Generalist biological artificial intelligence (GBAI) represents a transformative approach to modeling the ‘language of life’—the flow of information from DNA to cellular function. This Review synthesizes rapid advances in biological AI to interpret and generate DNA, RNA, proteins and cellular systems. We chart a course toward comprehensive systems that can concurrently process and predict across these domains, performing several critical biological tasks simultaneously. Substantial opportunities lie in synergizing language and structural AI, leveraging specialized models and improving AI agents for autonomous discovery. After addressing challenges in data, biological complexity, scalability and experimental validation, GBAI has the potential to deepen our understanding of disease pathways and biomarkers, advance automated therapeutic design and evaluation, and integrate within virtual cells to meaningfully simulate biological activity. This Review discusses the promises and pitfalls of biological AI algorithms and presents a vision for generalist biological artificial intelligence, in which models can perform diverse tasks across biological domains.

Viral sequences in diverse environments remain largely uncharacterized, impeding our comprehension of their genetic makeup, biological interactions, and potential applications. This underscores an urgent need for innovative analytical methods. Here, we present the VirHost Hunter framework, which employs phage tails and lysins, bypassing the requirement for full genomes, for efficient and high-resolution host assignment. By harnessing Protein Language Models and Vision Transformers, VirHost Hunter captures protein functional homology despite sequence dissimilarity, significantly boosting prediction accuracy. In the scenario of disease-associated gut bacteria, the calibrated VirHost Hunter surpasses existing methods, doubling phage host assignments, expanding taxonomic reach, and revealing previously uncharacterized phages targeting gut bacteria, including Akkermansia and Prevotella. Therefore, we establish a gut phage lysin database, enabling the synthesis of a lysin that effectively and specifically targets an obesity-promoting bacterium. VirHost Hunter’s precision and scalability mark a significant leap forward in virome research and present a promising avenue for microbiome therapies. Here, the authors present VirHost Hunter, an AI-based approach to phage–host assignment using tail and lysin proteins, showing it improves host resolution, expands functional discovery of gut phage, and enables targeted lysin identification for microbiome research.

Although sequence-discrete species appear to dominate microbial communities, readily distinguishing between distinct populations of a species recovered from different short-read metagenomic samples is challenging due to technical limitations associated with read length. To close this gap, we developed a novel algorithm to evaluate which reads in a metagenome belong to a target population based on the distribution of sequence identities of reads aligned to a reference sequence, which are filtered using a Kernel density estimation (KDE) as a flexible alternative to the commonly used static 95% nucleotide identity cutoff. Subsequently, we employed the average nucleotide identity of reads (ANIr) aligning above the KDE threshold, and resampling techniques for estimating the confidence intervals of ANIr values, to quantify intra-population sequence diversity and compare populations across globally representative marine samples. Most populations showed high ANIr in only a few samples at similar depths, and decreased ANIr and increased gene-content difference between samples where a closely related population is detected [e.g., same 95% ANI-based species]. Accordingly, ANIr correlated with the physical distance between the samples, and only a few truly cosmopolitan populations were identified. Among the latter, Alteromonas macleodii (97% average amino-acid identity -or AAI- to the type genome) and Prochlorococcus marinus (79% AAI) showed high relative abundance in both surface (0-200m) and deep (>1000m) samples. These results suggest that microbial communities under different environmental conditions share very few identical and abundant populations, and provide a highly needed methodology to track such populations over space and time, in marine or other habitats.

Group I introns are catalytic RNAs capable of self-splicing, yet structural insights into full-length precursor RNAs and post-splicing circularization have been limited. Here, we present cryo-EM structures of the Azoarcus pre-tRNAIle system across key catalytic stages, including the full-length precursor in Pre-1S conformations, the linear intron, and the circular introns, at 2.8-3.3 Å resolution. These structures reveal a preformed P1 helix, register shifts during splicing, and a conformational flip of G37 that stabilizes the circularization site. Biochemical analyses confirm a two-step circularization mechanism, generating a secondary circular product via an alternative ligation site. Together, our results provide an atomic-level view of a group I intron system through splicing and circularization. This work uncovers structural principles governing RNA conformational dynamics, catalysis, and circular RNA formation, with broad implications for ribozyme engineering. Cryo-EM structures of the Azoarcus group I intron capture its progression from pre-splicing to circular RNA products, revealing coordinated RNA rearrangements that drive catalysis and a two-step circularization pathway.

Peptide aggregation is a long-standing challenge in chemical peptide synthesis, limiting its efficiency and reliability. Although data-driven methods have enhanced our understanding of many sequence-based phenomena, no comprehensive approach addresses so-called non-random difficult couplings (generally linked to aggregation) during solid-phase peptide synthesis. Here we leverage existing peptide synthesis datasets, supplemented with further experimental data, to build a predictive model that deciphers the role of individual amino acids in triggering aggregation. We first identified and experimentally validated composition-dependent aggregation as a stronger predictor than sequence-based patterns. This insight enabled the development of a composition vector representation, allowing insights into the aggregation propensities of individual amino acids. Applying an ensemble of trained models, we predicted the aggregation properties of peptides and recommended the optimized use of aggregation-reducing tools. By elucidating each individual amino acid’s influence, this method holds the potential to accelerate synthesis optimization through existing data, offering a robust framework for understanding and controlling peptide aggregation. Aggregation during solid-phase peptide synthesis is a bottleneck, often rendering peptide synthesis costly and inefficient. Now the impacts of individual amino acids on aggregation have been characterized through machine learning and analysis of composition patterns. The generated models predict problematic peptides and optimize synthesis conditions, enabling chemists to use aggregation-avoiding routes.

Soils are heterogeneous and dynamic systems characterized by complex physical, chemical, and biological interactions. Understanding these interactions is critical, as they influence plant productivity, global biogeochemical cycles, and ecosystem resilience. While ecologists have long studied soils in field, greenhouse, and laboratory settings, their complexity and heterogeneity make it challenging to pinpoint key properties driving biological processes and derive mechanistic insights. Advancements in synthetic biology, which seeks to engineer and control biological processes in soils, have increased the demand for standardized and controllable experimental platforms. These platforms, referred to here as ‘synthetic soils’, are systems designed to reproduce selected physicochemical characteristics of natural soils in a simplified and defined format, allowing scientists to systematically change soil physicochemical properties (i.e. texture, mineralogy, pH) to study how biological components (i.e. microbes, plants, soil fauna, etc.) respond to, modify, or interact within these controlled environments. This review explores existing synthetic soils, their advantages, limitations, and applications in ecology and synthetic biology, and discusses potential directions for their future development.

RNAs exhibit complex dynamics in cells, including expression, splicing, localization, translation and degradation, and these processes are highly coordinated and tightly regulated both spatially and temporally. To better understand the biological function of diverse RNAs, approaches that allow monitoring of RNA with high spatiotemporal resolution are essential. Fluorescent RNAs (FRs), fluorescent protein–like entities consisting of RNA aptamers and their cognate fluorogenic dyes, have emerged as a promising approach for imaging RNA dynamics in live cells. We recently reported the development of several high-performance FRs, named Pepper, Clivia and Okra, that show advantageous properties, including high cellular brightness and photostability, low ion dependence and/or large Stokes shifts, and have been used to image diverse RNA species in live cells. In this protocol, we provide easy, efficient and generalizable strategies for using FRs to visualize different RNA species in bacteria and mammalian cells by expressing the RNA of interest tagged with one or more copies of the aptamer. We also provide a detailed procedure for multiplexed RNA imaging using orthogonal FRs and the steps to perform super-resolution live imaging of RNAs. The protocol typically takes 5–7 d, including cloning, transfection of mammalian cells or transformation of bacteria, live imaging and results analysis. This protocol is applicable to the real-time monitoring of the localization and dynamics of RNAs of interest in live cells. This protocol provides guidelines for using fluorescent RNAs, entities consisting of an RNA aptamer bound to its cognate fluorogenic dye, for live imaging of the localization and dynamics of different RNA species in bacteria and mammalian cells.

In this work, we describe an engineering approach that leverages ecological drift to generate Minimal Microbiomes; microbial consortia that are relatively simple, cohesive, and functionally complete. This process can be applied to any microbial ecosystem, provided that the target microbiome can be experimentally mimicked. Empirical support for this approach has emerged from multiple independent studies. We use simulations across diverse scenarios, significantly varying niche structures and biotic interactions, to explore the experimental conditions and source microbiome characteristics that favor successful outcomes, within a computational framework that also enables the study of microbial community assembly. Our results indicate that the effectiveness of this approach is constrained by several factors, and that perfect outcomes should not be routinely expected. Nevertheless, despite its drawbacks, this strategy remains a powerful tool for simplifying microbiomes and isolating key co-adapted populations, enabling the construction of low-diversity consortia that retain community function and present ecological cohesion.

Viral proteins interact with host proteins to hijack cellular pathways important for viral replication. Viral mimics are proteins whose structural similarity to host-mimicked proteins allows them to interact with mutual host targets. This mimicry poses a challenge for the host—how to avoid mimics without compromising essential interactions with host-mimicked proteins. Despite the prevalence of mimicry, the evolutionary dynamics between host and viral mimics remain largely unknown. We address this by integrating structural modeling, host–virus interaction networks, and comprehensive evolutionary analyses of host and viral proteins. We show that host proteins targeted by mimics and host-mimicked proteins are highly conserved, and that this is related to functional constraints imposed on host proteins. Host interface residues that interact with both mimics and host-mimicked proteins evolve slowly, while residues that exclusively interact with mimics evolve significantly faster. Surprisingly, viral mimics do not evolve rapidly, instead displaying complex evolutionary patterns. Our analysis reveals host’s limited capacity to escape mimicry and viral evolution to exploit this, and highlights how constraints lead to unexpectedly slow evolution of host–virus interaction networks.