Single-sample within-sample compositional methods (Module A, P1)

Five Module-A within-sample compositional transforms introduced in the OmicSelector single-sample scoring bank. All methods are single-sample and reference-cohort-free: each sample is normalised, scored, or projected using only its own panel values, with no requirement for an external training cohort to exist at deployment time. This is the property that makes within-sample methods suitable for clinical translation: a diagnostic test built on these primitives can be deployed by computing a per-patient score from the patient's own panel measurement, without re-fitting any normalization on a population reference.

Methods provided:

• ws_rclr_trimmed: robust trimmed centred log-ratio with optional zero-detection-only convention; tolerant to dominant contaminating species (haemolysis miRNAs) and to small-head-of-pool compositions typical of plasma/serum miRNA panels.

• ws_balance_ilr: orthonormal isometric log-ratio balances on a frozen sequential binary partition encoding circulating miRNA biology (red-blood-cell vs rest, platelet vs rest, let-7 family within-cluster, miR-17~92, miR-200, miR-371~373, dominant-vs-tail abundance balance).

• ws_alr_pivot: additive log-ratio with a frozen pivot pool of putatively low-variance circulating miRNAs serving as the geometric-mean denominator (anchors the transformation against a stable internal reference rather than choosing a single feature as denominator).

• ws_mad_logratio: median-centred log-ratio with optional per-sample MAD scaling; not Aitchison-valid but a robust descriptive transform for downstream learners that do not require zero-sum compositional coordinates.

• ws_dominance_score: panel-internal scalar QC dominance score reporting the fraction of total panel signal carried by the top-K features; flags pre-analytical failures (haemolysis, incomplete extraction) when calibrated against a Tier R reference distribution.

Helper utilities are provided alongside: ws_default_sbp returns the curated v1 sequential-binary-partition tree for circulating miRNA panels, and ws_default_pivot_pool returns the curated v1 ALR pivot pool (six low-variance miRNAs; miR-92a-3p excluded because recent serum/plasma literature treats it as erythrocyte-derived).

Details

Background: circulating miRNA biomarkers were proposed as minimally invasive cancer markers in 2008 (Mitchell et al., PNAS). In the intervening years no clinically actionable cancer-screening test has emerged from this source despite thousands of "candidate panel" papers. The reasons are predominantly technical (batch effects, hemolysis contamination, lack of standards, reference-cohort dependency, cross-cohort leakage, absence of provenance/claim-gating) rather than biological. These five within-sample compositional methods target two of those failure modes directly: reference-cohort dependency (each method scores a single sample using only its own panel values) and contamination by dominant erythrocyte-derived species (miR-451a, miR-486-5p, miR-16, miR-144-3p, miR-92a-3p - all explicitly handled by either exclusion from the rCLR centering subset or by having dedicated balances in the SBP tree).

Module A is one of six modules in the OmicSelector v2.4 toolkit. The other modules - Module B (hemolysis correction), Module C (frozen batch correction), Module D (outlier detection and conformal claim-gating), Module E (multi-cancer / domain generalisation), and Module F (self-supervised pretraining and ablation) - are documented in the single-sample scoring bank and the per-module help pages.

References

Aitchison J. (1986) The Statistical Analysis of Compositional Data. Chapman & Hall.

Egozcue J. J., Pawlowsky-Glahn V. (2005) Groups of parts and their balances in compositional data analysis. Mathematical Geology 37(7): 795-828.

Vandeputte D., Kathagen G., D'hoe K., et al. (2017) Quantitative microbiome profiling links gut community variation to microbial load. Nature 551(7681): 507-511. (Origin of the "detected_only" rCLR convention.)

Mitchell P. S., Parkin R. K., Kroh E. M., et al. (2008) Circulating microRNAs as stable blood-based markers for cancer detection. Proceedings of the National Academy of Sciences USA 105(30): 10513-10518.