Development of Computational and Statistical Methods for Fluorescence-Based Diffusion Measurements

Molecular diffusion plays a fundamental role in cellular physiology, yet accurate measurement of diffusion coefficients in living cells remains challenging due to limitations in existing analysis methods and phototoxicity constraints from prolonged laser exposure. This thesis addresses these challenges through development of two computational approaches for fluorescence–based diffusion measurements: IOCBIO FCS, an enhanced platform for correlation spectroscopy analysis, and fluorescence intensity trace statistical analysis (FITSA), a novel method based on direct Bayesian analysis of intensity traces.

Fluorescence correlation spectroscopy (FCS) and raster image correlation spectroscopy (RICS) are established techniques for measuring molecular dynamics through analysis of fluorescence intensity fluctuations. However, current analysis methods suffer from critical limitations. More fundamentally, conventional fitting approaches based on OLS incorrectly assume identical error variance across all autocorrelation function (ACF) data points and statistical independence between measurements, leading to underestimated uncertainties and unreliable model selection.

To address these software limitations in correlation-based methods, we developed IOCBIO FCS, a unified open–source Python platform integrating GPU-accelerated autocorrelation computation, comprehensive statistical inference frameworks (Bayesian and frequentist approaches with ordinary, weighted, and generalized least-squares error treatment), support for experimentally measured 3D PSFs, and capabilities for multiple-angle RICS analysis at variable scanning speeds. The platform enables spatial mapping of diffusion parameters across heterogeneous samples and characterization of anisotropic diffusion in organized cellular structures. Validation studies demonstrate that proper error modeling substantially affects parameter uncertainty estimates and model selection outcomes, with neglect of correlation structure leading to unreliable conclusions about system complexity.

The platform implements diffusion-concentration pre-analysis for quality control, enabling identification and exclusion of measurements contaminated by artifacts such as large fluorescent aggregates or unstable fluorescence before main parameter estimation. For RICS measurements, image–splitting strategies combined with concentration-based filtering systematically detect and remove images affected by rare high-intensity events that would otherwise distort spatial correlation analysis. Application of these RICS pre-analysis capabilities to study lipoprotein lipase oligomerization demonstrated practical utility of the filtering algorithms for characterizing diffusion in samples containing mixtures of different oligomeric states in the presence of large particle artifacts.

Proper treatment of correlated and heteroscedastic errors through covariance estimation requires thousands of repeated measurements under stable conditions – practically infeasible in live-cell experiments due to phototoxicity and cell viability constraints. To address the fundamental photon efficiency limitations of correlation-based techniques, we developed FITSA, a Bayesian method that analyzes fluorescence intensity traces directly rather than through derived ACFs. Key innovations include adaptive signal binning that adjusts temporal resolution based on local photon emission rates, subsection segmentation treating particle transits as independent events, and strategic seed point selection at maximum photon emission enabling efficient posterior sampling. Comparative analysis demonstrates that FITSA achieves convergence with substantially fewer iterations than earlier direct intensity analysis implementation, providing faster computation with improved robustness to prior specification. When comparing to FCS with rigorous covariance estimation – which requires thousands of repeated measurements – FITSA estimates diffusion coefficients with substantially fewer total photons, with advantages more pronounced for slower diffusion characteristic of intracellular environments. This substantial photon reduction enables shorter measurement durations, directly addressing phototoxicity concerns that limit live-cell applications of correlation spectroscopy.

These complementary tools address different experimental needs: IOCBIO FCS for complex scenarios including multi-component diffusion, triplet-state dynamics, and anisotropic transport; FITSA for photon-limited measurements where phototoxicity is critical.

Supervisors

• Supervisor: Marko Vendelin
• Co-supervisor: Martin Laasmaa

Opponents

• Assistant Professor Falk Schneider, Fluorescence and Membrane Dynamics (FMD) Lab, University of Warwick, Coventry, UK 
• Dr. Sergei Kopantšuk, Institute of Chemistry, University of Tartu, Tartu, Estonia

Time of defense

12 May 2026 at 10:00 at the Department of Cybernetics, room U2-303

Thesis

You can download PDF of the thesis at https://doi.org/10.23658/taltech.26/2026