abundances and z[!pos] = 0. This makes the representation invariant to multiplying a specimen by a positive scalar, without estimating any statistic from the scoring batch.

Usage

fit_ot_slicedlrt(X_train, y_train, meta_train = NULL, hp = list())

Arguments

X_train: Numeric matrix (samples $\times$ features) of non-negative abundance values. Columns must be uniquely named feature ids.
y_train: Integer/numeric 0/1 labels aligned to X_train; 1 is case/disease and 0 is control.
meta_train: Optional per-sample metadata. Accepted for interface uniformity and ignored by this method.
hp: List of frozen hyperparameters: n_slices positive integer number of random projection directions (default 100L); max_anchors_per_class positive integer anchor cap per class (default 200L); min_features positive integer feature-overlap floor for fitting and scoring (default 3L); eps positive finite variance floor (default 1e-6); and seed non-negative integer used for deterministic anchor subsampling and random directions while restoring the global RNG state (default 1L).

Value

A plain list of class ot_slicedlrt_model containing the frozen feature_universe, raw case_anchors, raw control_anchors, the full-universe sliced Gaussian representation repr, and resolved hp.

Details

Fitting stores raw training anchors rather than only fitted slice moments so that scoring can re-derive the same representation over the feature subset present at inference. The helper .ot_slicedlrt_repr() is used both at fit time over the full training feature universe and at score time over intersect(model$feature_universe, colnames(X)). Its only inputs are frozen anchors, frozen hyperparameters, and the present-feature dimension. Consequently, no test-batch centering, quantile, variance, bandwidth, or other cross-row statistic is estimated from scored specimens. If the present overlap contains fewer than hp$min_features features, score_ot_slicedlrt returns the documented neutral score 0 for every row.

References

Bonneel N, Rabin J, Peyre G, Pfister H. (2015) Sliced and Radon Wasserstein barycenters of measures. Journal of Mathematical Imaging and Vision 51: 22-45.

Neyman J, Pearson ES. (1933) On the problem of the most efficient tests of statistical hypotheses. Philosophical Transactions of the Royal Society of London. Series A 231: 289-337.

Examples

if (FALSE) { # \dontrun{
set.seed(1)
X <- matrix(stats::rgamma(80 * 30, shape = 2), nrow = 80,
            dimnames = list(NULL, paste0("miR-", seq_len(30))))
y <- rep(c(0, 1), each = 40)
X[y == 1, 1:6] <- X[y == 1, 1:6] + 25
X[y == 0, 7:12] <- X[y == 0, 7:12] + 25
model <- fit_ot_slicedlrt(X, y)
score <- score_ot_slicedlrt(model, X)
} # }