基于AME-MoE模型的微生物浓度检验方法OA

To address the issues of strong lag,high cost,and poor interpretability in traditional microbial concentration test methods,a new method of microbial concentration test based on an artificial microbial ecosystem and mixture-of-experts(AME-MoE)model was proposed.The model decomposed dynamically the complex microbial communities into specific bacterial species response modeled by multiple lightweight expert submodels,and introduced a Transformer to enhance the model's capture of nonlinear relationships,ultimately generating microbial concentration detection results.The research demonstrated that,referring to colony forming unit(CFU),the root mean square error of AME-MoE model was as low as 0.15lg(CFU/g),with a metabolic flux detection accuracy of 88.7%.Additionally,the single-sample inference latency time was only 22 ms,and the model exhibited excellent generalization capabilities in scenarios with data scarcity and in a noisy environment.AME-MoE model achieved high-precision,low-latency,and strongly generalizable microbial concentration intelligent detection,providing a decision-making tool with practical application value for hospital infection control,food safety,and real-time monitoring of environmental microbial risks.