Nature-Inspired Compressed Sensing for Transcriptomic Profiling From Random Composite Measurements.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Transcriptomic profiling measures gene expression, with technologies like single-cell RNA-Seq generating massive datasets.
• A key challenge lies in relating complex transcriptomic profiles to simpler, composite measurements.
• High-dimensional gene expression data analysis requires advanced computational methods.

Purpose of the Study:

• To develop a mathematical framework for reconstructing high-dimensional gene expression data from low-dimensional measurements.
• To address the challenge of exploiting relationships between transcriptomic profiles and random composite measurements.
• To improve the accuracy and efficiency of gene expression data analysis.

Main Methods:

• Proposed a novel mathematical framework named DECS (Differential Evolution with Compressed Sensing).
• Utilized differential evolution for global search and compressed sensing for local search.
• Leveraged the sparse nature of gene expression data to learn module dictionaries and levels.

Main Results:

• DECS successfully reconstructed high-dimensional gene expression data from low-dimensional inputs, achieving significant orders of magnitude improvement (e.g., 200x).
• Experimental comparisons showed DECS outperformed three benchmark methods in recovering gene expression patterns.
• Extensive benchmarks on nine Gene Expression Omnibus (GSE) datasets validated DECS's performance and sensitivity.

Conclusions:

• DECS provides a robust and effective method for gene expression data reconstruction.
• The framework offers valuable mechanistic insights by revealing underlying gene expression patterns.
• DECS represents a significant advancement in analyzing and interpreting complex transcriptomic data.