Cell division is a fundamental process essential for life, underpinning reproduction, development and tissue maintenance across all organisms and enabling population growth and evolutionary adaptation. Recreating this capability is, therefore, a central challenge in bottom–up synthetic biology, wherein the aim is to construct functional synthetic cells. In recent years, substantial progress has been made toward building a synthetic divisome through partial reconstitution of the protein machinery underlying cell division in vitro. Here, we review current strategies to mimic the key stages of division: symmetry breaking to define the division site, membrane deformation to drive constriction and, thus, shape changes of the cell, and the final abscission event. We critically assess the successes and limitations of these approaches and discuss how integrating multiple modules may enable the realization of a minimal, functional division system for synthetic cells. Cell division consists of three elementary steps: symmetry breaking, membrane constriction and abscission. Here, we review current strategies for achieving these steps in synthetic cells, assessing their successes and limitations with a view to integrating them within a divisome for synthetic cells.

Protein language models, as well as models that incorporate structural information or homologous sequences, estimate the sequence likelihood p(sequence), which reflects the protein fitness landscape and is commonly used in mutation effect prediction and protein design. It is widely believed in deep learning field that larger models perform better across tasks. However, for fitness prediction, language model performance declines beyond a certain size, raising concerns about their scalability. Here we showed that model size, training dataset and stochastic elements can bias the predicted p(sequence) away from real fitness. Model performance on fitness prediction depends on how well p(sequence) matches evolutionary patterns in homologs, which is best achieved at a moderate p(sequence) level for most proteins. At extreme predicted wild-type sequence likelihoods, models predict uniformly low or high likelihoods for nearly all mutations, failing to reflect the real fitness landscape. Notably, larger models tend to predict proteins with higher p(sequence), which may exceed the moderate range and thus reduce performance. Our findings clarify the scaling behavior of protein models on fitness prediction and provide practical guidelines for their application and future development. Larger language models do not always perform better at fitness prediction. Hou et al. found that performance depends on whether model outputs match evolutionary patterns in homologs, which is best achieved at moderate predicted sequence likelihoods.

Antimicrobial resistance is an escalating global threat, with soils serving as reservoirs and conduits for the dissemination of antibiotic resistance genes (ARGs). Pesticide use in agriculture contributes to ARG proliferation, and ~60% of agricultural soils contain multiple pesticide residues. However, how pesticide diversity influences ARG dynamics in active microbial populations (active ARGs) remains unclear. Here, we evaluate the effects of pesticide diversity on active soil ARGs through a long-term field experiment integrating bioorthogonal non-canonical amino acid tagging (BONCAT), fluorescence-activated cell sorting (FACS), and metagenomics. We show that both low and high pesticide diversity significantly increase active ARG abundance relative to untreated control, whereas total ARG levels remain largely unchanged. The underlying mechanisms differ with pesticide diversity. At low diversity, active ARG co-selection via efflux pumps in Acinetobacter baumannii is a prominent mechanism. At high diversity, elevated reactive oxygen species and SOS responses promote horizontal gene transfer of active ARGs, as validated by culture experiments. These findings demonstrate that increasing pesticide diversity accelerates the emergence and dissemination of active ARGs, highlighting the need for integrated pesticide management strategies that consider both application intensity and diversity to mitigate resistance risks under the One Health framework. Here, the authors conduct a long-term field experiment to show that applying diverse pesticide mixtures increases active antibiotic resistance genes in soils. Low pesticide diversity co-selects for specific resistance genes, whereas high diversity accelerates spread between bacteria via horizontal gene transfer.

Targeted protein degradation (TPD) is an emerging therapeutic modality, but how multiple factors jointly determine degradation efficacy remains poorly understood. Here we present a quantitative framework for modeling the cellular mechanism of action of TPD, allowing for prediction of cellular efficacy with practically obtainable parameters. Applying this model to published data on 41 targets reveals a common range of degradation rate between 0.1 and 10 min−1, reflecting a characteristic efficiency of the TPD process. We define the degradability landscape, which provides a holistic characterization of the degradation propensity of a target of interest and serves as a roadmap for degrader optimization. We show how degradability is influenced by key factors such as target half-life and E3 level. We further quantify functional inhibition for kinase targets, enabling direct comparison between degrader and small molecule inhibitors. Finally, we uncover degrader discovery opportunities by systematically identifying targets with the potential to achieve superior degradation to facilitate TPD translational development. Targeted protein degradation is an emerging therapeutic modality, but degradation efficacy remains unpredictable. Here, the authors present a quantitative framework that predicts cellular degradation efficacy, and shows how it can guide target selection and degrader optimization.

Although probiotic-based bionic strategies show therapeutic promise for inflammatory bowel disease, their clinical translation is limited by poor gastric acid survival, inefficient intestinal colonization and inadequate targeting. Inspired by the multi-level cooperative mechanism of defense protection–danger sensing–tissue repair observed in coral communities, we developed a core–shell bionic microcapsule reactor (MY-E@SS). Here we show that the multifunctional bionic shell enables safe delivery of engineered bacteria through the gastrointestinal tract. Upon reaching inflamed intestinal sites, these bacteria sense the pathological microenvironment and responsively release an anti-inflammatory peptide. In a male murine model of inflammatory bowel disease, this system exhibited excellent biocompatibility and pronounced therapeutic efficacy, restoring intestinal barrier integrity, attenuating systemic inflammation and oxidative stress, modulating respiratory metabolism, and reestablishing microbial homeostasis. Mechanistically, therapeutic effects were attributed to inhibition of TNF-α/NF-κB signaling pathway. This work provides an intelligent platform to modulate inflammatory microenvironments and advance therapies for complex diseases. The authors developed a bionic microcapsule containing engineered bacteria that sense gut inflammation and release targeted therapy. Microcapsule treatment in mouse model of inflammatory bowel disease restored intestinal barrier function.

Large language models (LLMs) are deep learning-based artificial intelligence models that have achieved remarkable success in natural language processing. Typically composed of neural networks with billions of parameters, they are trained on massive unlabeled datasets using self-supervised or semi-supervised learning. Beyond language, LLMs hold immense potential for addressing complex bioinformatics challenges. This review provides a comprehensive overview of transformer-based model applications in genomics, transcriptomics, proteomics, drug discovery, and single-cell analysis. We discuss critical components, including tokenization strategies for diverse biological data, transformer architectures, attention mechanisms, and pretraining approaches. We also survey currently available foundation models and their downstream applications across bioinformatics domains. Finally, we highlight major challenges that remain insufficiently addressed in prior reviews and outline future perspectives and design principles for next-generation biological language models, offering practical guidance for both users and developers.