AI-Powered Cancer Drug Research Has Exploded Since 2018, Landmark 15,554-Study Analysis Reveals

Artificial intelligence has quietly become one of the most powerful forces reshaping how humanity fights cancer, and for the first time, researchers have mapped the entire landscape of this revolution. A sweeping bibliometric analysis published in Clinical Cancer Bulletin has examined 15,554 publications spanning 2011 to 2025, offering the most comprehensive picture yet of how AI-driven anticancer drug design has evolved from a niche computational curiosity into a global scientific enterprise. The findings reveal a field in exponential ascent, with an average annual growth rate of 48.22 percent in publication output since 2018, a surge the authors attribute to landmark advances such as IBM Watson’s success in clinical trial matching and AlphaFold’s breakthroughs in protein structure prediction.

The scale of the analysis is itself remarkable. Researchers led by Mengyao Sun, Yue Yin, and Zejun Jia of Zhongshan Hospital, Fudan University, searched the Web of Science Core Collection using an elaborate query that combined artificial intelligence terms, ranging from machine learning and deep learning to large language models such as ChatGPT and BioGPT, with cancer terminology and drug design vocabulary. After rigorous screening that excluded veterinary studies, publications not employing AI methods, and research unrelated to anticancer therapy, the final dataset comprised 12,906 original research articles and 2,648 review articles. Each publication was then dissected using a multi-tool analytical arsenal including Excel 2025, CiteSpace version 6.4.1, VOSviewer version 1.6.20, and the Bibliometrix R package, with the entire procedure reported in alignment with international bibliometric reporting guidelines.

The geographic distribution of this research output tells a story of shifting global power in science. China emerged as the undisputed leader in total publication volume, contributing 41.9 percent of all publications with 6,514 articles, followed by the United States with 2,802 publications and India with 812. In 2022, China overtook the United States to become the leading annual producer. Yet the picture is more nuanced than raw numbers suggest. Among the top ten contributing countries, China had the lowest proportion of internationally collaborative publications at just 12.9 percent, while the United Kingdom led with a striking 55.9 percent of its papers co-authored across borders. Perhaps more tellingly, developed countries demonstrated significantly higher average citation counts than their developing counterparts, a statistically significant disparity reflecting differences in journal provenance, research infrastructure, funding intensity, and the strength of international networks.

At the institutional level, Harvard University generated the highest number of publications, followed closely by the Chinese Academy of Sciences and the University of California System. Cluster analysis revealed two prominent global scientific cooperation networks, one centered on the United States and the other on China, effectively dividing the field into two gravitational spheres. The pharmaceutical industry is deeply embedded in this landscape, with Roche, Pfizer, Novartis, AstraZeneca, and Merck among the contributing corporations. Funding data show the National Natural Science Foundation of China and the United States Department of Health and Human Services as the foremost sponsors, while AstraZeneca, Pfizer, Novartis, and Roche lead corporate investment, focusing primarily on early-stage drug discovery rather than clinical validation.

Individual researchers have also left indelible marks on the field. Professor Alex Zhavoronkov was identified as the most prolific author with 55 publications, and notably, half of the top ten high-output authors hail from Insilico Medicine, underscoring the outsized influence of industry in this domain. Professor Michael Patrick Menden received the highest number of citations, and every highly cited author is an expert in either information science or medicine, confirming the deeply interdisciplinary character of AI-assisted drug design. Interestingly, the majority of highly cited scholars are concentrated in Europe, home to institutions such as the European Molecular Biology Laboratory-European Bioinformatics Institute and the German Helmholtz Association, which fostered early integration of biology, chemistry, and computational sciences. Europe also nurtured pioneering companies like Exscientia and catalyzed AlphaFold itself. By contrast, although Chinese researchers hold four of the top ten positions in publication volume, none appeared on the highly cited list, a gap the authors suggest reflects the nation’s status as a rising star that must now prioritize research quality over quantity.

The keyword analysis paints a vivid portrait of what scientists are actually studying. The terms artificial intelligence, immunotherapy, and breast cancer dominated, with breast, prostate, lung, and liver cancers attracting the greatest attention. This concentration is no accident. Breast and prostate cancers rank among the most prevalent malignancies in women and men across Europe and the United States, and their favorable five-year survival rates, exceeding 90 percent for breast cancer and 98 percent for prostate cancer, create substantial commercial incentives. Breast cancer alone accounts for 7.7 percent of the global economic cost of cancer, making it the third most economically burdensome malignancy. Both cancers also possess well-established molecular classification systems and clearly defined druggable driver targets, which make them ideal testing grounds for AI technologies. Across cancer types, drug development converges on a limited set of validated targets: HER2, estrogen receptor, and CDK4/6 in breast cancer; EGFR tyrosine kinase inhibitors in non-small cell lung cancer; the androgen receptor in prostate cancer; and immune checkpoint inhibitors in hepatocellular carcinoma. This pattern, the authors note, reflects a persistent me-too and me-better development paradigm, with genuine first-in-class innovation remaining scarce.

The technological evolution of the field reads like a history of machine learning itself. In the early period from 2011 to 2012, support vector machines reigned supreme, with studies concentrated on specific diseases and drugs such as breast cancer, aromatase inhibitors, and tamoxifen. Between 2012 and 2017, random forests and artificial neural networks gained prominence as machine learning became systematically integrated into drug design for property prediction and molecular modeling. From 2018 to 2022, big data, web servers, and convolutional neural networks emerged as dominant themes, marking a transition to deep learning applied to massive datasets and online services that lowered barriers to entry. More recently, the scope has broadened from traditional structure-based drug design toward predicting pharmacodynamic efficacy, and from small molecules to innovative biotherapeutics including tumor vaccines, therapeutic antibodies, and antibody-drug conjugates. Immunotherapy has become a leading focal point, with AI being applied to neoantigen prediction, antigenic peptide design, and the optimization of T cell, dendritic cell, and natural killer cell therapies. Emerging hotspots include Toll-like receptor agonists as vaccine adjuvants, macrophage polarization, neutrophil extracellular traps, and the transcription factor STAT3.

Despite the dazzling growth, the analysis unflinchingly documents the field’s structural weaknesses. Tumors are extraordinarily complex biological systems: high-grade gliomas exhibit intratumoral heterogeneity, immunosuppressive microenvironments, glioma stem cells, and the physical barrier of the blood-brain barrier, while phenotypic plasticity, now recognized as a hallmark of cancer, allows tumor cells to dedifferentiate, resist drugs, and even switch lineages, as when lung adenocarcinoma transforms into small cell lung cancer. Most AI models are trained on static, reductionist datasets such as molecular structures or in vitro assays, blind to the dynamic, adaptive nature of tumors in living patients. The field also suffers from a paper-driven rather than need-driven orientation: algorithmic publications proliferate because entry barriers are low and publication is fast, while clinical translation remains sparse, with the probability of market approval hovering at approximately 5 percent even after phase 1 trials. Data fragmentation compounds the problem, with models trained on public databases like ChEMBL and TCGA that suffer from batch effects, inconsistent standardization, and shallow clinical annotations, creating what the authors describe as a data archipelago. Between 2019 and 2024, pharmaceutical companies using AI in Europe, the United States, and the Asia-Pacific region faced significant data breaches, highlighting the urgent challenges of privacy, security, and regulatory compliance, particularly when human genetic resources are involved. The black-box nature of many sophisticated models further conflicts with regulatory demands for clear mechanisms of action.

The path forward, the authors argue, demands a fundamental reorientation toward clinically driven innovation. Priorities include integrating multi-omics data spanning genomics, proteomics, metabolomics, lipidomics, and spatial transcriptomics; building specialized disease cohort databases including patient-derived xenograft models and organoid biobanks; and adopting federated learning frameworks that allow collaborative model training while protecting privacy. Closed-loop validation systems that combine AI with active learning and high-throughput wet-lab platforms such as CRISPR screens and microfluidic organ chips could finally connect computational predictions to biological reality. AlphaFold-style protein structure prediction, geometric deep learning frameworks for RNA-ligand interactions, and generative AI for de novo protein design are expanding the druggable space beyond traditionally undruggable targets, while machine learning-enhanced nanoparticles, liposomes, extracellular vesicles, and even nanorobots promise precision drug delivery. AI has already demonstrated the ability to cut research and development costs by more than 40 percent and compress timelines from years to months, and AI-optimized anticancer drugs such as CV8102, PRT3789, ISM6331, and ISM5043 have reached clinical trials. If the field can marry its computational firepower with biological insight, explainable algorithms, and rigorous clinical validation, the vision of truly AI-designed cancer medicines may finally move from promise to prescription.

Subject of Research: Bibliometric analysis of global research trends in AI-driven anticancer drug design from 2011 to 2025

Article Title: Global research status and trends in the AI-driven anticancer drug design: a bibliometric analysis of 2011–2025

Article References: Sun, M., Yin, Y., & Jia, Z. (2026). Global research status and trends in the AI-driven anticancer drug design: a bibliometric analysis of 2011–2025. Clinical Cancer Bulletin, 5(1), Article 8. https://doi.org/10.1007/s44272-026-00060-8

Image Credits: AI Generated

DOI: 10.1007/s44272-026-00060-8

Keywords: artificial intelligence, anticancer drug design, bibliometric analysis, immunotherapy, machine learning, drug discovery, breast cancer, clinical translation, multi-omics, AlphaFold, tumor microenvironment, China