In Silico Evaluation of a Novel Iterative Bayesian Sampling Strategy for Efficient Detection of Pathogenic Bacteria in Preharvest Produce and Environments

Introduction:  Sampling of preharvest environments and produce is, increasingly, being used as a tool to enhance food microbial safety. In most sampling plans, the sample locations are determined beforehand and all samples are collected at once. This, in part, is because traditional methods of microbial detection take one or more days to yield results. However, recent development of rapid microbial methods allow the users to get test results much faster, which makes iterative sampling strategy possible.

Purpose:  The goal of this study was to evaluate the effectiveness of traditional sampling plans and a novel iterative sampling strategy, based on Bayesian Global Optimization (BGO), on simulated fields with realistic contamination sources.

Methods:  The effectiveness of the iterative BGO sampling plan and three traditional sampling plans (random, stratified-random, and z-pattern) were evaluated using a simulation model. Preharvest fields with realistic contamination sources were generated in silico. Three types of contaminations were considered, point contamination, line contamination and planar contamination. It was assumed that pathogen presence was correlated with indicator bacteria level. The BGO plan uses prior results to inform the subsequent sampling locations, to maximize overall detection probability. The same number of samples was collected in each sampling plan (n=18).

Results:  In simulated fields with five by six plots and nine subplots/plot (270 total sampling locations and six contamination sites on average), the BGO sampling plan dramatically increased detection probability compared to traditional sampling plans (random: 0.30±0.11; stratified random: 0.32±0.11; z-pattern: 0.32±0.17; BGO: 0.63±0.23). The difference was highly significant (P<0.0001).

Significance: This study provides a novel iterative sampling strategy for microbial quality testing. The sampling strategy gives much better detection probability than traditional sampling plans in realistic scenarios. This alternative sampling approach would be particularly beneficial when implemented as part of testing program that monitors preharvest fields over the course of the cultivation cycle.