Optimising Multi-Protein Interaction Screening through Machine Learning Algorithm Development [Enhancing Förster/Fluorescence Resonance Energy Transfer (FRET) Detection in Flow Cytometry]
- Publication Type:
- Thesis
- Issue Date:
- 2023
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Förster or fluorescence resonance energy transfer (FRET) is a widely utilised technique to analyse protein-protein interactions (PPIs) for exploring biological processes. While microscopy was traditionally used to measure FRET, flow cytometry-based FRET has become more prominent in the last decade. Flow cytometry allows the multichannel high-throughput examination of FRET with great sensitivity and statistically robust quantification in many samples. As a superior alternative to traditional manual analysis, machine learning (ML) has attracted increasing interest in modern cytometry data analysis. It can automatically perform objective data-oriented investigations from large datasets with minimal human interventions.
Moreover, current FRET analyses are limited to chemically linked molecules for FRET calibration and lack high-performance data pre-processing. To achieve absolute single-cell quantification of natural (chemically unlinked) PPIs, this study established ML-powered algorithms for flow cytometry-based FRET detection of multi-protein interactions. This study presents designs, validations, performance tests, and application demonstrations of the algorithms in a FRET analysis workflow. The UltraFast algorithm presented F1 score over 0.91 for singlet data identification. The collaborative filtering-based algorithms' performance demonstrated error rates below 0.01% for both the correction of baseline subtraction error and autofluorescence prediction. Moreover, the downstream spectral unmixing process accomplished near zero (as low as 0.412) residual spillover and near zero (as low as 0.620) spread error, demonstrating more than 1000 folds of improvement compared to spectral unmixing without the abovementioned pre-processing steps. Together, the pre-processing pipeline with these developed algorithms achieved unbiased, accurate and robust flow cytometry data pre-processing, including singlet identification, fluorescence background-subtraction-error correction, autofluorescence prediction and removal, and FRET spectral unmixing.
The pre-processed flow cytometry data further allowed the complete quantification of two-protein three-colour FRET, three-protein six-colour FRET, and investigation of multiple simultaneous intercellular signalling activities. In particular, the FRET calibration and FRET efficiency are improved to the absolute single-cell level quantification compatible with chemically-unliked free-interacting molecules compared to the current method using linked calibration approaches. The new FRET analysis approaches have been tested on five different machine models, including conventional and full-spectrum flow cytometers, and validated using data generated from eleven FRET experiments. Further utilising the compositional data analysis (CoDA) techniques, this study also provided interpretations for the dynamic FRET energy competitions and the compositional activation levels of multiple cellular signalling pathways. In conclusion, this study provides powerful solutions for clinical diagnostics and therapeutic screenings, enabling the search for the next-generation PPI-specific and signalling pathway-specific cures for many human diseases.
Please use this identifier to cite or link to this item: