spqrp 0.1.0

Initial CRAN-targeted release. Native R port of the Python spqrp package.

Features

• Clustering pipeline (run_clustering()) – kNN graph in a PCA/UMAP/MDS embedding, optional iterative split of large components, ggplot visualization with patient-hue colouring and legend.
• Threshold-based evaluation (perform_distance_evaluation_on_ranked_proteins()) – pairwise sample classification from a percentile cutoff on pairwise distances, with FN/FP/percentile-overlay histogram.
• Protein ranking (train_with_normalise()) – pairwise random- forest classifier with three selectable backends; randomForest is the default (closest behaviour to Python’s imblearn.BalancedRandomForestClassifier). Importance values are normalised to sum to 1.0, matching sklearn’s clf.feature_importances_ convention.
• Isolation-forest outlier filtering (remove_outlier_samples()) – pure-R via the solitude package; default outlier_threshold calibrated empirically for solitude’s anomaly-score scale.

Verbosity

All functions are silent by default. Pass quiet = FALSE to any function that emits status output to see progress messages, per-call summaries, save-path hints, and cluster listings. Warnings about genuine data issues – e.g. samples dropped from analysis – fire regardless of quiet.

Documentation

• See articles/numerical-divergence.md for known cross-language divergences (UMAP, random-forest backends, isolation-forest scales, MDS solvers) and recommendations for cross-language comparison.
• See vignette("spqrp-mock-data") for a worked example on a small bundled cohort.