debCAM: a bioconductor R package for fully unsupervised deconvolution of complex tissues

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Dissecting complex tissues into distinct cellular or tissue subtypes is crucial for understanding biological processes.
• Bulk expression profiles often obscure the contributions of individual subtypes.
• Existing deconvolution methods may require prior knowledge or lack comprehensive subtype characterization.

Purpose of the Study:

• To develop a fully unsupervised deconvolution method for dissecting complex tissues into molecularly distinctive subtypes.
• To implement an R package, deconvolution by Convex Analysis of Mixtures (debCAM), for automated marker detection, subtype number determination, proportion calculation, and profile estimation.
• To demonstrate the utility of debCAM across diverse molecular data types.

Main Methods:

• Developed a novel unsupervised deconvolution algorithm based on convex analysis of mixtures.
• Implemented the algorithm as an R package, debCAM.
• Validated debCAM using gene expression, methylation, proteomics, and imaging data.

Main Results:

• debCAM successfully identifies tissue/cell-specific markers and determines the number of constituent subtypes.
• The package accurately calculates subtype proportions and estimates cell/tissue-specific expression profiles.
• Demonstrated robust performance and biomedical utility across multiple data modalities.

Conclusions:

• The debCAM software tool offers a comprehensive and unbiased approach for characterizing tissue remodeling.
• Enhanced data preprocessing and prior knowledge incorporation will further improve its applicability in various biomedical contexts.
• This unsupervised method facilitates deeper biological insights from bulk expression data.