Biomolecular condensates are a major driver of cellular organization; however, we lack a predictable and systematic approach to modulate the multivalent interactions underlying their formation. Here, we demonstrate that the AI-driven FragFold method enables robust and generalizable design of protein fragments to control biomolecular condensate formation. We apply this approach across diverse proteins: G3BP1, SARS-CoV-2 nucleocapsid, TDP-43, and focal adhesion kinase (FAK). Computationally screening 2,235 fragments, we selected 18 candidates for further investigation. Overall, we attain a 50% success rate (9/18 designs) in discovering condensate-controlling protein fragments, experimentally testing just 3-5 candidates per protein. For each condensate-forming protein, the success rate is at least 40%. Furthermore, FragFold-predicted fragment binding modes align with their condensate-inhibitory or enhancing activities, revealing both known and newly identified interactions underlying condensate formation. In FAK, a condensate-inhibitory fragment uncovered a domain interaction required for phase separation, and mutational analysis validated its importance. Notably, this inhibitory fragment also suppresses FAK condensate formation in living mammalian cells. Together, these results establish AI-guided protein fragment discovery as a generalizable strategy to dissect and control the molecular interactions that govern biomolecular condensates.

Accurate selection of start codons by ribosomes is a fundamental determinant of proteome composition. Although the 'Kozak sequence'—an 8-nucleotide sequence flanking the start codon—has long been viewed as the primary determinant of initiation in eukaryotes, it fails to explain the large diversity of start codon usage across transcripts. Here we combine massively parallel reporter assays, bioinformatics, machine learning, single-molecule imaging and cryo-electron microscopy to define the 'extended translation initiation sequence (eTIS)', an ~80-nucleotide sequence surrounding the start codon that governs initiation efficiency. A deep-learning model trained on eTIS features accurately predicts translation initiation across transcripts. Unexpectedly, we find that the Kozak sequence is not optimal for initiation as is widely presumed, and we identify the origin of this discrepancy. eTIS nucleotides that promote efficient initiation are enriched in the human transcriptome and are evolutionarily conserved, underscoring their functional importance. Biophysical and structural analyses reveal that specific eTIS residues—including the key +6 position and residues in the mRNA entry and exit channel—engage ribosomal proteins, rRNA and initiation factors to promote start codon recognition by stabilizing the ribosome at the start codon and facilitating the structural transitions required for initiation. Finally, optimization of the eTIS markedly enhances translational fidelity and protein output from therapeutic mRNAs, highlighting its practical utility. Together, these findings redefine the sequence logic of translation initiation and establish a framework for precise control of protein expression.

Urinary tract infections (UTIs) are among the most common bacterial infections globally and create a large burden on the healthcare system. Uropathogenic Escherichia coli (UPEC) account for the majority of UTIs and increase the risk of recurrence. The standard treatment is antibiotics and, with the rise of multi-drug resistant UPEC lineages, there is a need for alternative treatments and prevention. Colicins, bacteriocins targeting and produced by E. coli, have previously been shown to inhibit the growth of pathogenic E. coli and are a promising alternative. Here, we engineer commensal Bacteroidaceae to secrete colicins via outer membrane vesicle (OMV) targeting signal peptides to suppress E. coli in the mouse gut. Secreted colicins were assessed for their ability to kill primary clinical isolate UPEC strains, including epidemic multi-drug resistant ST131 strains, along with other pathogenic and type strains. Specifically, secreted colicin E7, from Phocaeicola vulgatus fully eliminated of several UPEC strains in culture. In mice, P. vulgatus secreting colicin E7 prevented the extended colonization of two clinical UPEC strains and restored microbiome diversity. Together, this work shows the viability of secreted, heterologous antimicrobials from P. vulgatus as prophylactic treatment against the colonization of pathogenic E. coli utilizing cross-phylum antagonism in the gut.

Wastewater surveillance has emerged as a critical tool for global epidemiology, yet the functional diversity of wastewater microbiomes remains poorly characterized at the protein level. Here, we present WasteFams, the first comprehensive database dedicated to the systematic exploration of protein families in wastewater metagenomic and metatranscriptomic studies worldwide. Integrating data from 580 metagenomes, 132 metatranscriptomes, and 1,709 reference genomes, WasteFams catalogs 3,887 non-redundant protein families (containing ⪰100 members) derived from over 105 million predicted proteins. Each protein family is enriched with multi-layered annotations, including AlphaFold3 structural predictions, taxonomic classifications, and biome-specific metadata. To further expand their functional annotation, we integrated deep genomic context analysis to link protein families to Mobile Genetic Elements (MGEs), Biosynthetic Gene Clusters (BGCs), Antibiotic Resistance Genes (ARGs), and CRISPR elements. Accessible through the EnvoFams portal, WasteFams provides a user-friendly interface featuring advanced search capabilities, sequence and structural similarity tools, and interactive visualization modules. As global initiatives increasingly leverage wastewater for public health and environmental insights, WasteFams can serve as a critical resource for discovering novel microbial functions, monitoring resistance mechanisms, and exploring the biotechnological potential of secondary metabolites within wastewater-engineered ecosystems.

Rapid and accurate prediction of protein-ligand bindings is essential for drug discovery. While generative AI has driven rapid advancements in structure-based approaches, sequence-based methods remain significantly faster and more cost-effective. Here, we present a weakly supervised deep learning framework integrating graph convolutional networks (GCN) for molecular encoding and bidirectional long short-term memory (BiLSTM) for protein modeling. The latter represents long-range dependencies better than the widely used convolutional neural network (CNN). Leveraging a bilinear attention network (BAN), this model learns protein-ligand pairwise interactions without requiring three-dimensional structural supervision. By using the publicly available BindingDB dataset, the model was trained, solely on affinity labels, and successfully classified binder and non-binders with AUROC of 0.96 and an AUPRC of 0.95. The model generates interpretable attention maps that serve as a "GPS" to locate binding sites. Remarkably, despite the lack of structural training data, it can pinpoint key contact residues confirmed by crystal structures. Our method could function as a scalable filter for giga-scale libraries, allowing rapid screening of drug candidates with direct structural insights into the protein-ligand interface.

Nitrogen is essential for all life forms, and microorganisms prefer ammonium as a nitrogen source. Due to the low affinity of glutamine synthetase (GS) for ammonium, E. coli must maintain high intracellular ammonium (NH4+) concentrations to sustain its rapid growth. Under ammonium limitation, E. coli imports ammonium through the transporter AmtB and incorporates it into glutamine by using GS. On the basis of structural and mutagenesis information, mechanisms have been proposed for the transport of ammonia (NH3) and protons by AmtB through spatially (partly) separate routes. These mechanisms do not explain the required coupling between proton and ammonia transports. How does the membrane potential push the ammonia inward so as to attain high concentrations near GS? We here compare six candidate kinetic models of E. coli ammonium transport and assimilation in terms of how they reproduce experimental data from the literature: three variants of the 'electro-binding model' in which the membrane potential affects AmtB-NH4+ binding, and three variants of the 'electro-flipping model' in which it influences the conformational flip of the transporter. The computer simulations decide that the electro-binding models are 28 times more plausible than the electro-flipping models and suggest that the transmembrane electric potential affects AmtB-NH4+ binding from the cytoplasmic side. The addition of kinetic and thermodynamic features to existing structural information plus our requirement of an explanation of the coupling, suggest a new spatiotemporal mechanism of coupling of ammonia and proton flows in AmtB. Further simulations show that GS and AmtB regulation is coordinated via both the uridylyltransferase/uridylyl-removing enzyme (UTase) and 2-oxoglutarate binding, allowing the cell to minimize futile cycling while maintaining rapid growth. The free energy cost of transport-related futile cycling exceeded that of the GS reaction itself. Moreover, AmtB enabled robust growth under varying ammonium concentrations and pH levels, albeit at a cost of futile cycling that became substantial at low ammonium. These findings highlight the crucial roles of GS and AmtB in E. coli's adaptations and provide new insights into the trade-off mechanism between nutrient acquisition and energy efficiency.

Methylotrophic yeasts such as Pichia pastoris are widely used for heterologous protein production because they contain strong and tightly regulated promoters. However, the use of methanol as an inducer presents several practical challenges, including toxicity, flammability, high oxygen demand during fermentation, and increased production costs. To overcome these limitations, researchers have been working on redesigning the AOX1 regulatory system and developing alternative induction strategies that do not rely on methanol. One promising approach is optogenetics, which uses light to control gene expression in a non-invasive way. These systems rely on light-sensitive proteins such as phytochromes, cryptochromes, LOV-domain proteins, and UVR8, allowing gene activity to be regulated in a precise and reversible manner without adding chemical inducers to the culture medium. This review brings together key advances in yeast optogenetics, with a focus on the EL222 system, highlighting its implementation for light-controlled heterologous protein production in P. pastoris and its broad application in synthetic biology and metabolic engineering in Saccharomyces cerevisiae. The growing versatility and scalability of EL222-based circuits highlight their potential to reshape both fundamental research and industrial bioprocessing through safer, more controllable, and energy-efficient gene regulation strategies.

Generative models are increasingly used for protein design, but the lack of standardized evaluation frameworks limits comparison across model classes and hinders translation to experimental success. Here, we introduce a unified sampling and benchmarking framework that enables controlled sequence generation across alignment, protein language, and structure-based models, and apply it to Tobacco etch virus (TEV) protease. Across hundreds of thousands of designed sequences, different models explore distinct regions of sequence space with no clear computational selection metrics to assess enzymatic function. Experimental evaluation reveals large differences in functional outcomes, ranging from non-functional variants to sequences with ~9-fold higher activity than wildtype. Machine learning-designed libraries achieve a 39.32% hit rate (percentage of variants matching or exceeding wildtype activity) compared to 6.06% for an error-prone PCR baseline. Structure-based models perform best overall, with hit rates of 74.4% and 66.8% for ESM-IF1 and ProteinMPNN, respectively. Commonly used selection metrics do not strongly correlate with experimental activity, highlighting a gap between in silico evaluation and enzyme function. Together, these results establish a generalizable framework for benchmarking generative protein models and demonstrate the necessity of experimental validation for guiding model development and sequence prioritization.

Wastewater surveillance has emerged as a critical tool for global epidemiology, yet the functional diversity of wastewater microbiomes remains poorly characterized at the protein level. Here, we present WasteFams, the first comprehensive database dedicated to the systematic exploration of protein families in wastewater metagenomic and metatranscriptomic studies worldwide. Integrating data from 580 metagenomes, 132 metatranscriptomes, and 1,709 reference genomes, WasteFams catalogs 3,887 non-redundant protein families (containing ⪰100 members) derived from over 105 million predicted proteins. Each protein family is enriched with multi-layered annotations, including AlphaFold3 structural predictions, taxonomic classifications, and biome-specific metadata. To further expand their functional annotation, we integrated deep genomic context analysis to link protein families to Mobile Genetic Elements (MGEs), Biosynthetic Gene Clusters (BGCs), Antibiotic Resistance Genes (ARGs), and CRISPR elements. Accessible through the EnvoFams portal, WasteFams provides a user-friendly interface featuring advanced search capabilities, sequence and structural similarity tools, and interactive visualization modules. As global initiatives increasingly leverage wastewater for public health and environmental insights, WasteFams can serve as a critical resource for discovering novel microbial functions, monitoring resistance mechanisms, and exploring the biotechnological potential of secondary metabolites within wastewater-engineered ecosystems.