Search

EP-4068152-B1 - RAPID, AUTOMATED IMAGE-BASED VIRUS PLAQUE AND POTENCY ASSAY

EP4068152B1EP 4068152 B1EP4068152 B1EP 4068152B1EP-4068152-B1

Inventors

  • OLSZOWY, Michael W.
  • REIF, OSCAR-WERNER
  • TRYGG, JOHAN
  • WALES, RICHARD
  • SJÖGREN, Rickard
  • EDLUND, CHRISTOFFER

Dates

Publication Date
20260506
Application Date
20210331

Claims (14)

  1. A method for training a machine learning model to predict a virus titer from a sequence of images of a cell culture containing a virus population, comprising the steps of: (1) obtaining a training set in a form of a plurality of images of virus-treated cell cultures from a plurality of experiments at a plurality of time points from a start time t 0 to a final time t final , wherein the training set images comprise fluorescence images of the cell culture labelled with a fluorescent marker, wherein the fluorescent marker is a fluorescent antibody binding to virus specific protein epitopes; (2) for each experiment, recording at least one numeric virus titer readout of the virus-treated cell culture at the final time t final ; (3) processing the images in the training set to acquire a numeric representation of each image in the sequence captured at a time before t final ; and (4) training one or more machine learning models to make a prediction of a final virus titer on the training set numeric representations, wherein the final virus titer ground truth is represented by the corresponding numeric virus titer readout.
  2. The method of claim 1, wherein the virus titer readout comprises (1) a number of infective particles or number of infective particles per unit volume, (2) a readout of a Tissue Culture Infective Dose 50 % Assay, or (3) the readout from a focus-forming assay.
  3. The method of any of claims 1-2, wherein the training set images comprise label-free light microscopy images.
  4. The method of any of claims 1-3, wherein processing step (3) comprises passing the microscopic images through a convolutional neural network (CNN) to thereby acquire an intermediate data representation of the microscopic images.
  5. The method of any of claims 1-4, wherein the processing step (3) comprises the steps of: a) segmenting individual cells from the microscopic images, b) calculating a cell-by-cell numeric description of each cell, and c) aggregating the numeric descriptions over all cells.
  6. The method of any of claims 1-5, wherein the machine learning model comprises one of a) a partial least squares linear model, b) an artificial neural network, c) a Gaussian process regression, and d) a neural ordinary differential equation model.
  7. The method of any of claims 1-6, wherein there are at least two time points in step 1) and wherein the period between the time points is less than or equal to 60 minutes.
  8. A method for predicting a virus titer of a cell culture to which a virus sample of unknown titer has been added, comprising the steps of: a) obtaining a time sequence of microscopic images of the cell culture; b) supplying a numeric representation of the time sequence of microscopic images obtained in step a) to one or more machine learning models trained in accordance with any one of claims 1-7, and c) making a prediction with the one or more trained machine learning models of the virus titer.
  9. The method of claim 8, wherein the cell culture further comprises specialist media aiding in imaging of the cell culture.
  10. An analytical instrument, comprising a system configured to hold one or more plates containing a cell culture and a virus sample, an integrated microscopic imaging system, and a machine learning model trained to make a prediction of a virus titer in the cell culture from one or more images in a time sequence of images of the cell culture obtained by the microscopic imaging system, wherein the prediction is made before the viral infection of the cell culture has proceeded to term, wherein the images comprise fluorescence images of the cell culture labelled with a fluorescent marker, wherein the fluorescent marker is a fluorescent antibody binding to virus specific protein epitopes.
  11. The analytical instrument of claim 10, wherein the instrument is further configured with a processing unit executing a training module, the training module providing set-up instructions for facilitating a user of the instrument conducting a training method with the instrument comprising the steps of: (1) obtaining a training set in the form of a a plurality of images of virus-treated cell cultures from a plurality of experiments at a set of time points from a start time t 0 to a final time t final , (2) for each experiment, recording at least one numeric virus titer readout of the virus-treated cell culture at time t final , (3) processing all microscopic images in the training set to acquire a numeric representation of each image; and (4) training one or more machine learning models to make a prediction of a final virus titer on the training set numeric representations, wherein the training comprises minimizing the error between the model prediction of a final virus titer and a ground truth.
  12. The analytical instrument of claim 11, wherein the processing step (3) comprises the steps of: a) segmenting individual cells from the microscopic image, b) calculating a cell-by-cell numeric description of each cell, and c) aggregating the numeric descriptions over all cells.
  13. The analytical instrument of claim 11, wherein the machine learning model comprises one of a) a partial least squares linear model, b) an artificial neural network, c) a Gaussian process regression, and d) a neural ordinary differential equation model.
  14. The analytical instrument of claim 11, wherein the processing step (3) further comprises a step of either (1) filtering out cells not infected by the virus, (2) filtering out dead cells, or (3) filtering out dead cells that did not die from a virus infection.

Description

Background This disclosure is directed to methods and systems for performing cell-based, functional virus counting assays, and more particularly to methods and systems for allowing such assays to be completed in much less time than usual. Functionality of viruses, such as active viruses, is currently measured in a variety of ways. The most widely used method is the standard plaque assay, which was first described in 1953. The assay measures virus function via the infection and lysis of target cells. The assay yields a plaque titer (concentration) indicative of the number of functional viruses, or plaque forming units, within a sample. The basic method is shown in Figure 1. First, cells are initially plated and grown to confluence. Then, virus samples of unknown concentration (titer) are serially diluted and added to the plates (often in the form of a petri-type dish or plate) containing the cells. Next, two to fourteen days later the cell monolayers are stained to reveal areas of lysis (plaques). Finally, the number of plaques is adjusted for dilution to determine virus titer of the original sample. Other functional assays, such as the fifty-percent tissue culture infective dose (TCID50), are derivative of the plaque assay. All involve incubation periods, virus with cells, of 2 to 12+ plus days depending upon the virus and cells used to measure functional infectivity. There is wide disparity in the ratios of total virus particles, functional and non-functional, to plaque-forming units (PFUs) for different viruses, as indicated by Table 1. Infectivity titers or infectious particle counts and total particle counts are essential for full virus characterization. The particle to PFU ratio can vary across orders of magnitude. Table 1VirusParticle-to-PFU ratioAdenoviridae20-100Alphaviridae1-2Semliki Forest virusHerpesviridae50-200Herpes simplex virusOrthomyxoviridae20-50Influenza virusPapillomaviridae10,000PapillomavirusPicomaviridae30-1,000PoliovirusPolyomaviridaePolyomavirus38-50Simian virus 40100-200Poxviridae1-100Reoviridae10Reovirus Source: http://www.virology.ws/2011/01/21/are-all-virus-particles-infectious/. Therefore, it has become critically important to understand both the functional virus titer, via a plaque assay, as well as the total virus particle number in virus samples, both for commercial applications, such as vaccine development and manufacture, as well as for safety in gene therapy. To address the need for total particle counts, more modern technologies, such as embodied in Sartorius' Virus Counter® product, electron microscopy and a number of indirect methods, have been developed to provide total particle counts. For some of these technologies, these counts can be measured in as little as 30 minutes. Unfortunately, heretofore there are no rapid surrogates for the legacy virus plaque assay of Figure 1. It is highly desirable to have both total particle and infectious particle counts contemporaneously, or nearly so. This disclosure presents a rapid, automated image-based virus plaque and potency assay that provides a plaque assay titer, and in turn, an infectious particle count, in hours, rather than in days. Thus, this disclosure now makes it possible to obtain total particle and infectious particle counts essentially contemporaneously. "Deep learning of virus infections reveals mechanics of lytic cells", Andriasyan et al., 2019, describes a deep learning approach to identify herpesvirus and adenovirus infections in the absence of virus-specific stainings. Procedures comprises staining of infected nuclei with DNA-dyes, fluorescence microscopy, and validation by virus-specific live-cell imaging. Deep learning of multi-round infection phenotypes identified hallmarks of adenovirus-infected cell nuclei. At an accuracy of >95%, the procedure predicts two distinct infection outcomes 20 hours prior to lysis, nonlytic (nonspreading) and lytic (spreading) infections. Summary In one aspect, described herein is a method for training a machine learning model to predict virus titer from an image, or sequence of images, of a cell culture containing a virus population. In this document, the term "machine learning model" refers to a computational system that has used optimization algorithms to learn and perform a task based on previous examples of desired input-output pairs. The trained machine learning model allows a prediction of virus titer to be made much earlier than in the standard virus plaque assay, for example in 6 or 8 hours (or possibly less) after initial inoculation of the cell culture with the virus sample, as compared to many days in the prior art. The method of training the machine learning model can include the steps of: (1) obtaining a training set in a form of a plurality of images of virus-treated cell cultures from a plurality of experiments at one or more time points from a start time t0 to a final time tfinal, (2) for each experiment, recording at least one numeric virus titer readout of