Understanding how opposing regulatory factors shape gene expression is essential for interpreting complex biological systems. A motivating observation, drawn from cancer epigenetics, is that removing an activating factor can sometimes lead to higher, not lower, expression of a gene that is also subject to repression. This counterintuitive behavior suggests that competition between activators and repressors for limited genomic binding sites may produce unexpected transcriptional outcomes. Prior theoretical work proposed this mechanism, but it has been difficult to test directly in natural systems, where layers of chromatin regulation obscure causal relationships. This paper introduces a fully synthetic, tunable genetic platform in a prokaryotic model system that isolates this competition mechanism in a clean and interpretable setting. The engineered construct contains a target gene with binding sites for both an activator and a repressor, together with a separate decoy region that carries overlapping binding sites for the same regulators. Activator and repressor functions are implemented using CRISPRa and CRISPRi, which permit independent control of regulator expression levels and binding affinities. Using this minimal system, the paper shows that increasing activator expression can reduce expression of the target gene when both regulators are present, consistent with the prediction that additional activator molecules displace the repressor from decoy sites and allow it to more effectively repress the target. By demonstrating how competition alone can invert expected regulatory responses, this synthetic framework provides a validated model for understanding similar paradoxical behaviors in natural regulatory networks and establishes a foundation for future studies in more complex mammalian contexts.

Protein-protein interactions are central for understanding biological processes. The ability to predict interaction partners is extremely valuable for avoiding costly, time-consuming experiments. It has been shown that AlphaFold has an unsurpassed ability to accurately evaluate interacting protein pairs. However, a protein can also form homomeric interactions, i.e. interact with itself. We found that AlphaFold yielded a significantly higher false-positive rate for identifying homodimers than for heterodimers. True Positive Rate (TPR) at 1% False Positive Rate (FPR) drops from 63% for heterodimers to 18% for homodimers. When we investigated the high-scoring false positives, i.e., non-homodimers with high AlphaFold scores when predicted as such, we found that their homologs were enriched for homomultimeric proteins. Using a simple logistic regression model that combines AlphaFold scores with structural and homology information, we increased the TPR (at 1% FPR) to 42 +/- 8% (5-fold cross-validation) from 19%. If we excluded the homology information, we achieved a TPR of 28 +/- 7%, which is still better than using AlphaFold metrics. Availability and implementation: All data are available from Zenodo DOI:\10.5281/zenodo.17738668 and all code from https://github.com/SarahND97/alphafold-homodimers

Intraspecies interactions shapes microbial community structure and evolution, yet the mechanisms determining competitive outcomes among closely related strains remain unclear. The soil bacterium Bacillus subtilis is a model for microbial social interactions, where quorum-sensing systems regulate cooperation and antagonism. Here, we take a multifaceted approach to dissect the role of quorum-sensing regulation in competitive fitness. Isolate NCIB 3610 carries a signal unresponsive RapP-PhrP module that alters quorum-sensing control and promotes faster growth. Modelling and mutant analysis demonstrate that the small differences in growth rate conferred by RapP-PhrP3610 are sufficient to drive competitive exclusion. The importance of quorum sensing control is further exemplified by experimental evolution of distinct wild isolates, which revealed recurrent mutations in the sensor kinase comP, which phenocopy complete comP or comA deletions and confer a growth-linked competitive advantage. Key quorum sensing mechanisms are abandoned even in structured microbial communities, where it might be expected that communal traits are favored. Furthermore, a phylogenomic survey of 370 B. subtilis genomes identified disruptive comP mutations in ~16% of isolates. However, growth rate alone does not explain all interaction outcomes as even isogenic strains with equivalent doubling times differ in competitiveness. Transcriptomic profiling and validation experiments implicated a type VII secretion system toxin as an additional effector. These findings reveal that disruption of quorum-sensing pathways, whether naturally or through selection, provides a rapid route to competitive advantage, highlighting a fundamental trade-off between communal signalling and individual fitness in microbial populations.

Large language models (LLMs) are increasingly used to extract knowledge from text, yet their coverage and reliability in biology remain unclear. Microbial phenotypes are especially important to assess, as comprehensive data remain sparse except for well-studied organisms and they underpin our understanding of microbial characteristics, functional roles, and applications. Here, we systematically assessed the biological knowledge encoded in publicly available LLMs for structured phenotype annotation of microbial species. We evaluated the performance of over 50 LLMs, including state-of-the-art models such as Claude Sonnet 4 and the GPT-5 family of models. Across phenotypes, LLMs reached accurate assignments for many species, but performance varied widely by model and trait, and no single model dominated. Model self-reported confidence is informative, with higher confidence aligning with higher accuracy, and can be used to prioritize phenotype assignment, effectively distinguishing between high- and low-confidence inferences. Overall, our study outlines the utility and limitations of text-based LLMs for phenotype characterization in microbiology.

Autonomous bioluminescence systems─genetically encoded platforms that integrate luciferase enzymes with complete substrate biosynthetic pathways─have emerged as transformative tools for real-time, noninvasive imaging in living systems. Unlike conventional substrate-dependent bioluminescence, these systems provide continuous light emission without external substrates, enabling long-term monitoring with minimal phototoxicity compared to the use of fluorescence. Here, we present a critical perspective on recent advances in the two best-characterized autonomous systems: bacterial and fungal bioluminescence systems. We assess their molecular mechanisms, protein engineering strategies, and emerging applications in single-cell imaging, multicolor biosensing, and whole-organism monitoring. By comparing their strengths and limitations, we highlight persistent challenges, such as low quantum yield in bacterial bioluminescence and substrate availability constraints in fungal bioluminescence, and discuss strategies to address them─including AI-guided mutagenesis, de novo protein design, and metabolic pathway optimization. We conclude by outlining application-driven design targets for the next generation of autonomous bioluminescent systems in biomedical research, environmental monitoring, and synthetic biology.

Polyethylene Terephthalate (PET) is an extensively used plastic whose durability and resistance to degradation contribute to growing environmental pollution and concerns. Enzymatic PET degradation, particularly via PETase from Ideonella sakaiensis, has emerged as a sustainable approach due to its ability to depolymerize PET under mild conditions. While research has largely focused on enhancing the enzyme′s thermal stability through distal mutations, less attention has been given to active-site engineering aimed at directly improving catalytic efficiency. Here, we used an automated in silico protein engineering platform called Gene Discovery and Enzyme Engineering (GDEE), designed to systematically explore mutations at the active site. By leveraging FastPETase (FP) as scaffold, we perform a high throughput generation of thousands of variants, evaluated them via docking studies with a PET substrate analogue, and ranked candidates based on binding affinity and catalytic geometry. We identified S238Y as a key mutation that enhanced PET film degrading performance at 40 °C when inserted in two of the most active PETase variants reported to date: 2.2-fold increase in the FP scaffold and 3.4-increase in the ThermoStable-PETase (TSP) background. Compared to wild type PETase, FP S238Y showed a 14.8-fold increase in bulk activity, translating into 9.4-fold more TPA and 20-fold more MHET by UPLC, while TSP S238Y reached a 25.8-fold increase (14.4-fold more TPA and 42.6-fold more MHET). This mutation also enhanced catalytic efficiency and resistance to enzyme concentration inhibition, especially in the TSP scaffold. Molecular dynamics confirm position 238 as a relevant modulator of ligand stabilisation. These findings underscore the potential of targeted active-site engineering, combined with structure-guided prediction, to accelerate the development of efficient mesophilic biocatalysts for plastic waste remediation.

Accessibility of the ribosome binding site (RBS) plays an outsized role in bacterial mRNA decay and translation. Antagonistic mRNA sequences that reduce accessibility and regulate expression have been widely documented near the RBS. To determine whether such sequences are also the primary effectors of expression when placed far from the RBS, we measured impacts of all possible 8-nucleotide substitutions (65,536 variants) at different positions in mRNA in Bacillus subtilis. While the vast majority of substitutions negligibly affect RNA levels, pyrimidine-rich substitutions resembling the anti-Shine-Dalgarno (aSD) sequence exhibit strong inhibitory effects. Even several hundred nucleotides downstream of the RBS, these aSD-like sequences base-pair with the RBS, promote RNA decay, and inhibit translation initiation. We find aSD-like sequences to be depleted throughout endogenous genes, likely due to selective pressure for expression. Taken together, our findings reveal widespread long-range RNA intramolecular interactions in vivo and uncover a key constraint on gene sequence evolution.

Detecting structural similarity at the local level between proteins is central to understanding function and evolution, yet most approaches require 3D models. In this work, we show that protein language models (pLMs), solely using sequence data as input, implicitly capture fine-grained structural signals that can be leveraged to identify such similarities. By mean-pooling residue embeddings over sliding windows and comparing them across proteins with cosine similarity, we find diagonal patterns that reflect locally aligned regions even without sequence identity. Building on this insight, we introduce a framework for detecting locally aligned structural regions directly from sequences, supporting the development of scalable methods for structural annotation and comparison.

Codon optimization involves selecting synonymous codons to match host-specific preferences. It is critical for heterologous expression but remains challenging due to the combinatorial design space. Under long-term evolutionary selection, natural coding sequences are near-optimal compromises between translational efficiency, accuracy, and regulatory constraints, providing a de facto standard for data-driven models. Recent deep learning-based language models therefore aim to learn the distribution of natural codon sequences and reuse it for design. However, existing approaches discard the rich semantic structure of taxonomic lineages, underutilize protein functional and evolutionary constraints, and often rely on masked-language objectives that lack a principled mechanism for sequence generation. Here we present CodonTranslator, a 150M-parameter decoder-only Transformer trained on 62 million CDS-protein pairs from over 2,100 species. CodonTranslator uses a pretrained language model to embed hierarchical species lineages and a pretrained protein language model to encode protein context, enabling interpolation across hosts and generalization to unseen species and proteins. Our results show that CodonTranslator implicitly learns the genetic code from data, faithfully reproduces species-specific codon usage, and designs coding sequences that match or surpass existing methods in both codon usage metrics and predicted biological stability.  https://github.com/poseidonchan/CodonTranslator.