Fit the structurally-informed ILR single-sample scorer (ss-struct-ilr)

Builds a genomic-cluster / GC-tertile Sequential Binary Partition from the training feature set + a miRNA annotation table, computes ILR balances per specimen via ws_balance_ilr, standardises them with train-only mean/sd, and learns a frozen diagonal-LDA linear rule (per-balance standardised case-minus-control mean difference). The returned object serialises every artefact needed to score one new specimen with no test-batch information.

Usage

fit_ss_struct_ilr(X_train, y_train, meta_train = NULL, annotation, hp = list())

Arguments

X_train

Numeric matrix (samples x features) of non-negative abundances; columns are uniquely named miRNA ids.

y_train

0/1 labels aligned to X_train (1 = case).

meta_train

Optional per-sample metadata; accepted for interface uniformity and ignored.

annotation

data.frame with columns feature (miRNA id), cluster (genomic-cluster id), gc (GC fraction).

hp

List of frozen hyperparameters: pseudocount (default NULL = the scale-invariant per-sample default of ws_balance_ilr), aggregate ("gmean"/"trimmed_gmean", default "gmean"), min_balance_coverage (default 0.5), min_part (default 1L).

Value

A ss_struct_ilr_model list: frozen sbp, balance_names, center, scale, weights, feature_universe, the resolved hp, and an annotation_features record.

See also

score_ss_struct_ilr, ws_balance_ilr