Nature Biotechnology

1Clinical metagenomic sequencing is increasingly used to detect pathogens directly from clinical samples, but sparse and uneven genome recovery caused by low microbial biomass and stochastic fragment loss makes species-level validation difficult. Under these conditions, the same observed coverage may arise either from genuine genome-wide representation or from reads confined to conserved loci shared among related taxa. Current heuristic thresholds based on read counts, breadth or related alignment summaries target the realised coverage rather than the generating process. We show that, under hidden fragment loss, coverage is the wrong inferential object for uniformly valid species-level inference. What remains informative is the spatial dispersion of surviving reads across genomic coordinates. This change of inferential object allows us to ask how much of the candidate genome must be represented in any source capable of reproducing the observed evidence. This yields the Metagenomic Alignment Validation Index (MAVI), a conservative lower bound on the fraction of the candidate genome required to explain the observed read dispersion. MAVI therefore provides an error-controlled, interpretable measure of genomic confidence: the minimum genomic content a competing source would need to share with the candidate to mimic the observed evidence. Evaluated across 643 long-read metagenomic samples, including 280 clinical samples, MAVI reproduced independently supported organism calls, resolved sparse low-coverage ambiguities and exposed workflow-level collapse of stochastic dispersion, without relying on heuristic coverage thresholds.

Plasmid construction underpins molecular biology and synthetic biology, yet validation is often limited to the inserted fragment rather than the whole plasmid, and around a third of laboratory-made plasmids carry sequence errors that can affect function. Sanger sequencing scales poorly across whole plasmids, while short-read approaches cannot resolve the repeated DNA parts, such as promoters, that are common in synthetic constructs. We present PICARD-seq, a rapid nanopore-based protocol that uses off-the-shelf Tn5 rapid barcoding reagents and a MinION to sequence pools of whole plasmids in under a day, and we systematically benchmark the computational pipelines available for analysing the resulting data. Using a curated set of 25 plasmids of known sequence spanning 3.0-20.6 kbp, including various dilution series and repetitive multi-part constructs, we ran five independent replicates of each pipeline. The ONT EPI2ME Clone Validation workflow was fast (13-18 min) but stochastic, varying between replicates for both plasmids assembled and what sequence was returned; Canu outperformed the default Flye assembler, and reducing the minimum coverage parameter from 60x to 20x substantially improved assembly of large, repetitive, and dilute samples. The ensemble assembler Autocycler was slower (81-111 min) but gave the highest and most consistent rate of recovering the expected sequence. Complementary read mapping with minimap2 distinguished genuine sequence differences from assembly artefacts. Applying PICARD-seq to problematic plasmids revealed backbone concatemers, a misincorporated promoter part, and a mixed population of rearranged molecules in a repetitive construct. PICARD-seq makes routine whole-plasmid validation practical and affordable for individual laboratories.

Metagenomic sequencing allows for detection of any number of pathogenic viruses from a single sample without a priori knowledge of its contents. This has created a substantial opportunity to characterize pathogens in humans and their environment, but the field is held back by a lack of bioinformatics tools for taxonomic profiling and reconstructing virus genomes. The ideal profiler would simultaneously do two things: report viruses at single-read sensitivity so that ultra-low-abundance pathogens are not missed and hold near-perfect specificity so that a pathogen is never falsely reported. Here we present EsViritu, a taxonomic profiler that pairs an iterative greedy clustering algorithm and lowest-common-ancestor taxonomic refinement with a curated Virus Pathogen Database to reconstruct virus genomes and generate virome profiles directly from reads. We benchmarked EsViritu against four leading metagenomic profilers (Kraken2, Centrifuger, Metabuli, and Sylph), running each program across a full parameter sweep and against the same Virus Pathogen Database for an apples-to-apples comparison. Across synthetic positive controls, pathogen-free negative controls, and mutated mock communities, EsViritu was the most accurate tool at reconstructing community composition: it led or tied every tool on relative abundance accuracy at all taxonomic ranks and led or tied on detection across ranks. Importantly, other tools frequently report the presence of multiple related taxa when only reads for a single genome are present, but EsViritu faithfully records the correct, single taxon and genome. EsViritu achieves read-level sensitivity for sequences as low as 85% average nucleotide identity to the nearest reference and reports essentially no false positives in pathogen-free host and synthetic samples. EsViritu also returns reconstructed consensus genomes whose nucleotide identity to the nearest reference closely tracks ground truth, yielding both a profile and genomes from a single run.

RNA isoform dysregulation drives disease pathogenesis and is the target of FDA-approved splice-switching therapeutics. However, multiplexed sequencing methods discard splice junction information because only 3' termini are barcoded and counted. Here, we repurpose acylation and click chemistries to conjugate bioorthogonal barcodes (bobcodes) directly onto multiple internal positions along cellular RNAs. Bobcoded RNAs from multiple samples are pooled for multiplexed cDNA synthesis, during which reverse transcriptase switches from each RNA template onto its tethered bobcode with greater than 99% accuracy in species mixing experiments. Bobcode attachment intervals set cDNA insert sizes without a library fragmentation step, and priming with poly(dT) or random hexamers selects between 3'-end counting and full-length isoform capture. A bioorthogonal barcode-sequencing (BOB-seq v0.1) drug screen identifies transcriptome-wide on- and off-target RNA splicing effects and outperforms existing multiplexing RNA sequencing methods in workflow simplicity, sample-to-sample variability, and barcoding accuracy. Bobcodes add isoform resolution to scalable multiplexed RNA sequencing.

Most human genes can end their messenger RNAs at more than one position, and the choice shifts between cell types. Single-cell sequencing reads that choice in individual cells, but two questions are seldom asked: are the reported sites real, and does the test for a cell-type difference hold its error rate? Here we introduce our tool PeakATail, which calls polyadenylation sites from reads still carrying a piece of the tail, ranks them by how many distinct molecules carry that evidence, discards sites where adenosine-rich genomic sequence mimics a tail, and pairs them with a differential test whose error rate is measured, not assumed. The evidence is scarce, so the tool is built to be right rather than complete. Across four libraries from two species, 71 to 83 percent of its calls fall within 100 bases of a curated atlas, over thirty times the shuffled-control rate; long reads from different donors support 76.5 percent of the human calls. The cost is sensitivity. Only one of six test configurations held its false-positive rate under shuffled labels; with it, 15,942 cell-type switches replicated across twelve lung cancer patients and none in ten shuffled-label runs. One pre-registered target was missed and is reported.

Microbial metagenomes encode vast catalytic diversity, but recovering enzymes of novel, yet-undescribed functionality typically requires whole-community assembly plus a means of identifying active proteins ab initio. The first step of the task is compute-intensive and misses low-abundance sequences. The second is complicated by our inability to predict novel functions. REBEAN, our DNA language model, sidesteps the latter by assigning each sequencing read a high-level Enzyme Commission (EC) class or a non-enzyme label without alignment. Here we build REFinder, a pipeline that addresses both steps by routing REBEAN-annotated reads to assemble only the putative enzymatic reads that have identified catalytic signatures. In our evaluation of 50 microbiome metagenomes, REFinder identified 1.2 to 2.1 fold more enzymes than could be recovered via homology-based annotation of the proteins from the corresponding full assemblies. Moreover, it was as much as 6.4-fold cheaper computationally than full assembly. Across all samples, REFinder identified at least three fourths and as many as 90% of the homology-accessible enzymes identified via full assembly of the complete metagenomes. Notably, a fraction of these, 22% to 45% per EC class, carried no similarity to Swiss-Prot proteins, i.e. a set of enzymes homology cannot annotate. For roughly two fifths of the over thirteen thousand such novel oxidoreductases from two saliva samples, ESMFold predicted structures aligned with TM-score[≥]0.7 to a characterized enzyme structure in the PDB - a substantial structural similarity without sequence homology. These results illustrate that targeted, alignment-free assembly turns even well-mapped microbiomes into a source of thousands of previously invisible but credible novel enzymes, raising our expectations for exploration of environmental microbiomes.

Local alignment tools struggle at querying against thousands of bacterial genomes, because the reference indexes are memory-intensive and base-level dynamic programming is slow. We present alamem (https://github.com/yunwilliamyu/alamem), an approximate local aligner that streams unprocessed reference databases and estimates average nucleotide identity (ANI) from Maximal-Exact-Match (MEM) chaining statistics. Alamem can query a bacterial genome against >85,000 prokaryotic genomes in under 30 seconds (with multithreading) using 4 GB of memory, making it over 16x faster and 21x more memory-efficient than existing local aligners using their default running modes. On top of alamem, we built alasight (https://github.com/graceoualline/alasight) for detecting signatures of horizontal gene transfer (HGT) by finding pairs of alamem hits to divergent genomes. Alasight achieves over an order of magnitude better sensitivity on data from real bovid microbiomes than existing mobile genetic element-detection software and provides depth and sparsity information supporting each putative HGT region.

Bacterial conjugation and CRISPR base editing offer complementary routes for programmable DNA transfer and heritable genomic writing, but their combined potential for information encoding in living cells remains unexplored. Here, we introduce CONCORD, a bacterial encoding platform that couples multiplexed bacterial conjugation to genomic base editing for synthesis-free in vivo DNA data writing. A recipient Writer strain expresses the base-editing machinery and is equipped with a writing template, whereas reusable donor Guide strains deliver the encoding sgRNA signals by conjugation. Using this setup, we encode binary information into genomic DNA in a high-throughput manner. Moreover, we systematically evaluated multiple orthogonal base editors for an expanded encoding repertoire and improved data recovery through redundancy encoding. We deploy the system to encode 45,689 bits into DNA including two culturally and historically significant artifacts: a greeting from NASA's Voyager Golden Record and a photograph of Navajo petroglyphs and recovered these datasets with 100% and 99.5% accuracy, respectively, without computational error correction. Connecting and engineering the unique mating and immune system of microbial cells we have established a framework for programmable information writing in bacterial populations, with applications for biological event recording and sustainable DNA data storage.

Recent efforts have demonstrated that systematic analysis of the millions of metagenomes in the Sequence Read Archive (SRA) enables groundbreaking biological discoveries. However, SRA-scale comparison of whole metagenomes has remained elusive. Here, we introduce the most comprehensive resource to date for experiment-to-experiment analysis of all SRA metagenomes, comprising FracMinHash sketches (470GB), hypergen sketches (19GB), and an all-pairs Jaccard similarity matrix (90GB), which we release online at https://github.com/RolandFaure/SRA-metagenome-sketches-databases. These resources enable a wide range of analyses, which we demonstrate with several applications. First, we use the FracMinHash sketches to uncover temporal trends, finding that the amount of newly discovered microbial sequence continues to grow, rather than saturate or shrink. Second, we use the hypergen sketches, which enable nearest-neighbor search, to identify co-assembly candidates and thereby improve assembly quality. Third, we use the similarity matrix to analyze all SRA metagenomes, identifying duplicated submissions and over 50K candidates for metadata correction. Together, our resources unlock systematic analysis of metagenomic data at unprecedented scale.

Metaproteomics analyses require protein sequence databases, yet the microbial proteins present in complex communities are not known in advance. Searching reference catalogs containing hundreds of millions of candidate sequences reduces identification sensitivity, while generating sample-specific databases through metagenomic sequencing adds time and cost. We introduce FastaLake, a computational workflow that uses full-length de novo peptide predictions by AlphaNovo to retrieve complete proteins from broad sequence resources and selects compact databases for conventional searching. In a defined eight-member community, FastaLake approached reference-search peptide yields without knowing its members. Using public sequence catalogs, FastaLake identified about as many peptides as matched metagenomic databases in two stool cohorts and 20-35% more in intestinal fluid at the same nominal 1% peptide false discovery rate. The resulting proteomes reveal personal functional profiles that persist across intestinal regions despite differences in leading species between intestinal fluid and stool. DNA, RNA and protein profiles agreed on broad differences between patients but only weakly on individual functions. All 726 acquisition-specific databases were smaller than the human reference proteome. Compact databases allowed non-specific searches of all 347 CapScan acquisitions, revealing histidine enrichment at peptide C termini in intestinal fluid. FastaLake enables metaproteomics analysis without matched metagenomic sequencing by constructing search databases from sequence resources and de novo peptide sequencing.

Viruses shape microbial metabolism through auxiliary viral genes (AVGs) that reprogram host metabolism, physiology, and gene regulation during infection. Identifying AVGs is complicated by ambiguous viral versus cellular assignment, inconsistent curation, and reliance on sequence homology that misses divergent functions. Here we present CheckAMG, which pairs annotation-based AVG prediction, curated with a per-gene viral scoring system and ontological mapping of metabolic function, with a finetuned genome-language model for annotation-independent discovery. Benchmarked against DRAM-V and VIBRANT across three ecosystems, CheckAMG produced more strongly supported auxiliary metabolic gene (AMG) calls, and uniquely among AMG tools reports auxiliary physiological (APGs) and regulatory (AReGs) genes. Applied to 1,005,980 fragmented or complete soil and human-gut viral genomes, CheckAMG identified 541,199 AVGs tracking biome-specific protein clusters and functions, including 109,107 proteins with no detectable sequence similarity to any reference database. CheckAMG thus provides a reproducible, scalable framework for identifying the genes through which viruses reprogram host biology.

Droplet-based and split-pool barcoding are the two dominant high-throughput single-cell RNA-seq paradigms, yet their systematic biases remain incompletely characterized. We compared seven matched 10X Genomics Chromium and Parse Biosciences Evercode WT sample pairs from five studies spanning multiple tissues and species focusing on gene counts and compositional differences rather than cell-type resolution. By separating Parse reads by priming strategy, we distinguished biases arising from priming chemistry from platform-level effects. These effects were largely independent, with platform-level differences substantially larger. Priming strategy introduced a GC-content bias within Parse, whereas the two platforms showed a consistent compositional shift across priming strategies, supported by differential capture of nuclear- and cytoplasmic-enriched transcripts. We interpret this shift as a relative enrichment of nuclear RNA in Parse. Finally, using contamination measures that do not assume a droplet-specific model, we found background RNA contamination in Parse to be comparable in magnitude to that in 10X, contrary to claims of substantially lower contamination. Together, these results indicate that expression measurements and cell-type composition are not directly interchangeable between the two platforms.

Understanding human tissue maintenance and remodeling requires linking clonal ancestry to molecular state, yet scalable joint measurement remains challenging. Here, we introduce Droplet-MT and an integrated computational workflow for joint transcriptome and methylome profiling at ~10,000 cells per library. Methylation epimutations enable mutation-independent clonal reconstruction from high-coverage methylomes, while reference-based label transfer assigns cells to established clones at ~0.2% CpG coverage, reducing sequencing cost. Clone-aggregated somatic variants provide genetic annotation without additional single-cell genotyping. Across ~21,000 paired profiles from healthy blood and longitudinal acute myeloid leukemia, we resolved small and expanded clones with heterogeneous lineage output. In leukemia, the framework distinguished changes in malignant clonal abundance, state remodeling within persistent clones, and recovery of healthy polyclonal hematopoiesis. Malignant monocytes remained transcriptionally and epigenetically distinct from healthy and recovery-associated cells despite treatment-associated remodeling. Droplet-MT enables scalable analysis of human tissues by linking clonal organization with molecular state and retrospective genetic annotation.

Single-cell RNA sequencing fundamentally depends on defining cell-to-cell similarity to resolve distinct cell states. However, conventional dimensional-reduction frameworks such as t-SNE and UMAP inherently rely on Euclidean geometry, a construct ill-suited for high-dimensional, sparse count matrices, and produce cluster boundaries that lack biological interpretability. Here, we introduce cell-driven information fusion, an end-to-end framework that constructs cell affinity directly from raw counts by integrating three biological signals: gene detection patterns, expression magnitude of dominant transcripts, and global profile correlation. A sparsity-adaptive operator merges these signals into a unified affinity matrix, while an entry-usage diagnostic traces which biological component governs local cluster boundaries. The framework comprises two complementary implementations: c-TSNE for low-to-moderate sparsity (below 80 percent), which consistently outperforms Euclidean baselines across six benchmarks, and c-UMAP for high-sparsity regimes. Evaluated across 29 datasets and two protocols against 11 baseline methods, c-UMAP achieved the highest mean normalized and adjusted mutual information (NMI and AMI) in every setting, with all improvements remaining significant after Bonferroni correction. Because a single fixed weight vector was applied across all 29 datasets per protocol, these gains reflect intrinsic algorithmic robustness rather than dataset-specific hyperparameter tuning. Furthermore, c-UMAP scales seamlessly to 1.14 million cells, running up to 22-fold faster than the leading baseline. Crucially, by operating directly on raw counts without variance-based gene selection, our approach avoids the median 46.9 percent loss of top differentially expressed genes per cell class, preserving essential lineage markers such as IAPP and FOXP3. Together, these results demonstrate that atlas-scale efficiency, raw-data preservation, and mechanistic interpretability are fully compatible.

Barcode-based multiplexing strategies can significantly increase sample throughput and decrease costs, while mitigating batch effects of single-cell RNA sequencing experiments. However, these approaches can be limited by inaccurate or inefficient demultiplexing, resulting in cell loss and reduced statistical power. Here, we present BEACON, a novel sample demultiplexing method that efficiently learns the background count distribution from data and removes it from individual cells, thereby improving classification accuracy. BEACON outperforms other state-of-the-art methods on multiple human data sets. We apply it to cancer cell line time course experiments in vitro, enabling the identification of genes associated with aggressive tumors in vivo. Finally, we adapt BEACON to multimodal protein-transcriptome profiling, enhancing protein signal recovery to identify a CD161-positive effector memory CD4 T-cell population with a Th17-like phenotype, which we prospectively validate. BEACON can therefore be applied to other droplet-based single-cell sequencing methodologies.

Ambient RNA contamination confounds expression analysis in droplet-based single-cell and single-nucleus RNA sequencing. Here we present AmbiDose, which combines an ambient profile estimated from empty droplets with cell-specific nominal removal budgets and expression-based protection of lineage-enriched genes. Its ambient dose guides count removal rather than estimating a calibrated contamination fraction. In a held-out human-mouse GEM-X mixture, AmbiDose removed 92.0% of cross-species UMIs while retaining 99.6% of same-species UMIs. It reduced off-target marker expression in kidney-tumor and fetal-liver cohorts and yielded the lowest proximal-tubule marker leakage among the tested methods across five strain-mixed kidney libraries. Simulations supported the contribution of cell-specific dose estimates but identified reduced endogenous retention or incomplete removal when ambient and endogenous expression overlapped. AmbiDose is available as a Python package compatible with Scanpy.

Recent work applies large language models (LLMs) to single-cell data, but the cell usually reaches the model as a short ranked list of genes. This list drops much of what defines a cell, so the model reasons from a partial view. Yet the information is not lost in the measurement, only in the text. Here, we introduce Cell-Lens, a training-free structured representation of the measured cell. It writes the cell as typed blocks, from paired measurements to marker-supported programs. On CellVerse, SOAR, CellPuzzles, and SC-Arena, across five model families, Cell-Lens improves performance by up to 27.1 points. It also narrows the model gap and reduces API-reported reasoning-token usage. The largest gains occur when a defining RNA marker falls out of the ranked list. In these cases, the paired protein data still capture the corresponding signal. Shuffling the added blocks removes the gain. The gain persists after selected label-associated proteins are removed. These controls link the improvement to cell-matched biological content. We also build and release CellSpectrum, a broader benchmark spanning eight sources and seven tasks with paired modalities. The gains hold there across tasks, tissues, and modalities. These results show that a better cell representation can improve single-cell reasoning without retraining the model. Code and CellSpectrum are publicly available at https://github.com/LiZaiyuan0619/Cell-Lens.

Large-scale single-cell epigenomic atlases characterize chromatin regulatory landscapes across diverse biological contexts, spanning cell types, tissues, individuals and species. Foundation models provide an opportunity to capture the full spectrum of cellular diversity in these atlases, yet current models remain largely confined to individual species by genomic coordinate dependence and overlook regulatory information encoded in DNA sequences. Here we introduce EpiZoo, a DNA sequence-aware foundation model for cross-species single-cell epigenomics. EpiZoo converts million-dimensional single-cell epigenomic profiles from diverse species into compact cell sentences that integrate DNA-encoded regulatory information, sequence-independent epigenomic context and accessibility-based importance. Built around a mixture-of-experts transformer and containing 2.6 billion parameters in total, EpiZoo is pretrained on our manually curated multi-species Omni-scATAC corpus of approximately 20.9 million cells to learn regulatory programs across species. On external datasets excluded from pretraining, EpiZoo achieves state-of-the-art performance in fundamental single-cell analysis tasks, including feature extraction, cell type annotation and data imputation. Its sequence-aware architecture enables extension to evolutionarily diverse species, and supports comparative analysis of regulatory conservation and divergence during primate evolution. Benefiting from this modeling design, EpiZoo enables context-aware prioritization of somatic mutations in cancer and prediction of cell-type-specific chromatin accessibility from DNA sequences across genomic regions and species.

Genetic perturbations can reshape both the state of a targeted cell and the behavior of neighboring cells. Measuring these effects systematically in intact tissues requires pooled screening methods that jointly identify perturbations, molecular phenotypes, and spatial context of individual cells. Here, we report an imaging-based pooled screening method that detects perturbation identities and endogenous RNAs without enzymatic amplification. Relative to our previous enzymatic-amplification-based workflow, amplification-free MERFISH detection with direct hybridization increased recovery of perturbation barcodes by ~10-fold and endogenous transcripts by 2.5-fold, improving perturbation assignment as well as molecular profiling of both perturbed and neighboring cells. Applying this method to a genetic mosaic mouse liver screen targeting ~200 genes in hepatocytes, we profiled 1.5 million cells in intact tissues and mapped both cell-autonomous responses of perturbed hepatocytes and cell non-autonomous changes in neighboring immune and vascular populations. These included localized accumulation of pDC-like cells near hepatocytes with impaired V-ATPase function and angiogenic responses in vascular cells near Vhl-perturbed hepatocytes with hypoxia activation. Our method provides a simplified, scalable approach to linking genetic perturbations with cell state, multicellular response, and tissue organization.

Accurate species-level profiling of shotgun metagenomes becomes difficult when an organism is represented by only sparse sequence coverage. Per-read k-mer classifiers can retain sensitivity under these conditions, but short reads from conserved or shared genomic regions can generate false-positive calls among closely related references. More conservative marker-gene and genome-containment approaches generally provide stronger species-level specificity but may lose sensitivity as genomic sampling becomes sparse. We present winnow-tax, a two-stage taxonomic profiling pipeline that separates sensitive candidate nomination from genome-level confirmation. Candidate species are nominated using Sylph, Kraken2, and a branch-rescue procedure, then reads are competitively recruited to a sample-specific reference set. Species presence is evaluated using read support, observed genome breadth, and a Lander-Waterman-based breadth ratio that compares observed breadth with the breadth expected at the measured mean depth. In a controlled synthetic community of 82 genomes with human DNA background, winnow-tax maintained a stronger precision-sensitivity balance than Kraken2/Bracken, Sylph, and MetaPhlAn 4 as genome coverage decreased, with its advantage concentrated near the low-coverage detection boundary. In the CAMI III Toy Longitudinal Human Gut benchmark, winnow-tax had higher sensitivity than Sylph (0.803 versus 0.672), lower precision (0.923 versus 0.970), and a modestly but significantly higher F1 score (0.858 versus 0.793). In a clinical enteric stool cohort with culture/PCR reference testing, winnow-tax achieved the best composite performance for Salmonella, detecting 37 of 48 composite-positive samples with one false-positive call. Together, these results support a profiling strategy in which weak taxonomic signals are retained during candidate generation but require genomic evidence distributed as broadly as expected for their sequencing depth before species presence is accepted.