Autoencoder Spectral Unmixing for Single-Cell Raman Analysis

Emil Alstrup Jensen, Lasse Ebdrup Pedersen, Anders Kristensen, Line H Clemmensen

Keywords: Deep Learning, Autoencoders, Spectral Unmixing, Raman Spectroscopy, Single-Cell

TL;DR: We apply physics-constrained autoencoders with volume regularization to unmix single-cell Raman spectra, guiding latent dimensionality and promoting diverse, interpretable endmembers

Abstract: Raman spectroscopy provides label-free, holistic molecular information at the single-cell level, but spectra are complex and challenging to interpret. We apply physics-constrained autoencoders with volume regularization to unmix single-cell Raman spectra, guiding latent dimensionality and promoting diverse, interpretable endmembers. On THP-1 and NK cell datasets, this approach improved peak definition, chemical interpretability, and captured biologically relevant variability.

Serve As Reviewer: ~Emil_Alstrup_Jensen1

Submission Number: 30