Probabilistic determination of food pathogens in qpcr
Patent Information
- Application Number
- PCT/EP2026/054134
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-02-19
- Filing Date
- 2026-02-16
- Publication Date
- 2026-08-27
Smart Images
Figure EP2026054134_27082026_PF_FP_ABST
Abstract
Description
[0001] P25-058 DAK
[0002] - 1 -
[0003] PROBABILISTIC DETERMINATION OF FOOD PATHOGENS IN qPCR
[0004] The hereby described invention discloses an improved method and a system for performing a real-time polymerase chain reaction (qPCR) using 5 Al based probabilistic determination.
[0005] Technical Field
[0006] The field of this invention pertains to the detection of food pathogens using real-time polymerase chain reaction (qPCR) techniques.
[0007] Background and description of the prior art
[0008] qPCR is a widely employed method for identifying the presence of specific 15 DNA sequences, such as those indicating food pathogens, within a given sample. This technique involves subjecting a sample to multiple cycles of temperature changes, which facilitate the amplification of DNA. At the conclusion of each cycle, fluorescence data is collected to determine the presence or absence of the target DNA sequence.
[0009] 20
[0010] Traditionally, the analysis of qPCR results occurs after the completion of all cycles, where fluorescence data is processed to yield a positive, negative, or invalid result. A positive result indicates the detection of the specific pathogen's DNA, while a negative result suggests its absence. An invalid result, often termed "no-amp," indicates that the test did not produce a conclusive outcome. The determination of a positive result typically depends on whether the fluorescence value crosses a predetermined threshold.
[0011] 30 The conventional approach, while effective, presents limitations in terms of time efficiency. The complete running stage of a qPCR test can be timeconsuming, often requiring up to an hour or longer to finalize results. DuringP25-058 DAK
[0012] -2 -
[0013] this period, food pathogen lab technicians must wait for the running stage to conclude before obtaining results, thereby hindering rapid decision-making and reducing sample throughput.
[0014] 5 In recent years, advancements in data analysis and machine learning have opened new avenues for enhancing qPCR techniques. Machine learning algorithms, particularly those utilizing historical fluorescence data, have shown potential in predicting qPCR outcomes during the running stage itself, thereby offering prospects for more efficient and timely analysis. This technological evolution aims to provide earlier indications of test results, thereby improving workflow efficiency and potentially enhancing the accuracy of pathogen detection.
[0015] Summary of the invention
[0016] 15
[0017] The task of this patent application is therefore to provide an improved approach to perform a pPCR reaction which is faster and therefore cheaper than the known approaches.
[0018] 20 This task has been solved by a method for performing a real-time polymerase chain reaction (qPCR), in particular a test, wherein a sample is exposed to multiple iterations of temperature ranges over a fixed time interval and after each iteration fluorescence data is collected for each sample and analyzed by a computer resulting in a positive, negative or invalid result characterized in that the computer uses an software-based machine-learning algorithm, trained on previously available fluorescence data, which is applied on the collected fluorescence data instantly after collection to predict the result of the qPCR test after each iteration. The innovative aspect of this method lies in the use of a software-based
[0019] 30 machine-learning algorithm, which is trained on previously available fluorescence data. This algorithm predicts the qPCR test result after each iteration in real-time. The technical effect of this feature is the provision ofP25-058 DAK
[0020] - 3 -
[0021] immediate feedback on the test outcome, significantly reducing the waiting time for results compared to traditional methods. The advantage is a more efficient workflow, allowing lab technicians to make quicker decisions and increase sample throughput.
[0022] 5
[0023] Advantageous and therefore preferred further developments of this invention emerge from the associated subclaims and from the description and the associated drawings.
[0024] One of those preferred further developments of the disclosed [method] comprise that the machine-learning algorithm is realized either in form of a neural network, in particular a network that is optimized for classification problems] such as a Feedforward neural network (ANN), or a Recourrent neural network(RNN), or a traditional algorithm such as Logistic regression, 15 Binary classification, or K-means clustering to implement a chemometric model. A neural network such as an ANN, as an advanced form of machine learning, can model complex patterns in fluorescence data, enhancing the prediction accuracy. The chemometric model uses statistical techniques to interpret chemical data, providing a robust framework for analyzing qPCR 20 results.
[0025] Another one of those preferred further developments of the disclosed method comprise that the chemometric model uses a logistic regression approach to calculate the prediction output. Logistic regression is a statistical method that models the probability of a binary outcome, such as positive or negative test results. This approach allows for precise prediction outputs, improving the reliability of the qPCR test results.
[0026] Another one of those preferred further developments of the disclosed 30 method comprise that the qPCR test is used to determine the existence of food pathogens in food matrixes. This application is critical in the context of food safety and quality control, as it directly addresses the need for rapidP25-058 DAK
[0027] -4 -
[0028] and accurate detection of harmful pathogens in food products. By utilizing the described qPCR method with a machine-learning algorithm, the process can efficiently analyze samples and predict the presence of pathogens such as bacteria or viruses that may pose health risks. It enhances the food 5 safety protocols, allowing for timely intervention and reducing the risk of contaminated food reaching consumers. This application leverages the strengths of the described method, particularly its real-time predictive capabilities, to provide a significant advantage in the field of food pathogen detection.
[0029] Another one of those preferred further developments of the disclosed method comprise that the prediction output of the machine-learning algorithm is a categorical value that determines if the sample will be marked as positive, negative or no-amp. This categorization aligns with traditional 15 qPCR result interpretations, facilitating seamless integration into existing workflows. No amp means that there is no amplification by a control that is placed within a sample and therefore the test has failed to run properly and cannot provide any reliable output. While in conventional qPCR analysis the categorical value has be evaluated against a predetermined threshold value 20 to determine if the sample will be marked as positive, negative or no-amp, in this new approach due to the probabilistical determination approach the trained machine-learning algorithm provides the result, positive, negative or no-amp, directly within its prediction.
[0030] Another one of those preferred further developments of the disclosed method comprise that for training of the machine-learning algorithm historical fluorescence data from previous real-time polymerase chain reactions (qPCR) is used as previously available fluorescence data to enhance prediction accuracy. By leveraging historical data, the algorithm 30 can enhance its prediction accuracy, adapting to various sample types and conditions encountered in food pathogen detection. The historical fluorescence data can be taken from previous qPCR runs in the sameP25-058 DAK
[0031] - 5 -
[0032] location, e.g. laboratory, and even in a very preferred embodiment from previous iterations of the same qPCR run. Furthermore it is also possible to use external historical fluorescence data provided from other locations and sources.
[0033] 5
[0034] Another one of those preferred further developments of the disclosed method comprise that the machine-learning algorithm is continuously updated by incorporating feedback from correct and / or incorrect predictions to improve future predictions. This adaptive learning process allows the algorithm to improve its accuracy over time, ensuring better performance in future qPCR tests / runs. Continuously most preferably means after every qPCR run, but alternatively it could be a specific number of qPCR test runs as well, like five or ten or any other suitabe number.
[0035] 15 Another one of those preferred further developments of the disclosed method comprise that a real-time feedback is provided by the computer which is dependent on a confidence level of the prediction, which is calculated by the software-based machine-learning algorithm parallel to the prediction result. This confidence level offers users an indication of the 20 reliability of the test outcome. The advantage is added assurance for lab technicians about the reliability of the predicted results. If the confidence level is low, further qPCR iterations for the current qPCR run are recommended.
[0036] Furthermore the invention includes a system for performing a real-time polymerase chain reaction (qPCR) test, comprising a qPCR thermal cycler configured to expose a sample to multiple iterations of temperature ranges over a fixed time interval, a fluorescence detector configured to collect fluorescence data from the sample after each iteration and a computer 30 comprising a processor and at least one memory storing a software-based machine-learning algorithm trained on previously available data;P25-058 DAK
[0037] - 6 -
[0038] wherein the system is configured to perform the previously disclosed method. This system provides a comprehensive solution for real-time qPCR analysis.
[0039] 5 Another one of those preferred further developments of the disclosed system comprise that the software-based machine-learning algorithm is implemented on the at least one memory via a cloud-based platform, allowing for remote access and analysis of the collected fluorescence data. This feature offers flexibility and scalability, enabling centralized data processing and analysis. It further allows for a faster result, since the cloudbased platform might have much more performant hardware resources like CPU-, GPU-power and the like. In that case a fast network connection to this cloud-based platform is recommended, which could be a standard wireless and / or cable based connection, depending on the available WAN- 15 ZLAN system at the used location.
[0040] Another one of those preferred further developments of the disclosed system comprise that the fluorescence sensor is an optical sensor, in particular a digital camera, which is incorporated in the system. The optical sensor is 20 preferably a digital camera with standard technology like CCD-, APS-CMOS technology, which is suitable for the capturing of the respective fluorescence light. Technically every suitable kind of image sensor could be used as long as its diameter allows an incorporation in the system and provides the required image quality.
[0041] A further solution includes a computer program product and a computer- readable storage medium and / or data carrier signal having stored thereon the computer program product, which comprises instructions which cause the involved computers to perform the method steps of the previously 30
[0042] described methods. This ensures that the software is tailored to the specific needs of the qPCR system, optimizing its performance and emphasizes theP25-058 DAK
[0043] - 7 -
[0044] importance of reliable data storage and retrieval in the qPCR testing process.
[0045] Detailed description of the invention
[0046] 5
[0047] The method, system and software product according to the invention and functionally advantageous developments of those are described in more detail below with reference to the associated drawings using at least one preferred exemplary embodiment. In the drawings, elements that correspond to one another are provided with the same reference numerals.
[0048] The drawings show:
[0049] Figure 1 : a fluorescence profile of a positive sample
[0050] Figure 2: an schematic overview about the required system parts to perform the qPCR run
[0051] Figure 3: an activity diagram visually representing the flow of data and decisions throughout the qPCR process
[0052] Figure 4: a diagram showing the prediction accuracy of an Al model using logistic regression
[0053] Figure 2 shows the respective relevant system 6 parts of the laboratory equipment for the preferred embodiment. The system 6 structure can also differ from embodiment to embodiment. Especially the kind of the involved computer 7 can differ greatly, depending on how much of the steps is performed by human users with the help of computers and application software or done automatically by specific computers using for instance Al enhanced software. For this preferred embodiment it is enough to say, that at least one computer 3, preferably a standard personal computer or 30 notebook or server / -s hosting the mentioned cloud memory, performs the qPCR which steps are further shown in a preferred working example in Figure 3.P25-058 DAK
[0054] - 8 -
[0055] The system 6 is situated within a qPCR laboratory environment. The lab technician 1 interacts with the system 6 to conduct the qPCR tests.
[0056] 5 An optical sensor 2 such as a camera embedded within the system 6, which is a qPCR machine, is responsible for detecting and collecting fluorescence data 3 during each cycle of the qPCR process. The fluorescence data 3 serves as the crucial input for the subsequent analysis and prediction tasks.
[0057] The collected fluorescence data is transmitted to the computer 7. This computer is equipped with the necessary processing capabilities to handle the incoming data 3. It utilizes software 5, which includes a machine learning (ML) algorithm 4, in preferably a chemometric model. The ML 15 algorithm 4 processes the fluorescence data 3 to generate predictions regarding the sample's 9 status, categorizing it as positive, negative, or noamp.
[0058] The computer 7 provides a user interface 8, which provides the lab
[0059] 20 technician 1 with real-time feedback and control over the testing process.
[0060] This user interface 8 allows the technician 1 to monitor the progress of the qPCR test, view prediction results, and make informed decisions based on the data 3 presented.
[0061] As mentioned earlier the traditional approach involves preparing the sample 9, running the sample through a qPCR instrument 6, then analyzing the results. The outcome of the analysis determines a “Positive” or a “Negative” result for each sample 9. A positive result in gernal indicates a food pathogen is present, while a negative sample indicates no detectable levels 30 of a specific food pathogen.P25-058 DAK
[0062] - 9 -
[0063] During the qPCR “running” stage, the sample is exposed to multiple iterations of temperature ranges over a fixed time interval. Each iteration of the interval is called a cycle. At the end of each cycle, fluorescence data 3 is collected for each sample.
[0064] 5
[0065] Upon completion of the running stage, the fluorescence data from all the cycles is processed to determine a result for each sample. This is the analysis stage. The result can be considered Positive, Negative, or “no amp”. Positive indicates that a specific pathogen’s DNA has been detected in the sample, Negative indicates no detection of a pathogen’s DNA, and “no-amp” indicates an invalid test. Positive indication is determined based on a sample’s 9 fluorescence value 3 crossing a predetermined threshold value. Figure 1 shows an example of a fluorescence profile of positive sample.
[0066] 15
[0067] In the working example of the current invention as shown in Figure 3, this method is now improved to overcome the previously mentioned shortcomings. The invented method begins similar, with the initiation of the qPCR running stage, during which the sample 9 is subjected to multiple 20 cycles of temperature variations via a qPCR thermal cycler 10. Each cycle is designed to facilitate the amplification of DNA sequences, a fundamental aspect of the qPCR process. During these cycles, fluorescence data 3 is systematically collected for each sample 9. This data 3 serves as the primary input for the subsequent analysis, providing a detailed account of the sample's behavior throughout the running stage. The collection of fluorescence data 3 is a continuous process, ensuring that each cycle's outcomes are captured accurately and comprehensively. Figure 3 illustrates an activity diagram of the working example, visually representing the flow of data and decisions throughout the qPCR process. This diagram provides a 30 clear overview of each step involved, from data collection to prediction and feedback, highlighting the dynamic nature of the invention.P25-058 DAK
[0068] - 10 -
[0069] Once the fluorescence data 3 is collected, it is now immediately provided to the chemometric model which is specifically trained for this purpose. The chemometric model, which is realized in a most preferred embodiment in the form of a neural network, processes the data 3 to generate a prediction 5 regarding the sample's 9 status. The output of this model is a categorical value, indicating whether the sample should be marked as positive, negative, or no-amp. This categorization aligns with traditional qPCR result interpretations, offering a seamless transition from manual to automated analysis. The chemometric model's prediction is probabilistic, meaning it estimates the likelihood of each potential outcome based on the data 3 provided. As the running stage progresses, the model continuously refines its predictions, aiming to converge with the results that would be obtained through traditional analysis methods.
[0070] 15 The invention is designed to enhance prediction accuracy over time. The chemometric model therefore adapts to various environmental conditions, sample matrices, and slight variances in sample preparation. This adaptability ensures that the model remains effective across different testing scenarios, accommodating the inherent variability of biological 20 samples. In some embodiments, any incorrect predictions made by the chemometric model are used as feedback to improve the algorithm's performance in future runs. This feedback mechanism allows the model to learn from its mistakes, gradually enhancing its predictive capabilities and reducing the likelihood of errors in subsequent tests.
[0071] Figure 4 demonstrates an example of the prediction accuracy of the used chemometric model using logistic regression. This figure illustrates how the model's accuracy improves as the number of cycles increases, showcasing the model's ability to learn and adapt over time. The accuracy is calculated 30 based on the model's predictions for a test set of 10,000 samples, reflecting its capacity to distinguish between positive and negative outcomes effectively. The graph in Figure 4 shows a progressive increase inP25-058 DAK
[0072] - 11 -
[0073] accuracy, with the model's performance converging towards the expected results with each additional cycle. This convergence underscores the model's robustness and reliability, particularly in the context of food pathogen detection, where accurate and timely results are paramount. 5
[0074] Overall, this working example demonstrates the practical application of the invention, offering a detailed account of how the Al-enhanced qPCR process operates. By incorporating machine learning into the qPCR workflow, the invention provides a powerful tool for rapid and accurate pathogen detection, addressing key challenges in the field of food safety and quality control.
[0075] 30P25-058 DAK
[0076] - 12 -
[0077] List of references
[0078] 1 Lab technician I user
[0079] 2 Sensor
[0080] 5 3 Fluorescence data
[0081] 4 ML algorithm I chemometric model 5 Software
[0082] 6 qPCR instrument I system
[0083] 7 Computer
[0084] 8 User interface
[0085] 9 Sample
[0086] 10 qPCR Thermal Cycler
[0087] 15
[0088] 20
[0089] 30
Claims
P25-058 DAK- 13 -Patent claims1. Method for performing a real-time polymerase chain reaction (qPCR) wherein a sample (9) is exposed to multiple iterations of temperature 5 ranges over a fixed time interval and after each iteration fluorescence data (3) is collected for each sample (9) and analyzed by a computer (7) resulting in a positive, negative or invalid result;characterized in thatthe computer (7) uses an software-based machine-learning algorithm (4), trained on previously available fluorescence data, which is applied on the collected fluorescence data (3) instantly after collection to predict the result of the qPCR test after each iteration.
2. Method according to claim 1 ,15 characterized in thatnetwork that is optimized for classification problems] such as a Feedforward neural network (ANN), or a Recourrent neural network(RNN), or a traditional algorithm such as Logistic regression, Binary classification, or K-means clustering to implement a 20 chemometric model.
3. Method according to claim 2,characterized in thatthe chemometric model uses a logistic regression approach to calculate the prediction output.
4. Method according to any of the previous claims,characterized in thatthe qPCR test is used to determine the existence of food pathogens in 30 food matrixes.
5. Method according to any of the previous claims,P25-058 DAK- 14 -characterized in thatthe prediction output of the machine-learning algorithm (4) is a categorical value that determines if the sample (9) will be marked as positive, negative or no-amp.
56. Method according to any of the previous claims,characterized in thatfor training of the machine-learning algorithm (4) historical fluorescence data from previous real-time polymerase chain reactions (qPCR) is used as previously available fluorescence data to enhance prediction accuracy.
7. Method according to claim 6,characterized in that15 the machine-learning algorithm (4) is continuously updated by incorporating feedback from correct and / or incorrect predictions to improve future predictions.
8. Method according to any of the previous claims,20 characterized in thata real-time feedback is provided by the computer (7) which is dependent on a confidence level of the prediction, which is calculated by the software-based machine-learning algorithm (4) parallel to the prediction result.
9. System for performing a real-time polymerase chain reaction (qPCR) test, comprising a qPCR thermal cycler (10) configured to expose a sample (9) to multiple iterations of temperature ranges over a fixed time interval, a fluorescence sensor (2) configured to collect fluorescence data (3) from 30 the sample (9) after each iteration and a computer (7) comprising a processor and at least one memory storing a software-based machinelearning algorithm (4) trained on previously available data;P25-058 DAK- 15 -wherein the system is configured to perform the method according to claims 1 to 8.
10. System according to claim 9,5 characterized in thatthe software-based machine-learning algorithm (4) is implemented on the at least one memory via a cloud-based platform, allowing for remote access and analysis of the collected fluorescence data (3).
11. System according to claim 9 or claim 10,characterized in thatthe fluorescence sensor (2) is an optical sensor (2), in particular a digital camera, which is incorporated in the system (6).1512. Computer program product comprising instructions which cause the computer of the system according to claims 9 to 11 to perform the method steps of claims 1 to 8.20 13. Computer-readable storage medium and / or data carrier signal having stored thereon the computer program product of claim 12 which cause the involved computer of the system according to claims 9 to 11 to carry out the method steps of claims 1 to 8.