RAD

RAD: Deconvolution and Phylogenetic Inference of Tumor Transcriptomes

RAD performs deconvolution of non-negative bulk RNA expression matrices B ∈ R+m × n to infer a cell community expression profile matrix C ∈ R+m × k and a fraction matrix F ∈ R+k × n, enabling reconstruction of tumor subpopulations and their evolutionary relationships.

Key Features:

• Matrix Deconvolution: Infers cell community expression profiles and fractional compositions from bulk RNA expression data.
• Gene Module Compression: Reduces noise and variability in RNA data to improve robustness and signal clarity.
• Hybrid Optimization Algorithm: Optimizes deconvolution for balanced robustness and accuracy in tumor subpopulation inference.
• Phylogenetic Analysis: Applies phylogenetic algorithms to model evolutionary adaptations across metastatic transitions by analyzing changes in expression programs and cell-type composition.

Scientific Applications:

• Clonal Evolution in Metastasis: Reconstructs tumor subpopulation dynamics from bulk transcriptomic data across progression states to identify evolutionary patterns and pathway dysregulation, including ECM-receptor interaction, focal adhesion, and PI3K-Akt signaling.

Methodology:

RAD processes bulk RNA expression matrices, applies gene module compression to reduce noise, performs hybrid optimization to estimate C and F matrices, and conducts phylogenetic inference to characterize evolutionary changes in expression programs and cellular composition during metastasis.