Cell systems最新文献

Advancements in artificial intelligence (AI) have transformed many scientific fields, with microbiology and microbiome research now experiencing significant breakthroughs through machine-learning applications. This review provides a comprehensive overview of AI-driven approaches tailored for microbiology and microbiome studies, emphasizing both technical advancements and biological insights. We first introduce foundational AI techniques and offer guidance on choosing between traditional machine-learning and sophisticated deep-learning methods based on specific research goals. The primary section on application scenarios spans diverse research areas from taxonomic profiling, functional annotation and prediction, microbe-X interactions, microbial ecology, metabolic modeling, precision nutrition, and clinical microbiology to prevention and therapeutics. Finally, we discuss challenges in this field and highlight some recent breakthroughs. Together, this review underscores AI's transformative role in microbiology and microbiome research, paving the way for innovative methodologies and applications that enhance our understanding of microbial life and its impact on our planet and our health.

The native microbiome influences numerous host processes, including neurological function. However, its impacts on diverse brain cell types remain poorly understood. Here, we performed single-nucleus RNA sequencing on the hippocampus of wild-type, germ-free mice, revealing the microbiome-dependent transcriptional landscape across all major neural cell types. We found conserved impacts on key adaptive immune and neurodegenerative transcriptional pathways. Mono-colonization with select indigenous microbes identified organism-specific effects on brain myeloid cell transcriptional state. Escherichia coli colonization induced a distinct myeloid cell activation state, increased brain-resident CD8⁺ T cells, and shaped amyloid phagocytic capacity, suggesting heightened disease susceptibility. Finally, E. coli-exposed 5xFAD mice displayed exacerbated cognitive decline and amyloid pathology, demonstrating the sufficiency of intestinal E. coli to worsen Alzheimer's disease-relevant outcomes. Together, these results emphasize the broad, species-specific, microbiome-dependent consequences on neural cell states and highlight the capacity of specific microbes to modulate disease susceptibility.

Synonymous codons are decoded at different speeds, but simple models predict that this should not drive protein output: translation initiation, not elongation, should limit the rate of protein production. We showed previously that the output of a series of synonymous fluorescent reporters in yeast spanned a 7-fold range corresponding to translation elongation speed. Here, we show that this effect is not due primarily to the established impact of slow elongation on mRNA stability. Rather, slow elongation further decreases the number of proteins made per mRNA. Our simulations, experiments on fluorescent reporters, and analysis of endogenous protein synthesis in yeast show that translation is limited on non-optimally encoded transcripts. Using a genome-wide CRISPRi screen, we find that impairing initiation attenuates the impact of slow elongation, showing a dynamic balance between rate-limiting steps of protein production. Our results show that codon choice can directly limit protein production across the full range of endogenous codon usage.

Scientific research has revealed only a minuscule fraction of the enzymes that evolution has generated to power life's essential chemical reactions-and an even tinier fraction of the vast universe of possible enzymes. Beyond the enzymes already annotated lie an astronomical number of biocatalysts that could enable sustainable chemical production, degrade toxic pollutants, and advance disease diagnosis and treatment. For the past few decades, directed evolution has been a powerful strategy for reshaping enzymes to access new chemical transformations: by harnessing nature's existing diversity as a starting point and taking inspiration from nature's most powerful design process, evolution, to modify enzymes incrementally. Recently, artificial intelligence (AI) methods have started revolutionizing how we understand and compose the language of life. In this perspective, we discuss a vision for AI-driven enzyme discovery to unveil a world of enzymes that transcends biological evolution and perhaps offers a route to genetically encoding almost any chemistry.

Classical phylogenetics assumes site independence, potentially overlooking epistasis. Protein language models capture dependencies in conserved structural and functional domains across the protein universe. Here, we ask whether MSA Transformer, which takes a multiple sequence alignment (MSA) as input, captures evolutionary distance and to what extent its representations reflect epistasis in protein sequence evolution, neither of which are explicitly available during training. Systematic shuffling of natural and simulated MSAs demonstrates that the model exploits column-wise conservation to distinguish phylogenetic relationships. Using internal embeddings, we reconstruct trees that are markedly consistent with those generated by maximum likelihood inference. Applying this approach to both the RNA-dependent RNA polymerase of RNA viruses and the nucleo-cytoplasmic large DNA virus domain, we recover both established and novel evolutionary relationships. We conclude that MSA Transformer complements, rather than replaces, classical inference for more accurate histories of protein families.

The maintenance of cellular homeostasis requires tight regulation of proteome concentration and composition. To achieve this, protein production and elimination must be robustly coordinated. However, the mechanistic basis of this coordination remains unclear. Here, we address this question using quantitative live-cell imaging, computational modeling, transcriptomics, and proteomics approaches. We found that protein decay rates systematically adapt to global alterations of protein synthesis rates. This adaptation is driven by a core passive mechanism supplemented by facultative changes in mechanistic/mammalian target of rapamycin (mTOR) signaling. Passive adaptation hinges on changes in the production rate of the machinery governing protein decay and allows for partial maintenance of the cellular proteome. Sustained changes in mTOR signaling provide an additional layer of adaptation unique to naive pluripotent stem cells, allowing for near-perfect maintenance of proteome composition. Our work unravels the mechanisms protecting the integrity of mammalian proteomes upon variations in protein synthesis rates. A record of this paper's transparent peer review process is included in the supplemental information.

Intrinsically disordered proteins or regions (IDPs or IDRs) adopt diverse binding modes with different partners, ranging from coupled folding and binding to fuzzy binding and fully disordered binding. Characterizing IDR interfaces is challenging both experimentally and computationally. State-of-the-art tools such as AlphaFold multimer and AlphaFold3 can be used to predict IDR binding sites, although they are less accurate at their benchmarked confidence cutoffs. Here, we developed Disobind, a deep-learning method that predicts inter-protein contact maps and interface residues for an IDR and its partner, given their sequences. It uses sequence embeddings from the ProtT5 protein language model. Disobind outperforms state-of-the-art interface predictors for IDRs. It also outperforms AlphaFold multimer and AlphaFold3 at multiple confidence cutoffs. Combining Disobind and AlphaFold-multimer predictions further improves performance. In contrast to current methods, Disobind considers the context of the binding partner and does not depend on structures and multiple sequence alignments. Its predictions can be used to localize IDRs in large assemblies and characterize IDR-mediated interactions.

Two recent studies in Cell Systems show why protein dynamics matter for prediction. By moving beyond static structures and embracing the dynamic "jigglings and wigglings" that Richard Feynman famously described, these approaches improve accuracy in binding site predictions for flexible systems despite challenges such as sparse training data. Together, they signal a shift toward models that try to capture the full energy landscape, paving the way for deeper insights into protein function.

The gut microbiome plays a critical role in human health, spurring extensive research using multi-omic technologies. Although these tools offer valuable insights, they often fall short in capturing the complexity of microbial interactions that associate with disease onset, progression, and treatment. Thus, integration of multi-omics datasets with metabolic models is needed to predict associations between microbial activity and disease. Here, we automated the reconstruction of 495 metabolic and gene expression models (ME-models), overcoming the main limitation preventing the wide use of this approach. We integrated them with multi-omics data from patients with inflammatory bowel disease (IBD), identifying taxa associated with variations in amino acids, short-chain fatty acids, and pH in the gut of IBD patients. In general, this approach provides testable hypotheses of the metabolic activity of the gut microbiota, and the automated pipeline opens the opportunity to study microbial interactions in other biologically relevant settings using ME-models.

Protein-protein interactions are fundamental to cellular processes, yet current deep learning approaches for binding site prediction rely on static structures, limiting their accuracy for disordered or flexible regions. We introduce dynamic geometric transformer (DynamicGT), a dynamic-aware model that integrates conformational dynamics into a cooperative graph neural network (Co-GNN) with a GT. Our model encodes dynamic features at both node (atom) and edge (interaction) levels, considering bound and unbound states to improve generalization. Dynamic regulation of messages passing between core and surface residues enhances detection of critical interactions for efficient information flow. Trained on a 1-ms molecular dynamics simulation dataset and augmented with AlphaFlow-generated conformations, the model was benchmarked extensively. Evaluation on diverse datasets containing disordered, transient, and unbound structures demonstrates that incorporating dynamics within a cooperative architecture significantly improves prediction accuracy where flexibility is key while requiring substantially less data than leading static approaches.