We present FLOWR.ROOT, an SE(3)-equivariant flow-matching foundation model that unifies pocket-aware 3D ligand generation with multi-endpoint binding affinity prediction (pIC50, pKi, pKd, pEC50) and pLDDT-based confidence estimation in a single backbone. One trained model supports de novo pocket-conditional generation, interaction- and pharmacophore-conditional sampling, scaffold hopping and elaboration, and fragment growing or replacement, enabled by a mixed isotropic–anisotropic prior placement strategy. Training proceeds in three stages: large-scale pre-training on billions of ligand conformations and millions of mixed-fidelity protein–ligand complexes, refinement on curated co-crystal data, and project-specific adaptation via parameter-efficient LoRA finetuning. Joint structure–affinity modelling enables inference-time importance-sampling guidance for single- and multi-objective design without external scoring functions. Case studies on kinase selectivity (CK2α/CLK3) and scaffold elaboration on TYK2, ERα, and BACE1 illustrate utility from hit identification through lead optimization. Structure-based generative modeling is rapidly reshaping drug discovery by enabling pocket-aware ligand design alongside predictive evaluation of binding properties. This manuscript introduces FLOWR.root, an SE(3)-equivariant flow-matching framework that jointly generates high-quality 3D ligands and predicts multi-endpoint binding affinities, demonstrating state-of-the-art performance, efficient domain adaptation, and practical impact across de novo design, scaffold elaboration, and lead optimization workflows.

Autonomous self-reproduction is a major goal of bottom-up synthetic biology aimed at building artificial cells. This requires that the genome be replicated by its self-encoded replication machinery. While the reconstituted E. coli chromosomal replication system, termed the Replication-Cycle Reaction (RCR) system, offers a promising platform for genome-scale replication, its generation from genetic information has not yet been achieved. Here we show that a 53 kb circular DNA, termed RCR module-genome, encoding all 26 RCR proteins, can self-replicate in a one-pot reaction when expressed using the protein synthesis using recombinant elements (PURE) system. We first built a prototype of the RCR module-genome and then optimized reaction conditions and solved expression bottlenecks to achieve robust self-replication. This artificial module-genome supports more than 28 doublings of recursive self-replication. This system, termed PRIMES (PURE-driven RCR for In-vitro Module-gEnome Self-replication), represents a milestone toward constructing self-reproducing artificial cells.

Soil microbiomes and neighboring plants both greatly influence plant growth, thereby shaping community dynamics. Although these processes have been studied extensively, they are rarely integrated in a mechanistic framework. Consequently, how they interact mechanistically remains poorly understood. Here we address these questions using a plant–soil feedback approach with six grassland species grown alone or with five different neighbors in sterile soil and in soils with legacies of conspecific and heterospecific plants. Focal plants experienced stronger growth suppression by neighbors in sterile soil, but this was alleviated by microorganisms in soils with legacies, and particularly in soils with heterospecific legacies. Plants consequently often performed better with a heterospecific neighbor than with a conspecific one and in many cases reached equal or even bigger size than when growing alone. Neighbors restructured the root-associated bacterial community of the focal plant towards that of the neighbor, with stronger convergence in neighbor legacy soils, and these shifts mirrored the growth responses of the focal plants. Importantly, the degree of this microbial convergence predicted growth responses. An independent inoculation experiment with natural bacterial communities cultured from focal plants supported the idea that neighbor-induced bacterial community shifts contribute to these growth responses. Our results reveal that neighbor-induced bacterial reassembly, modulated by soil microbial legacies, mediates plant–plant interactions. This mechanism provides new insights into how soil microbiomes can integrate with direct plant interactions, which can ultimately influence coexistence and competition in grassland ecosystems. In competition, plant performance in conspecific and heterospecific soils is determined not only by negative or positive plant–soil feedbacks but also by neighbours inducing changes in the root microbiome that subsequently regulate plant growth.

Mammalian biodesign increasingly demands decisions that are hard to make with intuition and trial and error alone. Placing virtual cell and tissue models at the centre of an artificial intelligence-orchestrated design–build–test–learn loop turns discovery into a disciplined engineering practice that prioritizes decision quality, shortens iteration and improves translational reliability.

Mutagenesis is a fundamental, yet poorly understood, source of genetic variation that underpins microbial evolution and adaptation. When the Bacillus subtilis replicative DNA polymerase (DNAP) PolC encounters DNA lesions induced by endogenous or exogenous insults, it stalls and disassembles. If error-free DNA damage tolerance (DDT) sub-pathways fail to circumvent the lesion, bipartite translesion synthesis (TLS) DNAPs (PolY1 or PolY2 together with PolA) may bypass the damage to resume DNA synthesis. Mismatch repair subsequently removes misincorporated nucleotides. Here, we investigate which proteins loaded at stalled replication forks influence mutation dynamics mediated by TLS DNAPs. We demonstrate that ΔrecA, ΔpolA, or ΔpolY1 ΔpolY2 mutations strongly reduce cell survival following DNA damage and mutagenesis. The accessory proteins DisA, RarA, RecD2, DinG, and Mfd, which physically interact with PolA and/or RecA, differentially affect cell survival after DNA damage in the absence of TLS DNAPs and differentially modulate mutagenesis by regulating the activity of distinct TLS DNAPs, highlighting their roles in error-prone DDT. We also reveal that SOS-independent mutagenesis operates in the ΔpolA ΔrecA background. Elucidating the regulatory network underlying TLS provides a framework to understand bacterial speciation and may uncover new avenues to limit antibiotic resistance emergence.

Bacterial transcription termination is a critical yet underexplored layer of gene regulation in microbial ecosystems. Existing computational tools, however, primarily focus on predicting transcript 3′ ends generated by Rho-independent terminators (RITs) in a few model species, leaving gaps in understanding those generated by Rho-dependent terminators (RDTs) and their diversity across Bacteria. We developed BATTER (Bacteria Transcript Three Prime End Recognizer), a deep learning-based framework for predicting bacterial transcript 3′ termini. BATTER leverages the observation that conserved stem-loop structures are frequently associated with 3′ ends of primary transcripts terminated by both RIT and RDT mechanisms across diverse bacterial clades. Compared with existing approaches, BATTER demonstrated superior performance and scalability, enabling a comprehensive analysis of 42,905 representative bacterial genomes. This large-scale application revealed that stem-loop structures exhibit clade-specific properties with greater variations between species than between gene families. Notably, BATTER uncovered that certain Cyanobacteria lineages, despite lacking rho homologs, harbor Rho utilization (RUT)-like sequences near 3′ ends, and preliminary experimental validation in E. coli supports their partial functionality in transcription termination. Additionally, BATTER systematically identified pervasive premature termination events in antimicrobial resistance (AMR) genes. BATTER enables large-scale comparative genomic analyses of transcription termination, providing a powerful framework to investigate termination-associated transcriptional regulation in microbial communities. The BATTER tool is available at https://github.com/xu-research-lab/BATTER.

Gene amplification, a common route to bacterial adaptation, often occurs through recombination between two copies of an insertion sequence (IS) element flanking a genomic region. Alternative non-canonical structures have also been proposed, in which a duplication is formed by a single IS element whose two ends join two distant chromosomal loci. However, the prevalence of such non-canonical structures and their role in bacterial adaptive evolution remain unclear. Here we developed AmpliFinder, a computational tool that uses short-read sequencing data to systematically identify pairs of IS–chromosome junctions that correspond to the two ends of the same IS element yet map to distant genomic loci flanking amplified regions. Applying AmpliFinder to 10,347 laboratory-evolved E. coli and Acinetobacter baumannii isolates, we identified 113 distinct de novo IS-associated amplifications and found that non-canonical amplifications are the most abundant mode of amplification. We validated the inferred architectures using ultra-long-read sequencing and propose a model for non-canonical amplification formation supported by the observation of nested intermediate structures. Quantifying enrichment for antibiotic-resistance genes in amplicons, we find that non-canonical amplifications more effectively and narrowly amplify genes under selection. These results highlight the role of non-canonical IS-based amplifications in the adaptive evolution of bacteria. Transposon-associated amplifications are dominated by a previously underappreciated architecture, where the transposon is found only between, not flanking, the amplicons. These amplifications facilitate the evolution of bacteria and their adaptation to antibiotics.