The biases revealed in protein sequence alignments have been shown to provide information related to protein structure, stability and function. Recent studies using Potts models show that single-site biases and pair correlations can improve predictions of protein fitness, activity and stability compared to simpler models. Here we use a Potts model to design groups of protein sequences with different amounts of single-site biases and pair correlations and determine the stabilities of sequences from each group. Surprisingly, sequences excluding pair correlations maximize stability compared to sequences that maximize pair correlations, suggesting that pair correlations contribute to other aspects of protein fitness. Consistent with this interpretation, we find that for three enzyme families, activity is greatly increased by maximizing pair correlations. The finding that elimination of covariant residue pairs increases protein stability suggests a route to enhance stability of designed proteins, although this stability may be offset by reduced enzyme activity. Sequence biases at individual positions can be used to design more stable protein variants, while correlation between pairs of residues has also been shown to be important for specifying protein structure and function. Here, a Potts model is used to separate the contributions of single-site biases and pair correlations, revealing that protein structure is stabilized by single-site biases rather than pairwise coupling, while pairwise coupling is important for protein function.

Unexploded ordnance from World Wars I and II continues to release 2,4,6-trinitrotoluene into marine sediments, yet microbial responses to this chronic contamination remain poorly understood. Here, we characterize the taxonomic and functional potential of sediment microbiomes at the historical submarine wreck UC-30 in the North Sea, combining 16S rRNA amplicon sequencing, shotgun metagenomics, and targeted GC-MS/MS analysis with a parallel aerobic laboratory enrichment. Minewell sediments showed distinct community shifts, with enrichment of Proteobacteria, notably Haliaceae and Rhodobacteraceae, alongside increased representation of oxidoreductases and stress-related enzyme classes, including glutathione S-transferases. Genes associated with TNT transformation, including Old Yellow Enzymes and nitroreductases, were modestly enriched in situ. The laboratory enrichment confirmed TNT removal and presence of N-ethylmaleimide reductase, an Old Yellow Enzyme implicated in TNT transformation. Functional and taxonomic parallels between field and enrichment communities indicate shared adaptive capacities under TNT exposure, positioning contaminated marine microbiomes as reservoirs of bioremediation potential. Long-term TNT exposure at historical shipwrecks can influence sediment microbial communities, according to combined wreck sampling from the North Sea and laboratory incubation tests.

Accurate enzyme turnover numbers are essential for building enzyme-constrained genome-scale metabolic models. However, collecting and curating these parameters remains a major bottleneck. Indeed,  values are scattered across multiple databases, reported under varying experimental conditions, and often missing for many enzymes. To address this challenge, we present WILDkCAT, a Python-based pipeline that enables the retrieval of  values from wild-type enzyme measured under user-specified pH and temperature ranges for a given metabolic model. The application to E. coli (iML1515) and Homo sapiens (Human-GEM) models demonstrated the ability of WILDkCAT to retrieve substantial  coverage and its applicability across diverse genome-scale models.

Microbial communities play fundamental roles in industrial processes and ecosystem stability. However, understanding how individual members and their interactions give rise to community-level function remains challenging because such functions emerge from complex interactions among diverse members. Here, we developed SubCom analysis, a subcommunity-based experimental–computational workflow for inferring candidate taxon-specific contributions and interaction contexts underlying microbial community function. Using an aniline-degrading microbial community, we generated paired composition–function data from 558 randomly assembled, low-complexity subcommunities constructed using a dilution-and-dispense strategy. We then trained decision-tree-based models to predict community function from composition, achieving high predictive performance (r = 0.77–0.89). Interpretation of the learned decision rules identified taxa with consistent functional association: specific Pseudomonas and Acinetobacter taxa were associated with increased community-level aniline utilization, whereas an Achromobacter taxon exhibited a negative association despite its presumed role in downstream metabolism. The models further suggested potential functional interactions, including attenuation of the positive contributions of Pseudomonas and Acinetobacter in the presence of a Corynebacterium taxon, highlighting functional relationships that are not readily inferred from genome-based approaches alone. An augmentation assay using representative isolates supported the predicted direction of several effects and enabled targeted improvement of community function. These results demonstrate the potential of SubCom analysis as a practical framework for inferring taxon-specific contributions and interaction contexts in complex, nonsynthetic microbial communities. Video Abstract

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.

The transition toward a sustainable bioeconomy necessitates the development of robust microbial platforms capable of C1 chemical valorization, particularly the conversion of methanol into high-value chemicals. Komagataella phaffii is a premier methylotrophic host for such applications due to its natural methanol utilization pathways and high-density fermentation capabilities. However, its potential as a C1-based synthetic biology chassis has been hampered by inefficient homologous recombination and the technical complexity of multigene pathway integration. To accelerate the design-build-test-learn (DBTL) cycle for methanol-based biomanufacturing, a more sophisticated and high-throughput genome engineering toolkit is required. We report the development of an HR-enhanced K. phaffii chassis (dKR) achieved by synergistically deleting Ku70 and overexpressing RAD52 at the chromosomal level. This engineering markedly improved the robustness and reproducibility of CRISPR/Cas9-mediated genome editing, increasing integration efficiency from 15.3% in the wild-type to 68.1% in the dKR chassis when using short 50-bp homology arms. The platform demonstrated exceptional capacity for vector-free multi-fragment assembly, achieving an integration efficiency of 86.5% for nine linear DNA fragments. Furthermore, the implementation of a tRNA-ribozyme CRISPR system enabled simultaneous multi-locus integration at three distinct genomic sites with 44.6% efficiency. Notably, the chassis supported direct in vivo plasmid assembly of ten fragments (9 fragments + 1 backbone) with 44.8% efficiency, effectively bypassing E. coli-based cloning. Functional validation via a carotenoid biosynthetic pathway resulted in β-carotene production while maintaining normal cellular growth and metabolic productivity. This study established a versatile genome-engineering platform that advances K. phaffii from a protein-expression host to a broadly applicable biological engineering chassis. By enabling streamlined cellular design and reproducible multigene integration, the platform supports applications spanning synthetic biology, metabolic engineering, and bioproduction. More broadly, the approach illustrates how rational manipulation of DNA repair pathways can enhance genome engineering performance in non-conventional hosts, providing transferable design principles relevant to biological engineering across multiple domains.

Plasmid host range (PHR) plays a key role in the spread of ecologically important genes, alongside applications in microbiome engineering and environmental biotechnology. PHR is a complex trait arising from the combination of plasmid, donor and recipient properties. Most studies of PHR use a single donor strain, leaving the role of the donor unexplored and often require genetically tagged recipient strains for counter-selection, which limits the use of non-genetically tractable strains. Here, we applied auxotrophic donor counter-selection in a relatively high-throughput and accessible screening format to characterize PHR across a diverse collection of environmental isolates without the need for recipient engineering. Specifically, we used two auxotrophic donors (Pseudomonas fluorescens and Pseudomonas putida) and plasmid pQBR57-tphKAB, an environmental plasmid engineered for terephthalic acid bioremediation. We screened a library of 101 soil isolates as potential recipients, including genera such as Pseudomonas, Bacillus and Xanthomonas. We only observed conjugation into other Pseudomonas, but donor identity was found to affect PHR, with P. fluorescens conjugating the plasmid into more recipient strains than P. putida. Phylogenomic analysis revealed that transconjugants clustered primarily with the Pseudomonas citronellolis lineage, previously isolated from soil. In strains that were close relatives of transconjugants but unable to acquire the plasmid, we observed five defence systems not present in transconjugants that may act as barriers to plasmid acquisition. Our approach demonstrates how auxotrophic donor counter-selection can be deployed at scale to screen PHR in environmental isolates and to investigate the influence of donor identity on plasmid conjugation.

Adenine base editors (ABEs), which enable A•T-to-G•C base editing, have emerged as a powerful tool with potential therapeutic applications. However, conventional ABEs suffer from bystander nucleotide conversions, limiting their utility for precise editing. Here we present a single-nucleotide resolution ABE (snuABE) created by fusing a nickase Cas9, nCas9-H840A, with the deaminase domain of ADAR (adenosine deaminase acting on RNA), which acts on DNA:RNA hybrids, instead of TadA, which acts on single-stranded DNA in conventional ABEs. snuABE requires a target-adenine guide RNA (tagRNA) that introduces a mismatch at the target adenine, enabling highly specific A-to-G editing by ADAR. Engineering ADAR from Pediculus humanus using the in silico protein evolution algorithm EvolvePro, along with 3′-end protection of the tagRNA, enhanced snuABE activity, yielding a median efficiency of 5.4% and a maximum efficiency of 50.0% across 32 targets in HEK293T cells. snuABE exhibits no detectable DNA off-target editing at predicted off-target or orthogonal R-loop sites, highlighting its potential as a precise and safe base-editing technology. Bystander edits are minimized by engineering an adenosine deaminase acting on RNA for DNA base editing.

Bacterial plant pathogens have ravaged crops since the dawn of agriculture and continue to pose a serious threat today. Bacteria and their plant hosts have co-evolved in an evolutionary arms race, with artificial selection due to agriculture tipping the scale in favor of the pathogen. This review gives an overview of plant pathogenic bacterial diversity, showing that pathogenicity has independently evolved numerous times, and that there is not one unifying trait determining plant pathogenicity. Instead, these bacteria represent repeated, independent evolutionary transitions driven by life in complex ecological networks, that include plant hosts, insect vectors, microbial competitors, and highly heterogenous abiotic environments. Their genomes reflect this interplay through a dynamic balance of architecture and flux. These structural features, along with highly variable pangenomes, capture the balance between genome stability and flux imposed by ecological constraints and epidemiological dynamics. Horizontal gene transfer via conjugative plasmids, prophages, integrative and conjugative elements, transposons, and in some lineages, natural competence, remains the major source of adaptive novelty, enabling rapid remodeling of virulence repertoires, metabolic capabilities, and antibiotic or heavy metal resistance genes. These changes create distinct selective landscapes. Agricultural practices such as chemical use, host resistance deployment, or seed trade, can drive recurrent bottlenecks, expansions, and admixture events that leave strong genomic signatures in pathogens. Finally, this review explores the genomic differences enabling the divergence of lifestyles, while also acknowledging knowledge gaps and future directions of research on the evolution of bacterial plant pathogens.

Virus-like particles (VLPs) are promising for delivering genome editors, yet the in vivo in vivo efficacy of VLP-mediated cytosine base editing remains limited. Here we identified insufficient inhibition of uracil DNA glycosylases as the underlying mechanism of low cytosine base editor (CBE) editing efficiencies in vivo. We engineered a previously reported CBE, transformer base editor (tBE), and developed a VLP delivery system to enhance the recruitment of uracil DNA glycosylase inhibitor proteins. tBE-VLPs achieved robust C-to-T editing in mouse liver and retina. A single injection achieved, on average, 46.0% editing at mPcsk9 and 64.2% at mHpd in the liver, as well as 24.2% at mVegfa in the retinal pigment epithelium, resulting in marked therapeutic benefits in mouse disease models. tBE-VLP4 induced no detectable off-target edits in vitro or in vivo and demonstrated superior specificity compared to AAV or lipid nanoparticle mRNA delivery. Our work establishes tBE-VLP4 as a precise, efficient system for in vivo cytosine base editing. In vivo cytosine base editing is made efficient with potent glycosylase inhibition.

Bacteria restrict viral replication not only through dedicated defense systems but also by entering physiological states that limit cellular resources, yet how phages overcome such host-imposed barriers remains unclear. The stringent response, driven by the alarmone nucleotides ppGpp and pppGpp, can impose a growth-restrictive state that hinders phage infection in specific phage–host contexts. Here, we show that alarmone signaling constrains bacteriophage T7 infection and that the portal protein Gp8 counteracts this barrier by engaging RelA and SpoT, inhibiting their synthetase activities and suppressing alarmone accumulation. Portal mutations that disrupt this interaction sustain alarmone elevation, delay lysis and impair replication in a manner relieved in alarmone-deficient hosts. Portal proteins from representative coliphages share related stringent-response-linked features, indicating that essential virion components can moonlight as antagonists of host stress physiology and that this mechanism is not unique to T7 and may extend to additional coliphages. Bacteria can protect themselves against viral infections by entering physiological states, such as the stringent response, that limit cellular resources. Here, the authors show that phages can use their portal proteins to suppress the bacterial stringent response, thus promoting infection.