Core Concepts
A framework combining deep latent space mapping and Bayesian optimization can efficiently generate synthetic microbiome samples that predict low levels of multidrug-resistant Salmonella, Listeria, and Campylobacter.
Abstract
This study presents a novel approach to generate synthetic microbiome samples that can predict reduced prevalence of multidrug-resistant (MDR) foodborne pathogens like Salmonella, Listeria, and Campylobacter. The key components of the framework are:
Multi-class classification model: A neural network is trained to predict the MDR profiles of the three pathogens based on microbiome samples. Oversampling techniques are used to address class imbalance in the dataset.
Autoencoder: An autoencoder is employed to learn a low-dimensional latent representation of the microbiome samples, enabling efficient generation of synthetic samples.
Bayesian optimization: Four acquisition functions (expected improvement, upper confidence bound, Thompson sampling, probability of improvement) are evaluated to efficiently search the latent space and identify synthetic microbiome samples that predict low MDR levels, requiring fewer iterations compared to random sampling.
The results show that the expected improvement, upper confidence bound, and probability of improvement acquisition functions consistently outperform Thompson sampling in identifying synthetic microbiomes with reduced MDR predictions. The study also provides insights into the key microbiome taxa, such as Bacteroides, Ruminococcaceae, and Chlamydiaceae, that influence the reduction of MDR pathogens. This framework demonstrates the feasibility of creating bespoke synthetic microbiomes with customized MDR profiles, which can accelerate microbiome research and address antimicrobial resistance challenges.
Stats
Salmonella MDR prediction for synthetic sample: 4.4e-6
Listeria MDR prediction for synthetic sample: 1e-15
Campylobacter MDR prediction for synthetic sample: 2e-3
Quotes
"By combining deep latent space mapping and Bayesian learning for efficient guided screening, this study demonstrated the feasibility of creating bespoke synthetic microbiomes with customized MDR profiles."
"The validity of our results has been thoroughly tested against reasonable baselines, ensuring the robustness of our findings."