Optimising multiplex molecular diagnostics for infectious diseases through advanced signal processing and machine learning
File(s)
Author(s)
Xu, Ke
Type
Thesis
Abstract
Rapid, low-cost Nucleic Acid Amplification Tests (NAAT) are essential for the diagnosis of infectious diseases. However, existing multiplexing polymerase chain reaction (PCR) or loop-mediated isothermal amplification (LAMP) tests struggle with reliability and cost. Amplification Curve Analysis (ACA) offers a data-driven alternative for single-tube multiplex tests, but kinetic factors hinder its reliability. Addressing these data and instrumentation challenges motivates the data-driven multiplexing framework advanced in this thesis.
In addition, colorimetric LAMP (cLAMP) offers an inexpensive, equipment-light choice for point-of-care NAAT. However, fully automating its readout remains difficult, highlighting the need for a universal, hardware-independent colour-detection method, which serves the other aim of the thesis.
This thesis charts data-driven approaches to multiplexing pathways from amplification curves to colorimetric tests for the detection of infectious diseases, integrating signal processing, machine learning and biochemical design. Chapter 1 familiarises readers with background information and elaborates on the challenges for data-driven implementation in NAAT. Chapter 2 presents an Adaptive Mapping Filter that removes non-specific and low-efficiency digital PCR curves, providing a standard data processing procedure. Chapter 3 delivers Smart-Plexer 2.0, using robust kinetic features and clustering to automate single-well multiplex PCR assay selection. Chapter 4 tackles the domain shift of PCR data with a transformer-based adversarial network, enabling label-free adaptation. Chapter 5 automates cLAMP by pairing deep image segmentation with adaptive colour identification, achieving 99% accuracy on clinical samples. Chapter 6 re-engineers LAMP kinetics with a primer-ratio optimisation, generating distinct Time-to-Positive for separate targets, expanding single-channel multiplexing in real-time LAMP.
In conclusion, this thesis advances multiplex diagnostics from biochemical and informatics perspectives. It demonstrates the integration of adaptive filtering, machine-driven assay selection, domain-robust learning, and machine learning-based automation for improving diagnostic precision and accessibility, providing a blueprint for next-generation assays that are not only more accurate but also smarter—capable of learning from every output they generate.
In addition, colorimetric LAMP (cLAMP) offers an inexpensive, equipment-light choice for point-of-care NAAT. However, fully automating its readout remains difficult, highlighting the need for a universal, hardware-independent colour-detection method, which serves the other aim of the thesis.
This thesis charts data-driven approaches to multiplexing pathways from amplification curves to colorimetric tests for the detection of infectious diseases, integrating signal processing, machine learning and biochemical design. Chapter 1 familiarises readers with background information and elaborates on the challenges for data-driven implementation in NAAT. Chapter 2 presents an Adaptive Mapping Filter that removes non-specific and low-efficiency digital PCR curves, providing a standard data processing procedure. Chapter 3 delivers Smart-Plexer 2.0, using robust kinetic features and clustering to automate single-well multiplex PCR assay selection. Chapter 4 tackles the domain shift of PCR data with a transformer-based adversarial network, enabling label-free adaptation. Chapter 5 automates cLAMP by pairing deep image segmentation with adaptive colour identification, achieving 99% accuracy on clinical samples. Chapter 6 re-engineers LAMP kinetics with a primer-ratio optimisation, generating distinct Time-to-Positive for separate targets, expanding single-channel multiplexing in real-time LAMP.
In conclusion, this thesis advances multiplex diagnostics from biochemical and informatics perspectives. It demonstrates the integration of adaptive filtering, machine-driven assay selection, domain-robust learning, and machine learning-based automation for improving diagnostic precision and accessibility, providing a blueprint for next-generation assays that are not only more accurate but also smarter—capable of learning from every output they generate.
Version
Open Access
Date Issued
2025-06-23
Date Awarded
2025-08-01
Copyright Statement
Attribution-Non Commercial-No Derivatives 4.0 International Licence (CC BY-NC-ND)
Advisor
Rodriguez-Manzano, Jesus
Holmes, Alison
Georgiou, Pantelis
Sponsor
Imperial College London
Publisher Department
Department of Infectious Disease
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)