IntrinsicScan

IntrinsicScan is a Bioconductor-oriented implementation of a label-blind, feature-wise intrinsic binary partitioning framework.

For each continuous feature, IntrinsicScan:

  1. fits an exact one-dimensional two-medoid partition under L1 loss using only the feature values;
  2. reveals the binary phenotype only after the partition is fixed;
  3. quantifies orientation-free phenotype concordance using Youden J;
  4. provides an exact conditional hypergeometric test for the fixed-partition J;
  5. optionally evaluates resampling stability with deterministic, feature-specific random-number streams; and
  6. optionally performs fixed-partition genome-wide permutation calibration.

The package is designed to separate intrinsic boundary discovery from phenotype concordance, inferential evidence, and resampling stability. It is not a supervised cutoff optimizer, a general clustering algorithm, or a bimodality test. A fitted binary partition does not by itself establish the existence of two discrete biological states.

Bioconductor-style input

The main function accepts a feature-by-sample matrix, SummarizedExperiment, or ExpressionSet. It returns a SummarizedExperiment whose rows are ordered by the IntrinsicScan ranking.

fit <- intrinsicScan(se, phenotype = "group", positive = "case")
rowData(fit)
intrinsicResults(fit)

No normalization or missing-value imputation is performed. Input measurements must already be preprocessed, continuous, complete, and finite.

Installation

IntrinsicScan is currently being prepared for Bioconductor submission. The source repository is available at:

https://github.com/memedesimo98/IntrinsicScan

After release through Bioconductor, installation will use:

if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
BiocManager::install("IntrinsicScan")

Development status

Version 0.99.3 is a pre-submission Bioconductor development version. The validated numerical engine is intentionally kept stable while package infrastructure, documentation, and submission metadata are finalized.

AI-assisted development disclosure

Development of IntrinsicScan was assisted by OpenAI ChatGPT (GPT-5.6 Sol). The model contributed to iterative method critique and development, R package engineering, testing and validation strategy, Bioconductor compliance, documentation, and scientific drafting.

All methodological decisions, interpretation of results, validation of the implementation, and responsibility for the accuracy, reliability, licensing, and maintenance of the package remain with the human authors.

Assisted-by: OpenAI ChatGPT (GPT-5.6 Sol)

License

IntrinsicScan is distributed under the MIT license. See LICENSE for details.