New Platform Leverages AI and Quantum Computing to Predict Salmonella Antimicrobial Resistance

Antimicrobial-resistant Salmonella strains are a growing public health concern due to the overuse of antimicrobials and the rise of genetic mutations. Accurate prediction of resistance is crucial for effective treatment, yet traditional methods such as bacterial antimicrobial susceptibility tests (ASTs) are inefficient. Predictive models based on whole-genome sequencing (WGS) data also face challenges, particularly overfitting due to the high dimensionality of genomic data. Additionally, these models struggle with imbalances in the number of resistant and sensitive samples, further limiting their reliability. Now, researchers have developed an innovative prediction platform that integrates advanced artificial intelligence (AI) and quantum computing technologies to improve accuracy and efficiency in antimicrobial-resistance prediction.

The new solution, developed by scientists at Sichuan University (Chengdu, China), is called the Salmonella Antimicrobial-Resistance Predictive platform based on Large Language Models (SARPLLM). It employs a two-step feature-selection process that begins with a chi-square test and conditional mutual information maximization to identify key resistance genes through pan-genomic analysis. The SARPLLM algorithm itself is based on the Qwen2 large language model and incorporates low-rank adaptation (LoRA) to convert Salmonella samples into sentence-like structures for AI processing.

To resolve data imbalance, the researchers created the QSMOTEN algorithm—an advancement of the SMOTEN method that uses quantum computing to encode genomic features into quantum states and calculate sample distances more efficiently, reducing computational complexity from linear to logarithmic scale. In addition, the team launched a user-friendly online platform built with Django as the back-end and Echarts for knowledge graph visualization. It includes modules for resistance prediction, pan-genomic results, gene-antimicrobial interaction mapping, and data downloads.

Experimental validation showed that SARPLLM achieved higher F1-scores than existing models for multiple antimicrobials, while QSMOTEN demonstrated strong performance on both virtual and physical quantum machines in computing genomic similarities. The study, published in Engineering, confirms that combining AI with quantum computing can significantly accelerate and enhance resistance prediction. The platform’s practical utility is strengthened by its ability to provide online data processing, visualization, and downloadable results.

However, researchers noted that current large language models still struggle to fully grasp the complex biological and genetic foundations of antimicrobial resistance, and that the accuracy of predictions is influenced by the quality of training data. The team plans to integrate multi-source data and domain-specific knowledge in future iterations and to advance more stable quantum hardware to improve performance further.