The year 2025 saw substantial changes for Nature Biomedical Engineering, with the addition of new team members and the appointment of a new Chief Editor, Rita Strack. Nevertheless, our team’s commitment to publishing exciting and impactful work remained steadfast. Here we celebrate the end of the year by highlighting a few of our editorial team’s favourites from 2025 in some of the key areas that define Nature Biomedical Engineering.

In the genome editing space, Wang, Cong and colleagues1 made a splash with CRISPR-GPT, a tool that bridges areas of machine learning and gene editing. The model leverages the reasoning abilities of large language models, combined with human input and built-in expert knowledge, to automate and optimize experimental design and analysis for specific genome-editing tasks. It can generate end-to-end workflows, offering recommendations on CRISPR systems, gRNA design, delivery methods, protocols and validation assays and even predicting off-target effects and editing outcomes. Separate work by Dong, Zhou, Chen and colleagues introduces a valuable gene-editing platform using Cas12a-knock-in mice2. Combining this model with an existing transgenic line further enables simultaneous dual-gene activation and knockout, underscoring the versatility of the Cas12a-knock-in mouse system for the study of complex genetic interactions. A final paper to note in this field, from Asokan and colleagues, shows the development of two kidney-specific adeno-associated virus (AAV) vector variants discovered by cycling structure-guided AAV capsid libraries through mouse, pig and non-human primate kidneys and human organoids3. The AAV-kidney variants demonstrate improved transduction in vivo in pig and non-human primate kidneys compared to parental AAV9, showing potential to address current limitations in therapeutic gene transfer to the kidneys.

The microbiome engineering field witnessed a methodological advance from Wang and colleagues4, reporting an in situ platform, BACTRINS, that uses CRISPR-associated transposon systems for the inactivation of toxin genes in bacterial cells. The platform is validated in a mouse model of Shiga toxin–producing Escherichia coli infection. As a result, Shiga toxin gene is inactivated and the bacteria not only become non-pathogenic but can further promote protection by expressing the nanobody payload to reduce pathogen attachment to the gut epithelia.

It was a strong year as well for machine learning at the journal, with studies advancing the application of artificial intelligence in understanding, diagnosing and treating disease. Among many noteworthy contributions, a few stood out for their potential long-term impact. For example, DrugGPT from Liu and colleagues5 is a chatbot that can help clinicians to make individualized decisions on drug treatments and dosages. Importantly, the work bypasses problems with hallucinations and misinformation associated with large language model-based chatbots for accurate, evidence-based recommendations. Another standout comes from Yan and colleagues6, who explored the use of realistic synthetic data for training foundation models, seeking to overcome a fundamental limitation associated with training large models — that of acquiring massive training datasets. Through their RETFound model for retinal images, they demonstrate that their data-efficient strategy does not compromise performance. Notably, work from Dar and colleagues7 shows that generative artificial intelligence models meant to generate realistic synthetic data can memorize and recreate specific patient data, resulting in patient-reidentification, offering an important warning about privacy to developers, especially those seeking data-efficient training strategies.

Point-of-care diagnostics papers published this year offered advances in performance, automation and methodological development. Shafiee and colleagues8 report a bioluminescence system, LUCAS, that leverages an enzyme cascade reaction and uses a magnetic bead-based immunoassay to reach prolonged bioluminescence, high sensitivity and high accuracy, without any nucleic acid extraction or amplification. Validated for detection of SARS-CoV-2, and pathogens such as human immunodeficiency virus, hepatitis B virus and hepatitis C virus from patient samples, the platform is implemented on a portable, fully automated device. Another outstanding advance for point-of-care diagnostics comes from Ning, Hu and colleagues9, who use a microchip device that uses fingerstick whole-blood microsamples for detection of Mycobacterium tuberculosis in immunocompromised individuals. The method is based on quantification of antigen-specific T cell responses and does not need the complicated equipment typically required in laboratories or hospitals. This achievement pushes immunology to the point-of-care level.

The journal has also featured a rich collection of tissue and disease modelling papers in 2025, showcasing their diverse applications across various organs. Wang, Leong, Shi and colleagues develop a 3D printed model of healthy and diseased penile physiology, mimicking the complex vasculature of the corpus cavernosum10. After engineering animal models of penile dysfunction, the authors used the 3D-printed tissue seeded with endothelial cells to restore normal erectile and mating functions in rabbits and pigs. Nature Biomedical Engineering further emphasized the power of established tissue models in accelerating therapeutic developments. In a 3D microphysiological human-derived cardiac model, Srivastava, Murthy, Healy and colleagues rapidly screen and identify the ideal lipid composition for improved heart-specific delivery of lipid nanoparticle–mRNA complexes and enhanced gene editing outcomes11. They validate their results in vivo in mouse models of myocardial infarction, demonstrating the power of established tissue models in accelerating therapeutic developments.

In cancer biology, Miyako and colleagues12 leverage microbiology to develop a tumour-tropic bacterial consortium that induces selective intratumoural thrombosis and vascular collapse. Such bacteria-based cancer therapies lead to complete tumour regression without immune involvement or genetic engineering.

In immunotherapy, the standout advances we published use tools such as synthetic biology and protein engineering to tackle hurdles in oncology. Wilson and colleagues13 commit to overcome the challenge of systemic delivery of STING agonists (mostly hampered by poor pharmacokinetics and limited tumour accumulation) by engineering albumin-hitchhiking nanobody–STING agonist conjugates. Such conjugates extend circulation, concentrate in tumours and synergize with checkpoint blockade, offering a programmable route to rewire tumour immunity. El Hebieshy, Verdoes, Scheeren and colleagues14 address the persistent issues of heterogeneity and poor control over multivalency in antibody-based therapeutics with ubi-tagging, a rapid, site-specific conjugation platform. This technology enables modular assembly of bispecific T cell engagers and dendritic cell-targeted vaccines, overcoming inefficiencies of current chemoenzymatic methods. Together, these studies exemplify how mechanistically distinct advances can overcome long-standing barriers in cancer therapy.

Last but not least, in neuroscience, mechanistic work by Pan and colleagues15 reported evidence that cerebellum processes motor frequency with precision and generalizability across individuals. The aim is to generate a model for motor control that is universal and could potentially replace personalized artificial intelligence models that rely on personalized training sets, allowing them to act on patients that have already lost the required brain functions for training sets. The work is validated in in vivo models by using optogenetics for mice and current stimulation for humans.

We are delighted to close the year with a showcase of our favourite articles, which are a small sampling of the many important papers it was our privilege to publish this year. We look forward to welcoming the new year with equally outstanding content on innovative bioengineering strategies with therapeutic and translational value across disciplines. We thank all our readers, reviewers and authors for their support and wish you a Happy New Year!