The art of PCR assay development: data-driven multiplexing

Abstract

The present thesis describes the discovery and application of a novel methodology, named Data-Driven Multiplexing, which uses artificial intelligence and conventional molecular instruments to develop rapid, scalable and cost-effective clinical diagnostic tests.

Detection of genetic material from living organisms is a biologically engineered process where organic molecules interact with each other and with chemical components to generate a meaningful signal of the presence, quantity or quality of target nucleic acids. Nucleic acid detection, such as DNA or RNA detection, identifies a specific organism based on its genetic material. In particular, DNA amplification approaches, such as for antimicrobial resistance (AMR) or COVID-19 detection, are crucial for diagnosing and managing various infectious diseases. One of the most widely used methods is Polymerase Chain Reaction (PCR), which can detect the presence of nucleic acids rapidly and accurately. The unique interaction of the genetic material and synthetic short DNA sequences called primers enable this harmonious biological process. This thesis aims to bioinformatically modulate the interaction between primers and genetic material, enhancing the diagnostic capabilities of conventional PCR instruments by applying artificial intelligence processing to the resulting signals.

To achieve the goal mentioned above, experiments and data from several conventional platforms, such as real-time and digital PCR, are used in this thesis, along with state-of-the-art and innovative algorithms for classification problems and final application in real-world clinical scenarios. This work exhibits a powerful technology to optimise the use of the data, conveying the following message: the better use of the data in clinical diagnostics enables higher throughput of conventional instruments without the need for hardware modification, maintaining the standard practice workflows.

In Part I, a novel method to analyse amplification data is proposed. Using a state-of-the-art digital PCR instrument and multiplex PCR assays, we demonstrate the simultaneous detection of up to nine different nucleic acids in a single-well and single-channel format. This novel concept called Amplification Curve Analysis (ACA) leverages kinetic information encoded in the amplification curve to classify the biological nature of the target of interest. This method is applied to the novel design of PCR assays for multiple detections of AMR genes and further validated with clinical samples collected at Charing Cross Hospital, London, UK. The ACA showed a high classification accuracy of 99.28% among 253 clinical isolates when multiplexing. Similar performance is also demonstrated with isothermal amplification chemistries using synthetic DNA, showing a 99.9% of classification accuracy for detecting respiratory-related infectious pathogens.

In Part II, two intelligent mathematical algorithms are proposed to solve two significant challenges when developing a Data-driven multiplex PCR assay. Chapter 7 illustrates the use of filtering algorithms to remove the presence of outliers in the amplification data. This demonstrates that the information contained in the kinetics of the reaction itself provides a novel way to remove non-specific and not efficient reactions. By extracting meaningful features and adding custom selection parameters to the amplification data, we increase the machine learning classifier performance of the ACA by 20% when outliers are removed. In Chapter 8, a patented algorithm called Smart-Plexer is presented. This allows the hybrid development of multiplex PCR assays by computing the optimal single primer set combination in a multiplex assay. The algorithm's effectiveness stands in using experimental laboratory data as input, avoiding heavy computation and unreliable predictions of the sigmoidal shape of PCR curves. The output of the Smart-Plexer is an optimal assay for the simultaneous detection of seven coronavirus-related pathogens in a single well, scoring an accuracy of 98.8% in identifying the seven targets correctly among 14 clinical samples. Moreover, Chapter 9 focuses on applying novel multiplex assays in point-of-care devices and developing a new strategy for improving clinical diagnostics.

In summary, inspired by the emerging requirement for more accurate, cost-effective and higher throughput diagnostics, this thesis shows that coupling artificial intelligence with assay design pipelines is crucial to address current diagnostic challenges. This requires crossing different fields, such as bioinformatics, molecular biology and data science, to develop an optimal solution and hence to maximise the value of clinical tests for nucleic acid detection, leading to more precise patient treatment and easier management of infectious control.