From photons to phenotypes: decoding organoid biochemistry via deep learning-enhanced raman spectroscopy
File(s)
Author(s)
Georgiev, Dimitar Dimitrov
Type
Thesis
Abstract
Three-dimensional neural organoids have emerged as powerful in vitro models for studying human brain development, disease and drug response. Yet their analysis remains constrained by standard imaging and characterisation techniques, which are invasive, require exogenous labelling and offer limited multiplexing. There is a pressing need for advanced analytical and imaging approaches that can interrogate spatial organisation and dynamics in intact organoid systems under physiologically relevant settings, without prior knowledge of specific molecular targets or the use of external labels.
To address this challenge, this thesis develops computational methods for non-invasive, label-free biochemical analysis using Raman spectroscopy (RS). First, it introduces RamanSPy – an open-source Python package for RS data analysis designed to improve standardisation, reproducibility, accessibility and interoperability. RamanSPy provides a comprehensive library of tools for spectroscopic analysis that supports day-to-day tasks, integrative analyses, method and protocol development, and the integration of advanced data analytics. Next, hyperspectral unmixing algorithms for RS based on autoencoder neural networks are developed and systematically validated using synthetic and experimental benchmark datasets. Unmixing autoencoders consistently outperform standard methods, achieving more accurate qualitative and quantitative analysis and enabling enhanced volumetric Raman imaging of a monocytic cell.
Building on these tools, a Raman imaging platform for unsupervised, spatially resolved biochemical analysis in neural organoids is established. This approach enables non-invasive, label-free mapping of cellular and subcellular structures in both cryosectioned and intact organoids. Using this platform, volumetric imaging of a neural rosette within a neural organoid is demonstrated. Furthermore, changes in biochemical composition during early developmental stages in intact neural organoids are interrogated, revealing spatiotemporal variations in lipids, proteins and nucleic acids.
Overall, this work expands the methodological and computational toolkit for RS chemometrics and establishes a versatile framework for high-content, label-free (bio)chemical phenotyping with broad applications in organoid research, cell biology, disease modelling and drug discovery.
To address this challenge, this thesis develops computational methods for non-invasive, label-free biochemical analysis using Raman spectroscopy (RS). First, it introduces RamanSPy – an open-source Python package for RS data analysis designed to improve standardisation, reproducibility, accessibility and interoperability. RamanSPy provides a comprehensive library of tools for spectroscopic analysis that supports day-to-day tasks, integrative analyses, method and protocol development, and the integration of advanced data analytics. Next, hyperspectral unmixing algorithms for RS based on autoencoder neural networks are developed and systematically validated using synthetic and experimental benchmark datasets. Unmixing autoencoders consistently outperform standard methods, achieving more accurate qualitative and quantitative analysis and enabling enhanced volumetric Raman imaging of a monocytic cell.
Building on these tools, a Raman imaging platform for unsupervised, spatially resolved biochemical analysis in neural organoids is established. This approach enables non-invasive, label-free mapping of cellular and subcellular structures in both cryosectioned and intact organoids. Using this platform, volumetric imaging of a neural rosette within a neural organoid is demonstrated. Furthermore, changes in biochemical composition during early developmental stages in intact neural organoids are interrogated, revealing spatiotemporal variations in lipids, proteins and nucleic acids.
Overall, this work expands the methodological and computational toolkit for RS chemometrics and establishes a versatile framework for high-content, label-free (bio)chemical phenotyping with broad applications in organoid research, cell biology, disease modelling and drug discovery.
Version
Open Access
Date Issued
2025-10-04
Date Awarded
2026-02-01
Copyright Statement
Attribution-NonCommercial 4.0 International Licence (CC BY-NC)
License URL
Advisor
Barahona, Mauricio
Stevens, Molly
Sponsor
UK Research and Innovation
Grant Number
EP/S023283/1
Publisher Department
Department of Computing
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)