The Web of Science Core Collection (WoSCC) is the most well-known and influential database of scientific literature and is the preferred database for bibliometric analysis (22). The WoSCC search was conducted on July 25, 2024. After removing duplicates through automated and manual verification, the final dataset included 2,408 publications from 2014 to 2024. All publications were weighted equally in the analysis to maintain methodological consistency across different research domains. The published period of this study is set between 2014 and 2024. The search terms are as follows: TS = (“anti-infective agent*” OR “anti-bacterial agent*” OR “antimicrobial agent*” OR “antimicrobial use” OR “antibiotic use” OR “antimicrobial residue” OR “antimicrobial resistance” OR “antibiotic resistance” OR antibiotic* OR antifungal* OR antiviral* OR antiparasitic*) AND TS = (“artificial intelligence*” OR “deep learn*” OR “machine learn*” OR “neural network*” OR “compu* intelligent*” OR robot). Only original articles and reviews written in English were included, resulting in a total of 2,408 publications being analyzed in our study (Figure 1).

Data downloaded from the WoSCC include fully recorded and cited references. We utilized three analysis tools: VOSviewer 1.6.17, CiteSpace 6.3.R3, and Microsoft Excel 2019. Key bibliometric measures include: co-authorship, co-citation, and co-occurrence. For example, co-authorship analysis focused on examining collaborative relationships among authors, nations, or institutions, as indicated by jointly authored papers. The analysis of co-occurrences offers a quantitative approach to identifying relationships among components, while co-citation analysis was employed to evaluate the strength of connections between frequently cited elements.

VOSviewer, developed by Leiden University’s Science and Technology Research Center in the Netherlands, is a tool for creating and exploring network data maps. It offers cluster, superimposed, and density views to evaluate research trends and hotspots (23). In the maps generated by VOSviewer, each node represents an element, with larger nodes indicating more reflections, and broader link widths between nodes indicating a higher level of cooperation. Microsoft Excel was employed to depict the worldwide production and evolutionary trends of relevant papers and to generate charts regarding the rankings of various factors.

CiteSpace is a citation visualization analysis software focused on revealing the potential knowledge contained within the scientific literature (24). Using CiteSpace, we analyzed keyword/reference clustering and timelines. The parameters for CiteSpace included: ① each slice represents a year from 2014 to 2024; ② single node type selection; ③ selection criteria based on the g-index, k = 15; and ④pruning performed using the pathfinder method. The impact factor (IF) and category quartile data were derived from the Journal Citation Reports (JCR) 2023. Scientific researchers, countries, journals, institutions, and journals were evaluated based on the H-index, a composite index measuring the quantity and quality of academic output.

A collection of 2,408 publications was obtained (key findings are summarized in Supplementary Table S2). As shown in Figure 2, publication output increased moderately from 2014 to 2020, followed by a sharp acceleration that peaked at 549 publications in 2023 (22.7% of total publications). This dramatic increase coincides with significant advances in AI applications for antimicrobial research. The growth pattern can be divided into three distinct phases: (1) an initial steady phase (2014–2019), reflecting the early adoption of AI technologies in antimicrobial research; (2) a transition phase (2019–2021), likely influenced by the emergence of more sophisticated machine learning algorithms and increased computing power; and (3) a rapid expansion phase (2021–2023), marked by breakthrough developments such as AlphaFold’s protein structure predictions and advanced deep learning applications in drug discovery. The steep acceleration post-2020 also correlates with increased funding initiatives and international collaborations focused on AI-driven solutions for antimicrobial drug development.

The top 10 countries ranked by their number of publications on the application of AI in antimicrobial agents are highlighted in Table 1. The United States leads with 707 publications, followed by China with 581, and India with 233. The international cooperation network, shown in Figure 3, reveals collaborative relationships among 64 countries. As shown in Figure 3A, the geographic distribution of research collaborations shows strong clustering in North America, East Asia, and Western Europe. Figure 3B further demonstrates that these collaboration networks are particularly dense between the United States and China.

Leading in publications within this field, the Chinese Academy of Sciences stands out with 53 publications, closely followed by Harvard Medical School and the University of California San Diego, with 43 and 37 publications, respectively, (Table 2). Notably, although Massachusetts Institute of Technology (MIT) ranks fourth with 28 publications, it holds the highest TLS value. This demonstrates that research from the MIT is both extensive and highly interconnected, highlighting its prominent role and influence in global research collaborations. Additionally, Figure 4A reveals the cooperation network among the top 89 institutions by publication volume, identifying two primary clusters. These clusters are distinguished by color, with red representing the cluster centered around the Chinese Academy of Sciences and green for Harvard Medical School. Table 3 lists the top 10 authors by publication volume in this field. Sean Ekins leads with 18 publications, followed by Chia-Ru Chung with 15, and Mahmoud Huleihel with 14. These numbers reflect both their productivity and their impact on the research community. Figure 4B further shows the collaborative networks among these authors.

Table 4 highlights significant contributions from leading journals in the application of AI to antimicrobial agents. “SCIENTIFIC REPORTS” tops the list with 73 publications, followed by “FRONTIERS IN MICROBIOLOGY” with 67, and “ANTIBIOTICS BASEL” with 50. The publication rankings correlate closely with their citation impacts, with “SCIENTIFIC REPORTS” achieving the highest co-citation count at 4,700. “NATURE COMMUNICATIONS” has the highest IF among the top 10 journals in both publication volume and citations. Furthermore, Figure 5 represents the extensive interconnectedness within scientific research, revealing robust citation flows among fields such as medicine, molecular biology, and health sciences. This comprehensive view highlights the collaborative and multidisciplinary nature of research in AI for antimicrobial applications.

The 10 most frequently co-cited references in research on AI and antimicrobial agents are detailed in Table 5. In 2021, the most frequently co-cited reference was “Highly accurate protein structure prediction with AlphaFold” by Jumper, originally published in Nature (25). This was followed by “A deep learning approach to antibiotic discovery” which appeared in Cell in 2020 (26). The two papers demonstrate significant advancements in the application of AI in antimicrobial agents: AlphaFold’s accurate protein structure predictions facilitate drug discovery and the development of antibiotics, while a deep learning approach has led to the identification of novel antibiotics such as Halicin, offering new avenues to address antibiotic resistance. CiteSpace visualized co-cited references in Figure 6A, and we constructed a network map of the top seven clusters shown in Figure 6B.

Clustering reveals dominant themes and trends by grouping similar references. A timeline analysis in Figure 7 further traces the evolution of these trends. Clusters #0 “antibacterial peptide (ABP)” #1 “antibiotic resistance prediction” and #3 “artificial intelligence” have been continuously active research hotspots since their emergence, significantly advancing solutions in global health through innovations in therapy development, drug-resistance management, and medical AI applications.

In the field of AI and antimicrobial agents, a total of 483 keywords were identified, with 33 of these appearing more than 50 times (Figure 8A). Supplementary Table S1 presents the top 20 keywords, which are the most frequently occurring. These keywords are grouped into several thematic categories: AI and data analysis, antibiotic and antimicrobial resistance, microbiology-related terms, and concepts related to biomolecular and drug development. “Machine learning” stands out as the most frequently mentioned keyword, followed closely by “antibiotic resistance” and “prediction” in terms of occurrence.

Cluster analysis was used to describe the relationships among the themes of keywords, revealing interconnections between AI and antimicrobial agents, as shown in Figure 8B. Timeline analysis further traced the evolution of these themes, as depicted in Figure 9, identifying eight clusters (#0 to #7), highlighting shifts in research focus and emerging trends. The enduring research clusters from 2014 to 2024 in the field of AI and antimicrobial agents—#0 sepsis, #1 artificial neural networks, #2 antimicrobial resistance, #3 antimicrobial peptides, #4 drug repurposing, and #5 molecular docking—highlight the persistent integration of AI in tackling key challenges in antimicrobial therapy.

The burst analysis of keywords depicted in Figure 10 underscores significant shifts in research focus and the emergence of new themes within this field. Key trends such as the sustained interest in “ABPs” and “artificial neural networks” from 2014, coupled with the rising concern over “drug resistance” and “Mycobacterium tuberculosis” beginning in 2019, reflect crucial developments and policy shifts. Additionally, the analysis highlights growing attention to “tetracycline,” “MALDI-TOF MS,” “big data” and “strategy” indicating a broadening of research scopes and the integration of new technologies and methodologies.

This study analyzed papers over the past decade on AI in antimicrobial drugs, highlighting a marked increase in publications since 2021, peaking in 2023. This trend reflects the urgent need to combat global AMR, emphasizing the vital role of AI. Significant international collaborations, especially among leaders like the United States, China, and India, are essential for fostering technological innovation and interdisciplinary approaches to complex health challenges. The United States leads globally due to its strong research infrastructure and substantial support for new drug development (27). China capitalizes on its extensive medical data and rapid AI innovation (28), while India benefits from its robust IT sector and cost-effective drug development (29).

The contributions of institutions like the Chinese Academy of Sciences, Harvard Medical School, and UC San Diego in AI for AMR are underscored by leading researchers like Sean Ekins, Chia-Ru Chung, and Mahmoud Huleihel. Their prolific work advances the field and highlights the critical role individual researchers play in shaping AMR research (30–33). These institutions stand out due to their robust infrastructure and substantial funding, enabling significant advancements in AMR solutions. Meanwhile, “SCIENTIFIC REPORTS” plays a key role in advancing AI research on antimicrobial drugs, offering a platform for diverse multidisciplinary studies. In contrast, “NATURE COMMUNICATIONS” with its high IF, distinguishes itself by publishing influential, high-quality research that drives global scientific agendas. Additionally, the extensive citation networks connecting disciplines such as medicine and molecular biology further highlight the essential role of interdisciplinary collaboration in addressing AMR challenges.

The intersection of AI and biomedical research has given rise to significant advancements in the fight against microbial threats. Among these, three studies stand out for their innovative approaches and substantial impact on the field of antimicrobial research.

As the top-ranked co-cited reference (25), the study “Highly accurate protein structure prediction with AlphaFold” by Jumper, published in Nature (2021), marks a pivotal advancement at the juncture of AI and biomedical research. This groundbreaking work on accurately predicting protein structures significantly enhances the field of antimicrobial drug design. By delivering detailed structural predictions of pathogenic proteins, AlphaFold enables the development of targeted antimicrobial therapies that are not only more effective but also less prone to resistance. Additionally, it provides novel insights into ARmechanisms by predicting structural changes in proteins due to mutations, aiding in the design of new drugs to counteract resistance. The rapid and efficient predictions made by AlphaFold facilitate accelerated research and development cycles, essential for keeping pace with swiftly evolving microbial threats. This progress not only advances antimicrobial drug research but also demonstrates the transformative potential of applying advanced computational technologies to biological research, thereby opening new pathways for addressing complex health challenges.

As the second most co-cited reference (26), the paper “A Deep Learning Approach to Antibiotic Discovery” by Stokes, published in Cell (2020), exemplifies the profound impact of computational techniques on antimicrobial drug discovery. This innovative study employed a deep learning model to screen over 107 million compounds efficiently, leading to the discovery of Halicin, a novel antibiotic effective against a broad spectrum of resistant pathogens, including Mycobacterium tuberculosis. This research not only illustrates the capacity of computational models to both accelerate and innovate within the antibiotic discovery process but also tackles the pressing issue of antibiotic resistance. It introduces a scalable and novel approach that rejuvenates the antibiotic development pipeline, potentially addressing global health threats.

As the third most co-cited reference (34), “CARD 2020: antibiotic resistome surveillance with the comprehensive antibiotic resistance database” by Alcock, published in Nucleic Acids Research (2020), highlights the essential role of informatics in combating AMR. The Comprehensive Antibiotic Resistance Database (CARD) offers a meticulously curated collection of DNA and protein sequences, along with detection models and bioinformatics tools designed to elucidate the molecular basis of bacterial AMR. This work significantly advances AMR surveillance and analysis capabilities, supporting widespread applications in public health, clinical settings, agriculture, and environmental studies. Through ongoing updates and community-led improvements, CARD not only deepens our understanding of AMR patterns but also strengthens global efforts to effectively manage and reduce antibiotic resistance.

In the field of antimicrobial drugs and AI, significant advancements have been realized through the strategic integration of machine learning technologies. These developments, centered around the key terms “machine learning,” “antibiotic resistance,” and “prediction,” form the backbone of current research efforts aiming to combat the rising challenge of ARefficiently.

In recent years, machine learning has significantly advanced the prediction of antimicrobial drugs and detection of antibiotic resistance. Utilizing deep learning and other machine learning models, researchers have improved diagnostic accuracy. Notably, Stokes (26) used a deep neural network to discover Halicin, a new antibiotic effective against resistant pathogens in mouse models. The evolution of deep learning has also popularized molecular graph representations, enhancing the prediction of antimicrobial activity through graph neural networks (35). Furthermore, Arango-Argoty (36) created DeepARG, a tool for predicting ARgenes using deep learning, and Khaledi (37) applied machine learning to help physicians select targeted antibiotics based on genomic data. AR has become a major challenge in the global public health arena. According to Murray, in 2019, there were 4.95 million deaths associated with bacterial antibiotic resistance, of which 1.27 million deaths were attributed directly to bacterial antibiotic resistance (38). The World Bank estimates that by 2050, ARcould reduce global GDP by up to 3.8% (39). Although the overuse or misuse of antibiotics is a primary driver of resistance, other interconnected factors such as environmental contributors, agricultural practices, and global disparities in antibiotic exposure also exacerbate the prevalence and spread of resistance. For example, the widespread use of antibiotics in agriculture leads to the spread of resistant bacteria through the food chain and environment, impacting human health (40), while urban wastewater treatment plants have become hotspots for antibiotics and resistance genes entering the environment (41). Current antimicrobial prediction models, such as deep learning and random forests, have shown great potential in resistance prediction. However, these models face challenges related to data imbalance, quality control, and missing data (42, 43). Future research should focus on improving model accuracy through data balancing techniques and integrating multiple data sources (such as clinical, genetic, and chemical data) to provide more comprehensive predictions (44).

Recent clinical implementations of AI models in antimicrobial research have shown promising results in several key areas. For example, the clinical validation of deep learning models for antibiotic discovery has shown tangible outcomes. As demonstrated by Stokes, deep learning approaches led to the discovery of novel antibiotics effective against resistant pathogens, particularly against MRSA and resistant Enterococci (26). Furthermore, Wong successfully employed graph neural networks to predict both antibiotic activity and cytotoxicity across over 12 million molecules, identifying compounds with strong antibiotic activity and low toxicity (13). This systematic approach shows great promise for clinical applications. However, challenges remain in widespread clinical adoption, including the need for extensive validation across diverse patient populations and integration with existing hospital information systems.

Based on our analysis of publication trends and citation patterns, specific AI models and research priorities emerge as particularly promising for near-term clinical impact. The bibliometric data suggests that graph neural networks, as demonstrated by Wong show particular promise for immediate clinical application due to their proven capability in simultaneously predicting both antibiotic activity and cytotoxicity (13). Additionally, deep learning models for antimicrobial resistance prediction, as shown in Khaledi’s work, demonstrate strong potential for rapid clinical implementation in resistance prediction (37). These approaches are particularly valuable as they can be integrated into existing clinical workflows with minimal disruption.

This study has some limitation, including the use of English only articles from the WoSCC database. Exploring regional databases and non-English publications could offer a more comprehensive view of AI applications in AMR, particularly from regions with high AMR burden but potentially underrepresented in WoSCC. On the positive note, WoSCC is quite expansive but the cost is that great research done in other languages or databases may not be taken into consideration. In addition, the fact that citation delays might lead to underrepresentation of recent high-quality publications could affect the accuracy of trend analyses. However, one cannot deny the fact that this study is appropriate for exploring promising research topics in the development of antimicrobial resistance, revealing some specific research areas that hold the highest potential, as well as the subsequent areas for potential research. The exclusive use of WoSCC is defendable since it along with other limitations offers broad fulltext access and citation analysis, which PubMed or Embase, for instance, does not afford for a bibliometric study.