Evaluation of the Impact of Skewness, Clustering, and Probe Sampling Plan on Aflatoxin Detection in Corn

Machine learning assisted biosensing technology: An emerging powerful tool for improving the intelligence of food safety detection

Recently, the application of biosensors in food safety assessment has gained considerable research attention. Nevertheless, the evaluation of biosensors' sensitivity, accuracy, and efficiency is still ongoing. The advent of machine learning has enhanced the application of biosensors in food security assessment, yielding improved results. Machine learning has been preliminarily applied in combination with different biosensors in food safety assessment, with positive results. This review offers a comprehensive summary of the diverse machine learning methods employed in biosensors for food safety. Initially, the primary machine learning methods were outlined, and the integrated application of biosensors and machine learning in food safety was thoroughly examined. Lastly, the challenges and limitations of machine learning and biosensors in the realm of food safety were underscored, and potential solutions were explored. The review's findings demonstrated that algorithms grounded in machine learning can aid in the early detection of food safety issues. Furthermore, preliminary research suggests that biosensors could be optimized through machine learning for real-time, multifaceted analyses of food safety variables and their interactions. The potential of machine learning and biosensors in real-time monitoring of food quality has been discussed.

Modeling Preharvest Cyclospora cayetanensis Sampling and Testing for Various Water and Produce Sampling Plans

As of August 2023, the two U.S. Food and Drug Administration (FDA) official detection methods for C. cayetanensis are outlined in the FDA Bacteriological Analytical Manual (BAM) Chapters 19b (produce testing) and 19c (agricultural water testing). These newly developed detection methods have been shown to not always detect contamination when present at low levels. Yet, industry and regulators may choose to use these methods as part of their monitoring and verification activities while detection methods continue to be improved. This study uses simulation to better understand the performance of these methods for various produce and water sampling plans. To do so, we used published FDA test validation data to fit a logistic regression model that predicts the methods' detection rate given the number of oocysts present in a 10-L agricultural water or 25 g produce sample. By doing so, we were able to determine contamination thresholds at which different numbers of samples (n = 1, 2, 4, 8, 16, and 32) would be adequate for detecting contamination. Furthermore, to evaluate sampling plans in use cases, a simulation was developed to represent C. cayetanensis contamination in agricultural water and on cilantro throughout a 45-day growth cycle. The model included uncertainty around the contamination sources, including scenarios of unintentionally contaminated irrigation water or in-field contamination. The results demonstrate that in cases where irrigation water was the contamination source, frequent water testing proved to be more powerful than produce testing. In scenarios where contamination occurred in-field, conducting frequent produce testing or testing produce toward the end of the season more reliably detected contamination. This study models the power of C. cayetanensis detection methods to understand the sampling plan performance and how these methods can be better used to monitor this emerging food safety hazard.

Aggregative Soil Sampling Using Boot Covers Compared to Soil Grabs From Commercial Romaine Fields Shows Similar Indicator Organism and Microbial Community Recoveries

Aggregative boot cover sampling may be a more representative, practical, and powerful method for preharvest produce soil testing than grab sampling because boot covers aggregate soil from larger areas. Our study tests if boot cover sampling results reflect quality and safety indicator organisms and community diversity of grab sampling. We collected soil samples from commercial romaine lettuce fields spanning 5060 m2 using boot covers (n = 28, m = 1.1 ± 0.4 g; wearing boot covers and walking along the path), composite grabs (n = 28, m = 231 ± 24 g; consisting of 60 grabs of 3-5 g each), and high-resolution grabs (n = 72, m = 56 ± 4 g; taking one sample per stratum). Means and standard deviations of log-transformed aerobic plate counts (APCs) were 7.0 ± 0.3, 7.1 ± 0.2, and 7.3 ± 0.2 log(CFU/g) for boot covers, composite grabs, and high-resolution grabs, respectively. APCs did not show biologically meaningful differences between sample types. Boot covers recovered on average 0.6 log(CFU/g) more total coliforms than both grabs (p < 0.001) where means and standard deviations of log-transformed counts were 3.2 ± 1.0, 2.6 ± 0.6, and 2.6 ± 1.0 log(CFU/g) for boot covers, composite grabs, and high-resolution grabs, respectively. There were no generic E. coli detected in any sample by enumeration methods with LODs of 1.3-2.1 log(CFU/g) for boot covers and 0.5 log(CFU/g) for both grabs. By 16S rRNA sequencing, community species diversity (alpha diversity) was not significantly different within collection methods. While communities differed (p < 0.001) between soil sampling methods (beta diversity), variance in microbial communities was not significantly different. Of the 28 phyla and 297 genera detected, 25 phyla (89%) and 258 genera (87%) were found by all methods. Overall, aggregative boot cover sampling is similar to both grab methods for recovering quality and safety indicator organisms and representative microbiomes. This justifies future work testing aggregative soil sampling for foodborne pathogen detection.

Simulation Evaluation of Power of Sampling Plans to Detect Cronobacter in Powdered Infant Formula Production

Cronobacteris a hazard in Powdered Infant Formula (PIF) products that is hard to detect due to localized and low-level contamination. We adapted a previously published sampling simulation to PIF sampling and benchmarked industry-relevant sampling plans across different numbers of grabs, total sample mass, and sampling patterns. We evaluated performance to detect published Cronobacter contamination profiles for a recalled PIF batch [42% prevalence, -1.8±0.7 log(CFU/g)] and a reference, non-recalled, PIF batch [1% prevalence, -2.4±0.8 log(CFU/g)]. Simulating a range of numbers of grabs [n=1 to 22,000 (representing testing every finished package)] with 300 g total composite mass showed that taking 30 or more grabs detected contamination reliably (<1% median probability to accept the recalled batch). Benchmarking representative sampling plans ([n=30, mass grab=10g], [n=30, m=25g], [n=60, m=25g], [n=180, m=25g]) showed that all plans would reject the recalled batch (<1% median probability to accept) but would rarely reject the reference batch (>50% median probability of acceptance, all plans). Overall, (i) systematic or stratified random sampling patterns are equal to or more powerful than random sampling of the same sample size and total sampled mass, and, (ii) taking more samples, even if smaller, can increase the power to detect contamination.

A Validated Preharvest Sampling Simulation Shows that Sampling Plans with a Larger Number of Randomly Located Samples Perform Better than Typical Sampling Plans in Detecting Representative Point-Source and Widespread Hazards in Leafy Green Fields

Commercial leafy greens customers often require a negative preharvest pathogen test, typically by compositing 60 produce sample grabs of 150 to 375 g total mass from lots of various acreages. This study developed a preharvest sampling Monte Carlo simulation, validated it against literature and experimental trials, and used it to suggest improvements to sampling plans. The simulation was validated by outputting six simulated ranges of positive samples that contained the experimental number of positive samples (range, 2 to 139 positives) recovered from six field trials with point source, systematic, and sporadic contamination. We then evaluated the relative performance between simple random, stratified random, or systematic sampling in a 1-acre field to detect point sources of contamination present at 0.3% to 1.7% prevalence. Randomized sampling was optimal because of lower variability in probability of acceptance. Optimized sampling was applied to detect an industry-relevant point source [3 log(CFU/g) over 0.3% of the field] and widespread contamination [-1 to -4 log(CFU/g) over the whole field] by taking 60 to 1,200 sample grabs of 3 g. More samples increased the power of detecting point source contamination, as the median probability of acceptance decreased from 85% with 60 samples to 5% with 1,200 samples. Sampling plans with larger total composite sample mass increased power to detect low-level, widespread contamination, as the median probability of acceptance with -3 log(CFU/g) contamination decreased from 85% with a 150-g total mass to 30% with a 1,200-g total mass. Therefore, preharvest sampling power increases by taking more, smaller samples with randomization, up to the constraints of total grabs and mass feasible or required for a food safety objective. IMPORTANCE This study addresses a need for improved preharvest sampling plans for pathogen detection in leafy green fields by developing and validating a preharvest sampling simulation model, avoiding the expensive task of physical sampling in many fields. Validated preharvest sampling simulations were used to develop guidance for preharvest sampling protocols. Sampling simulations predicted that sampling plans with randomization are less variable in their power to detect low-prevalence point source contamination in a 1-acre field. Collecting larger total sample masses improved the power of sampling plans in detecting widespread contamination in 1-acre fields. Hence, the power of typical sampling plans that collect 150 to 375 g per composite sample can be improved by taking more, randomized smaller samples for larger total sample mass. The improved sampling plans are subject to feasibility constraints or to meet a particular food safety objective.

Understanding Current Methods for Sampling of Aflatoxins in Corn and to Generate a Best Practice Framework

Aflatoxin contamination in corn is a significant issue, posing substantial health threats to humans and animals. Aflatoxin testing protects consumer health, ensures the safe global trade of corn, and verifies compliance with legislation; however, effective sampling procedures are essential to ensure reliable results. While many sampling procedures exist, there is no evidence to indicate which is the best approach to ensure accurate detection. Using scientific and gray literature sources, this review analyzed sampling procedures to determine an optimum approach to guide the development of standard practices. Results revealed that sampling is the major source of error in the accurate assessment of aflatoxin levels in food and crucial for obtaining reliable results. To guarantee low variability and sample bias-increased sample size and sampling frequency, the use of automatic dynamic sampling techniques, adequate storage, and homogenization of aggregate samples for analysis are advised to ensure a representative sample. However, there is a lack of evidence to support this or indicate the current utilization of the reviewed procedures. Inadequate data prevented the recommendation of sample sizes or frequency for optimum practice, and thus, further research is required. There is an urgent need to make sampling procedures fit-for-purpose to obtain accurate and reliable aflatoxin measurements.