Source: UNIVERSITY OF FLORIDA submitted to NRP
MACHINE LEARNING ENABLED DETECTION OF SPOILAGE AND FOODBORNE PATHOGENS USING PAPER CHROMOGENIC ARRAYS OF DYE-IMPREGNATED POROUS NANOSILICA
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
ACTIVE
Funding Source
AFRI COMPETITIVE GRANT
Reporting Frequency
Annual
Accession No.
1030547
Grant No.
2021-67021-39520
Cumulative Award Amt.
$420,604.39
Proposal No.
2022-11982
Multistate No.
(N/A)
Project Start Date
Sep 1, 2023
Project End Date
Aug 31, 2026
Grant Year
2023
Program Code
[A1511]- Agriculture Systems and Technology: Nanotechnology for Agricultural and Food Systems
Recipient Organization
UNIVERSITY OF FLORIDA
G022 MCCARTY HALL
GAINESVILLE,FL 32611
Performing Department
(N/A)
Non Technical Summary
Fast and reliable pathogen detection in food is critical to public health, and in particular, in preventing foodborne illness outbreaks. Here, we propose a novel system to detect viable spoilage and pathogenic microorganisms in complex food matrices using a paper chromogenic array (PCA) enabled by machine learning. The PCA includes a paper substrate impregnated with 22 chromogenic dyes, in which exposure to volatile organic compounds released by microorganisms of interest elicits color changes. These color changes are digitized and used to train a machine learning (ML) algorithm, including a state-of-the-art multi-layer convolutional neural network, giving it strain-specific, high-accuracy pathogen detection and quantification capabilities. The outputs of automated pattern recognition include microbial identities, strain-specific microbial population estimates, and spoilage status of the food substrate. The proposed work includes the development of the approach for the seven most common pathogens and spoilage-causing microbes for four model foods and an approach for database construction and ML training that can be easily extended to other microbial targets and food commodities. The speed, reliability, low cost, and versatility of this approach give it the potential to significantly advance nondestructive in-the-field detection of microbial contamination in food.
Animal Health Component
30%
Research Effort Categories
Basic
30%
Applied
30%
Developmental
40%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
7127299110050%
5032410208050%
Knowledge Area
712 - Protect Food from Contamination by Pathogenic Microorganisms, Parasites, and Naturally Occurring Toxins; 503 - Quality Maintenance in Storing and Marketing Food Products;

Subject Of Investigation
2410 - Cross-commodity research--multiple crops; 7299 - Research equipment and methods, general/other;

Field Of Science
1100 - Bacteriology; 2080 - Mathematics and computer sciences;
Goals / Objectives
We address the following priorities in USDA-NIFA-AFRI-006609 and the program of Nanotechnology for Agricultural and Food Systems (A1511): 1) Innovative ideas and fundamental sciences to develop nanotechnology-enabled solutions for food security through improved productivity, quality, and reducing food waste/loss; Enhanced food safety and biosecurity. 2) Nanotechnology-enabled smart sensors for accurate, reliable, and cost-effective early and rapid detection of pathogens and contaminants in foods. Portable, field-deployable, and agriculturally affordable sensors and devices for real-time detection and screening to identify targets requiring no additional laboratory analyses; The tools of big data. Fast and reliable pathogen detection in food is critical to public health, and in particular, in preventing foodborne illness outbreaks. Here, we propose a novel system to detect viable spoilage and pathogenic microorganisms in complex food matrices using a paper chromogenic array (PCA) enabled by machine learning (ML). There are four main goals: 1) Streamline dye selection and dye optimization processes using general-purpose mixed-integer nonlinear optimization; Standardize PCA assembly using photolithography and paper microfluidic fabrication techniques; 2) Construct a PCA database and training ML algorithm for multiplex identification of viable microbial targets; 3) Assess specificity and sensitivity of the PCA-ML platform and report training and testing accuracy; 4) Validate the PCA-ML platform as a nondestructive surveillance tool on real food models.
Project Methods
1) For the first objective, we will select 22 dye spots (each with one or more dyes and a single deposition thickness) using a library of 71 chromogenic dyes responsive to metabolic volatile organic compounds (VOCs) from viable pathogens. Optimization using the general-purpose mixed-integer nonlinear programming code BARON will be used to select the 22 dye spots based on response data for individual VOCs emitted by microorganisms on the target model food. A similar approach will optimize dye thickness and reduce the number of parameter combinations at which the PCA must be tested in order to construct a good database. We will also use state-of-the-art photolithography and paper microfluidics fabrication techniques to standardize PCA assembly. 2) We will develop ML pattern-recognition algorithms to streamline knowledge database construction, image preprocessing, and subsequent algorithm training. We will deploy and train traditional AI models and deep learning models to prioritize specificity and identification in multiplex cocktails. Finally, we will develop novel algorithms combining transfer learning and best-subset selection to provide interpretable ML models easily confirmed and trusted by users. For this approach, we will develop novel polynomial-time rounding and randomization algorithms and theoretical quality guarantees for their learning rates. 3) Limits of detection (LODs) and quantitative pattern-data relationships are determined by VOC transport to the paper microarray and subsequent kinetics of reaction with the chromogens. To model dye response, we will compute VOCs transport and kinetics in the headspace and in and on the paper matrices, validated against the experiment. Computations will employ an open-source spectral-element computational fluid dynamics code with good capabilities to model laminar and turbulent flows and transport and reaction. We will also investigate whether dye deposition onto the paper microarray using a nano- or mesoporous powder provides significantly better sensitivity compared to direct deposition.4) The PCA, ML algorithms, and sampling hardware described in Objectives 1-3 will be validated using four model foods (fresh-cut Romaine lettuce, pear, ground turkey, and frozen raw tuna), inoculated with single-cultures of the top-seven pathogens or multiplexed cocktails thereof. This will constitute an evaluation of functionality in monitoring both food safety and spoilage for real samples.

Progress 09/01/23 to 08/31/24

Outputs
Target Audience:The proposed project and the importance of the work will be communicated to researchers and experts in the field at conferences and meetings, as well as invited talks/seminars. The audiences include students, faculties, and researchers in the field of food science and microbiology. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?One master's student at the University of Florida, Gainesville, FL, was trained under this project. How have the results been disseminated to communities of interest?Boce Zhang. Seminar. Challenges & opportunities of nanopore & smart sensing in the "new era of smarter food safety". University of Georgia - Center for Food Safety. 2024. Griffin, GA. Boce Zhang. Lecture. Challenges & innovations in "the new era of smarter food safety"-prospects of omics & AI. University of Maryland, Department of Nutrition and Food Science. 2024. Virtual. Boce Zhang. Speech. Machine learning enabled nondestructive paper chromogenic array detection of multiplexed viable pathogens on food. Institute of Biological Engineering (IBE) 2024 - Biological Sensing and Diagnostics. 2024. Atlanta, GA. Jia, Z., Zhang, B. Surveillance of pathogenic bacteria on a food matrix using machine-learning-enabled paper chromogenic arrays. IAFP. 2024. Long Beach, CA. Jia, Z., Zhang, B. Surveillance of pathogenic bacteria on a food matrix using machine-learning-enabled paper chromogenic arrays. SCIX. 2024. Raleigh, NC. What do you plan to do during the next reporting period to accomplish the goals?Manuscript under preparation 1. Jia, Z., Wang, D., Green, M., Roche, M., Block, E. M., Yu, H., Luo, Y., Zhang, B.* Real-time surveillance of pathogens in ground chicken under fluctuating temperatures using artificial intelligence powered paper chromogenic array senor. Under preparation 2. Jia, Z., Luo, Y., Russo, H. B., Tang, E. C., Holliday, E. G., Rootes, T. R., Yu, H., Zhang, B.* Nondestructive surveillance of viable but nonculturable Salmonella in low-moisture foods using nanoparticle-based paper chromogenic arrays and machine learning. Under preparation

Impacts
What was accomplished under these goals? During this period, we developed and applied a PCA-ML system to continuously monitor multiplex viable pathogens in poultry during storage under different dynamic temperature conditions. The PCA-ML system successfully identified Salmonella, E. coli, and Listeria monocytogenes in both monoculture and cocktail culture, and discriminated these pathogens from background microflora in the control samples. Notably, the PCA-ML system could also predict the population of each pathogen while detecting the pathogen species. Additionally, we developed and validated the capability of a nanoparticle-enabled PCA-ML system to continuously detect VBNC pathogens in low-moisture foods, using peanut butter as our model. The nanoparticle-enabled PCA comprised a paper microarray impregnated with nine nanoparticle-based chromogenic dyes. Our study tested the sensitivity of gold and silicon nanoparticles to react with standard VOCs, finding that silicon nanoparticles exhibited a more sensitive response to VOCs, which were then used for nanoparticle-enabled PCA fabrication. After exposure to different sample scenarios (control, VBNC, and normal Salmonella), the nanoparticle-enabled PCA displayed unique color patterns. This result demonstrates that the nanoparticle-enabled PCA could recognize VBNC Salmonella and differentiate the pathogen from normal Salmonella and background microflora. The color changes of the nanoparticle-enabled PCA were digitalized, and a database was constructed for ML analysis. We also planned for and purchased materials for a project utilizing a collection of non-toxic dyes. We conducted a literature search for non-toxic chromogenic dyes and screened previously selected dyes, creating a pool of seventeen potential chromogenic dyes for use in PCA design. The development of the PCA-ML platform using these selected non-toxic dyes is expected to be a focus of research in the coming term. We wrote a review paper that will serve as a chapter in the 113th volume of Advances in Food and Nutrition Research. In this review, we discussed current research and remaining obstacles for machine learning-enabled colorimetric sensors for the detection of foodborne pathogens. This review included references to previous research efforts from our team towards machine learning-enabled foodborne pathogen sensing and discussed similar research to increase visibility and interest in this type of research.

Publications

  • Type: Book Chapters Status: Accepted Year Published: 2024 Citation: Emma Holliday, Boce Zhang. 2024. Machine learning-enabled colorimetric sensors for food safety. Advances in Food and Nutrition Research - Smarter Food Safety. 113. Academic Press/Elsevier Cambridge, MA.