NMF-RI

NMF-RI performs unsupervised blind spectral unmixing of fluorescent emissions in multispectral flow and image cytometry data to separate fluorochrome signals for multiplex immunostaining analyses.

Key Features:

• Unsupervised blind spectral unmixing: Separates fluorescent emissions without requiring control samples, avoiding characterization of individual fluorochromes.
• Non-negative Matrix Factorization (NMF) enhancements: Builds on NMF with initialization based on theoretical spectra, automated selection of sparse data, and a re-initialized multilayer optimizer to improve convergence and accuracy.
• Robust testing and validation: Evaluated on synthetic datasets across varying colocalization levels, signal-to-noise ratios, spectral resolution, and initialization errors.
• Performance comparison: Benchmarked against traditional spectral unmixing algorithms on multispectral flow and image cytometry systems and reported to consistently outperform them on highly spatially and spectrally mixed data.
• Low computational cost: Designed to reduce computational requirements for large-scale unmixing tasks.

Scientific Applications:

• Multiplex immunostaining preprocessing: Enables unmixing of multiplex immunostaining samples without experimental control samples to support downstream analyses of multiple biomarkers.
• Multispectral flow cytometry: Applicable to unmixing highly mixed liquid cytometry datasets.
• Multispectral image cytometry and tissue imaging: Applicable to unmixing highly mixed spatially colocalized signals in solid-sample imaging datasets.
• Complex biomarker studies: Supports analysis of numerous biomarkers in complex biological systems where spectral and spatial overlap is extensive.

Methodology:

Non-negative Matrix Factorization with initialization based on theoretical spectra, automated selection of sparse data, and a re-initialized multilayer optimizer; validation and benchmarking performed using synthetic datasets varying colocalization levels, signal-to-noise ratios, spectral resolution, and initialization errors, and compared to traditional spectral unmixing algorithms on multispectral flow and image cytometry systems.