Signalling Entropy Across Measurement Scales: A Compositional Dilution Lemma and Cross-Modality Invariance for...

Area of Science:

• Computational Biology
• Information Theory
• Genomics

Background:

• Cancer transcriptomic dysregulation is complex and varies across measurement types.
• Existing analytical frameworks may not fully capture the multifaceted nature of gene expression changes in cancer.

Purpose of the Study:

• To develop a unified information-theoretic framework for analyzing cancer transcriptomic dysregulation.
• To establish methods for assessing cross-modality invariance and prognostic value of transcriptomic features.

Main Methods:

• Utilized Shannon entropy, signalling entropy rate, and Gaussian total correlation to quantify gene expression aspects.
• Developed a linear-time algorithm for signalling entropy rate and a Compositional Dilution Lemma.
• Employed large-scale datasets (700,202 single cells, 3942 bulk samples) across five cancer types for validation.

Main Results:

• Demonstrated significant pan-cancer tumor elevation in transcriptomic dysregulation (p<10-7).
• Validated cross-modality invariance, with null hypothesis not rejected (p>0.5).
• Prognostic Cox regressions showed good concordance (0.696±0.018) and immune deconvolution identified distinct cancer subtypes.

Conclusions:

• The developed framework provides a robust method for analyzing cancer transcriptomic data across modalities.
• Findings support the invariance of transcriptomic features across measurement types and highlight their prognostic potential.
• The study successfully partitions cancers into myeloid- or lymphoid-driven classes based on immune deconvolution.