Search for a command to run...
Antimicrobial resistance (AMR), when bacteria develop the ability to survive medicines meant to kill them. It is one of the biggest health threats we face today[1]. During AMR, bacteria become resistant to antibiotics, and common infections become harder to treat. This problem is getting worse worldwide[2]. Scientists are now using generative artificial intelligence (GenAI), a type of AI system that can create new synthetic data, along with machine learning (ML; computer programs that learn from large datasets), deep learning (DL; an advanced form of ML that uses complex neural networks called deep neural networks), and statistical models to predict how AMR will spread. They are using these technologies across different fields of biomedical science to analyze the data[3–6]. This new approach could help us fight resistance before it becomes a crisis. Traditional methods for tracking AMR have limitations: they deliver incomplete, delayed data and reveal only past events[7]. Building on these new technologies, GenAI can generate synthetic data to fill surveillance gaps and simulate scenarios of resistance evolution. These technologies quickly analyze vast data such as hospital records, prescription patterns, and lab results. They may help to detect patterns humans might miss, helping predict where and when resistance may appear. ML is used to predict AMR to different antibiotics from data on the gene-to-genome composition of pathogens. Recently, Kim et al published an extensive review of ML-based AMR prediction. This review shows how ML is increasingly used to predict AMR from genomic data[8]. Researchers use several ML algorithms to understand AMR (Table 1). It offers faster diagnostics and better surveillance of AMR. The review also discusses current problems of AMR. Sometimes, ML models lack interpretability of their results. Therefore, it underscores the need for transparent, clinically reliable ML models. These models are needed to bridge the gap from research to real-world practice. Table 1 - Advanced AI related algorithm in AMR study. Sl. No. New algorithm Learning method Remarks Reference 1. Neural networks This models lightly encouraged by the human brain’s structure along with the DL models, it also capable of modeling complex nonlinear relationships, but the large amounts of data required It enable adaptive, data-driven modeling of complex, non-linear relationships in AMR studies, thereby improving the accuracy of resistance prediction and pattern recognition across diverse microbial datasets [PMID: 35 373 160] 2. Random forest Decision trees sets having internal nodes comprising a series of questions about appropriate features, diverse answers are focused to separate child nodes till reaching the final class label This enhance AMR studies by providing robust, interpretable classification of resistance phenotypes through ensemble learning, effectively handling high-dimensional genomic and phenotypic data while reducing overfitting [PMID: 18 779 814] 3. Logistic regression The type of regression algorithm along with logistic curve, it links weights to each input features It serves as a transparent and statistically robust baseline in AMR studies, enabling interpretable estimation of the association between microbial features and resistance outcomes while facilitating risk prediction and hypothesis testing [https://doi.org/10.1038/nmeth.3904]. 4. Rule based IF-THEN statements sets It contribute to AMR studies by encoding expert knowledge and known resistance mechanisms into transparent decision rules, allowing explainable and clinically interpretable prediction of antimicrobial resistance patterns [PMID: 27 671 088] Along with the condition and a prediction 4. Support vector machines Separates labeled training data via constructing an optimal hyperplane, grouping appropriate genes, k-mer, or SNV features together In AMR studies leverage predefined biological rules and resistance determinants to deliver highly interpretable, explainable predictions that align closely with established antimicrobial resistance mechanisms [PMID: 17 160 063] 5. Recurrent neural network This neural network algorithm designed to process sequential data by using internal memory to remember previous inputs Such algorithm excel at processing sequential data, making them valued in time series analysis of bacterial growth patterns and antibiotic treatments [PMID: 38 927 169] 6. Gradient boosting machines It builds a predictive model in a stage-wise fashion by combining a series of simple, “weak” models, typically decision trees This algorithm used in the discovery of genetic markers associated with antibiotic resistance [PMID: 37 815 417] In addition to ML, DL models are also useful for tracking resistance. This method can work with different factors, such as the antibiotics people use, economic conditions, and genetic changes caused by bacterial mutations. Bringing together multiple sources of information gives us a complete picture of how resistance develops and spreads. The method can help us diagnose and treat antibiotic resistance[9]. Statistical models are also useful for studying antibiotic resistance. Despite the promise of these advanced tools, real challenges remain. The biggest is data quality. Many countries, especially poorer ones, lack effective surveillance systems – often where AMR is worst. Without quality data, predictions cannot be accurate. GenAI can generate synthetic datasets, but they must be carefully validated. Another issue involves understanding how these models work in practice. DL and GenAI models can make accurate predictions, but they often operate like a “black box.” Doctors and policymakers need to understand why a model makes certain predictions. This understanding builds trust and helps people to use these tools. Moreover, the landscape of AMR is constantly evolving. New resistance mechanisms emerge over time. Although new antibiotics are being developed, with some currently in clinical trials, it is essential that researchers, doctors, and other stakeholders adopt new methods and models to keep up with and analyze them. To control AMR, GenAI should be used. GenAI-driven real-time monitoring systems should be developed to monitor data continuously. It should continuously analyze genetic, clinical, and environmental data. In this regard, GenAI can provide early warnings. It will be guiding both public health and clinical responses. Additionally, GenAI holds promise for drug discovery, especially antibiotic discovery, to combat the threats posed by AMR. However, realizing the full potential of GenAI requires more than just technology. Countries must invest in data collection systems. Scientists from different countries must work together. We need rules for fairly using GenAI in healthcare. Most importantly, these tools must be available to all countries, not just wealthy ones. Using ML, DL, and GenAI to predict AMR is a major step forward. Instead of just watching resistance spread, we can now predict where it will go next and forecast its future. GenAI enables simulating future scenarios and testing solutions before implementation. This gives us time to prepare and respond. We can potentially stop outbreaks before they start. At the same time, there should be clear rules for AI and its reporting. In this regard, Agha and his colleagues recently developed a guideline for AI reporting (termed the TITAN guideline). It tried to establish a standard for AI reporting. It might serve as a benchmark[10]. During the fight with AMR, success requires teamwork, collaboration, and strong data systems. We must analyze real-world AMR data in the light of public health using these new tools and technologies. The challenge is urgent. Our ability to treat infections, especially those enabled by AMR, depends on how well we use these advanced tools. AMR could bring us back to a time before antibiotics existed, but these predictive technologies, especially GenAI, offer hope to prevent that future.