Image-based and machine learning-guided multiplexed serology test for SARS-CoV-2

Vilja Pietiäinen, Minttu Polso, Ede Migh, Christian Guckelsberger, Maria Harmati, Akos Diosdi, Laura Turunen, Antti Hassinen, Swapnil Potdar, Annika Koponen, Edina Gyukity Sebestyen, Ferenc Kovacs, Andras Kriston, Reka Hollandi, Katalin Burian, Gabriella Terhes, Adam Visnyovszki, Eszter Fodor, Zsombor Lacza, Anu Kantele, Pekka Kolehmainen, Laura Kakkola, Tomas Strandin, Lev Levanov, Olli Kallioniemi, Lajos Kemeny, Ilkka Julkunen, Olli Vapalahti, Krisztina Buzas, Lassi Paavolainen, Peter Horvath, Jussi HepojokiCell Reports Methods 2023 · OpenReview

Abstract

This paper describes a scalable and automated, high-content microscopy-based miniaturized immunofluorescence assay (mini-IFA) for the detection of SARS-CoV-2 antibodies. Unlike conventional IFA, which typically requires cells infected with live pathogens, this assay utilizes transfected cells expressing individual viral antigens (Spike, Membrane, Nucleocapsid, and the Receptor-Binding Domain). The system integrates a custom neural network-based image analysis pipeline for the automated, multiplexed detection of IgG, IgA, and IgM. Validation against hundreds of patient samples demonstrated high correlation with both human expert visual observations and conventional ELISA results. The mini-IFA proves particularly effective for IgG and IgA detection and can distinguish between vaccinated and naturally infected individuals. This approach allows for high-throughput serological testing in standard biosafety level environments and is rapidly adaptable to emerging pathogens.

Introduction

Serological testing is a cornerstone of pandemic management, providing critical data on population-level immunity, infection rates, and vaccine efficacy. However, existing diagnostic tools often force a compromise between data richness and scalability. The enzyme-linked immunosorbent assay (ELISA) is the current gold standard for high-throughput testing but typically measures responses against only one antigen at a time. Conversely, traditional Immunofluorescence Assays (IFA) provide detailed visual information about antibody binding but are labor-intensive, subjective, and often require high-containment Biosafety Level 3 (BSL-3) facilities because they rely on live virus cultures.

To address these limitations, we developed a "mini-IFA" platform. This system combines the visual depth of IFA with the automation of machine learning, all while operating under standard BSL-2 conditions by replacing live viruses with transfected cells expressing specific viral proteins.

Methodology

Antigen Expression and Assay Design

Instead of infecting host cells with live SARS-CoV-2, Vero E6 cells were transfected with plasmids encoding four key viral proteins: Spike (S), Nucleocapsid (N), Membrane (M), and the Receptor-Binding Domain (RBD). These cells were cultured in 384-well plates and exposed to patient serum. To enable multiplexing, fluorescently labeled secondary antibodies were used to simultaneously detect three distinct immunoglobulin classes: IgG, IgA, and IgM.

Machine Learning Pipeline

The assay utilizes a high-content imaging approach coupled with an automated analysis pipeline. We employed the BIAS software for deep learning-based cell segmentation. A custom neural network was then used to classify individual cells into five distinct categories: positive, negative, atypical, small bright, and trash. This automated scoring system processes thousands of cells per well, eliminating the inter-observer variability inherent in manual IFA interpretation.

Experimental Validation

Dataset and Training

The study utilized a comprehensive dataset of 948 serum or plasma specimens from 890 donors. This included 181 samples from RT-PCR confirmed SARS-CoV-2 patients and 500 pre-pandemic negative controls collected in 2017. To ensure the robustness of the ML model, experts manually labeled 55,496 individual cells across 16 different plates, providing a high-quality ground truth for the neural network.

Performance Metrics

The automated system demonstrated exceptional performance across several benchmarks:

  • Classification Accuracy: The ML model achieved a specificity of 0.95–0.97 and a sensitivity of 0.79–0.89 for individual cell classification.
  • Expert Correlation: When compared to a consensus of six human experts, the system achieved an Area Under the Curve (AUC) of 0.89 to 0.98.
  • ELISA Comparison: The mini-IFA showed a high Spearman correlation (0.81–0.82 for IgG) with traditional ELISA results, confirming its diagnostic reliability.

Discussion

Clinical and Laboratory Advantages

A primary advantage of the mini-IFA is its safety profile. By utilizing transfection-based antigen expression, the assay avoids the need for infectious materials, allowing diagnostic labs to perform complex serology without BSL-3 infrastructure. Furthermore, the multiplexed nature of the test provides a more nuanced immune profile than single-antigen ELISAs. For instance, the system can distinguish between vaccinated individuals (who show antibodies primarily against the Spike protein) and those previously infected with the virus (who exhibit antibodies against the Nucleocapsid and Membrane proteins).

Physiological Relevance

Unlike ELISA, which uses purified proteins immobilized on plastic surfaces, the mini-IFA expresses proteins within a cellular environment. This ensures that the antigens maintain more natural folding and localization, potentially leading to more accurate antibody recognition and fewer false results caused by denatured protein structures.

Limitations and Future Outlook

While highly effective for IgG and IgA, the detection of IgM was found to be less specific, likely due to the inherent non-specific staining patterns associated with that antibody class. Additionally, while the platform is adaptable, introducing a new pathogen requires an initial investment of expert time to label cells and train the model for new phenotypes.

Future applications of this platform include long-term kinetic studies to monitor how antibody levels against different viral components wane over time. The "plug-and-play" nature of the transfection-based system makes it a promising tool for rapid response to future emerging infectious diseases.

Conclusion

The mini-IFA represents a significant advancement in serological testing, combining the accuracy of automated machine learning with the biological richness of immunofluorescence. By providing a safe, high-throughput, and multiplexed diagnostic tool, this method offers a scalable solution for both clinical diagnostics and large-scale epidemiological research.

Summary updated

Abstract

We present a miniaturized immunofluorescence assay (mini-IFA) for measuring antibody response in patient blood samples. The method utilizes machine learning-guided image analysis and enables simultaneous measurement of immunoglobulin M (IgM), IgA, and IgG responses against different viral antigens in an automated and high-throughput manner. The assay relies on antigens expressed through transfection, enabling use at a low biosafety level and fast adaptation to emerging pathogens. Using severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) as the model pathogen, we demonstrate that this method allows differentiation between vaccine-induced and infection-induced antibody responses. Additionally, we established a dedicated web page for quantitative visualization of sample-specific results and their distribution, comparing them with controls and other samples. Our results provide a proof of concept for the approach, demonstrating fast and accurate measurement of antibody responses in a research setup with prospects for clinical diagnostics.
BibTeX
@inproceedings{pietiainen2023image,
  author = {Vilja Pietiäinen and Minttu Polso and Ede Migh and Christian Guckelsberger and Maria Harmati and Akos Diosdi and Laura Turunen and Antti Hassinen and Swapnil Potdar and Annika Koponen and Edina Gyukity Sebestyen and Ferenc Kovacs and Andras Kriston and Reka Hollandi and Katalin Burian and Gabriella Terhes and Adam Visnyovszki and Eszter Fodor and Zsombor Lacza and Anu Kantele and Pekka Kolehmainen and Laura Kakkola and Tomas Strandin and Lev Levanov and Olli Kallioniemi and Lajos Kemeny and Ilkka Julkunen and Olli Vapalahti and Krisztina Buzas and Lassi Paavolainen and Peter Horvath and Jussi Hepojoki},
  title = {{Image-based and machine learning-guided multiplexed serology test for SARS-CoV-2}},
  booktitle = {Cell Reports Methods},
  year = {2023},
  url = {https://openreview.net/forum?id=fQVMBT7YYB},
}

Loading related content...

Figure