At the heart of LoL-align lies a deceptively simple idea: two residues in different proteins are likely to be homologous if their spatial neighborhoods look alike. The algorithm scores each candidate residue pair by examining the distances between that pair and an established set of aligned anchor pairs. For a candidate pair (j, l), the method compares the intramolecular Cα–Cα distances in the query protein with the corresponding distances in the target protein, within a 20-angstrom neighborhood of the anchor set. Each such comparison yields a log-odds value — the logarithm of the probability that the pair stems from a homologous alignment divided by the probability that it does not — and these values are summed to produce what the authors call the LoL-score. A small neural network with a single hidden layer, trained on positive pairs drawn from the same SCOPe superfamily and negative pairs from different folds, converts raw distance geometry into these probabilities.

The algorithm then works in an iterative loop that gradually grows the alignment from a few trustworthy seeds. It begins with a vectorized forward–backward computation of posterior alignment probabilities based on Foldseek’s 3Di structural alphabet and amino acid substitution scores. From the resulting matrix, the ten aligned seven-mers with the highest summed posterior probabilities are extended along the diagonal, and the three best candidates serve as seed anchors. For each seed, three steps repeat until convergence: the LoL-score matrix is computed from the current anchors, the forward–backward algorithm yields fresh posterior probabilities, and all residue pairs whose probabilities approach the maximum within a small offset are added to the anchor set. The alignment thus expands locally in three-dimensional space, anchoring itself ever more firmly in geometrically consistent residue correspondences.

Two operational modes make LoL-align versatile. In single-domain mode, final scores are normalized by (LQ × LT)^0.25 to improve ranking of compact proteins. In the default multidomain mode, if the anchor expansion stalls, the algorithm folds the 3Di and amino acid score matrices into the LoL-score matrix and reruns the forward–backward computation. This allows the alignment to leap across structural domains that lie far apart in space — outside the 20-angstrom neighborhood of any anchor — and stitch together full-length, multidomain proteins that rigid superposition methods struggle to align. Because many real-world queries are full-length proteins with multiple independently moving domains, this flexibility is arguably the tool’s most consequential design choice.

Speed presented the central engineering challenge, and the team solved it by rewriting the forward–backward algorithm itself. Earlier probabilistic alignment implementations ran unvectorized and in logarithmic space, where costly log and exp operations guard against numerical over- and underflow. LoL-align’s authors reformulated the update equations so the dynamic programming matrices can be processed with SIMD instructions on modern CPUs, computing several values per instruction. They divided the target protein into fixed-length blocks, rescaled each row by its block maximum — with scaling factors stored in log space — and thereby performed all other computations in linear space without numerical instability. An algebraic reformulation of the gap-recurrence, using precomputed constant vectors and a cumulative sum, made the gap-dependent term vectorizable despite its dependence on the immediate left neighbor. A fast C++ implementation reuses a single buffer across forward, backward and posterior passes, and employs branch-free SIMD approximations of logarithm and exponential functions to squeeze out further throughput.

The benchmarking results are striking. On the SCOPe40 single-domain database — 11,211 proteins clustered at 40 percent sequence identity, with 3,566 queries — LoL-align achieved average sensitivity up to the first false positive that was 7.3 percent higher than Dali and 9.9 percent higher than TM-align at the superfamily level. In precision–recall analysis it attained an area under the curve 14.5 percent above Dali and 2.3 percent above TM-align. These gains matter because superfamily-level detection represents remote homology: proteins whose shapes reveal shared ancestry invisible to sequence comparison. In an era when automated annotations must maintain low false discovery rates, ranking more true relatives before the first spurious hit translates directly into more functional insight per search.

Full-length searches against the AlphaFold Database present an even tougher test, and here the team constructed a reference-free benchmark: 2,000 query proteins sampled from the 2.3 million cluster representatives in the clustered AFDB were searched with Foldseek, and the top hits were re-aligned and ranked by each tool. Alignment quality was assessed with an LDDT-based G-score, the geometric mean of residue-wise precision and sensitivity, with alignments scoring below 0.5 classified as false positives. LoL-align proved considerably more sensitive than both Dali and TM-align, and when all alignments were re-ranked by G-score to isolate quality from ranking, LoL-align’s alignments achieved the highest scores. In runtime, it was roughly five times faster than TM-align and twenty times faster than Dali — meaning the sensitivity of the field’s gold standard now comes at a fraction of the cost.

Recognizing that a global similarity score alone cannot always judge whether a local alignment reflects true homology, the team trained a small neural network classifier to predict the probability of a homologous match from local alignment features. The 40-dimensional feature vector draws on alignment scalars such as the LoL-score and LDDT, an aligned-neighbor fraction summarized in histograms over a Cα contact graph, and sequence-separation topology descriptors. Trained with fivefold cross-validation on SCOPe, with folds kept disjoint between training and validation partitions, the classifier achieves calibrated precision levels: on the single-domain benchmark, a LoL-score threshold of 0.65 corresponds to 99 percent precision, 0.60 to 95 percent, and 0.50 to 75 percent. Calibrated probabilities let researchers set error budgets explicitly, a quality rarely offered by traditional alignment scores.

Perhaps the most practical aspect of the release is its integration into Foldseek, the widely used structure search engine that compresses proteins into a 20-state 3Di alphabet for blazingly fast prefiltering. In the combined workflow, Foldseek’s prefilter rapidly narrows hundreds of millions of structures to candidate hits, and LoL-align then refines and re-ranks the top hits with its more accurate probabilistic alignment. The SIMD forward–backward algorithm has also been released separately in MMseqs2 under an MIT license, potentially benefiting probabilistic sequence alignment applications well beyond structure comparison. LoL-align is freely available through the Foldseek web server and the open-source repository, and replacing TM-align with LoL-align inside the multiple-structure alignment tool Foldmason yielded slight improvements as well.

The broader significance extends beyond one software release. The comparison of predicted structures at planetary scale is becoming the successor to BLAST-style sequence searching, enabling evolutionary analyses and functional annotations across protein universes that sequence methods could never reach. LoL-align demonstrates that the classical trade-off between sensitivity and speed is not immutable: by combining a principled probabilistic model of structural homology with clever numerical engineering and hardware-level parallelism, the tool matches or exceeds the sensitivity of the slowest, most careful methods while running at near-prefilter speed. For biologists mining the AlphaFold Database for remote relatives, novel folds and evolutionary signatures, the era of choosing between accuracy and scale may finally be drawing to a close.

Subject of Research: Development and benchmarking of LoL-align, a probabilistic protein structure alignment algorithm

Article Title: LoL-align: sensitive and fast probabilistic protein structure alignment

Article References: LoL-align: sensitive and fast probabilistic protein structure alignment. (n.d.). https://doi.org/10.1038/s41592-026-03249-w

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03249-w

Keywords: protein structure alignment, LoL-align, remote homology detection, AlphaFold Database, Foldseek, forward–backward algorithm, SIMD vectorization, SCOPe benchmark, log-odds score, LDDT G-score, computational biology, Nature Methods

A new study published in Nature Methods by Jacob Schreiber, of the Research Institute of Molecular Pathology at the Vienna BioCenter and UMass Chan Medical School, aims to close that gap. The paper introduces tangermeme, a free, open-source Python package described as an everything-but-the-model toolkit for genomic deep learning. The name of the design philosophy is deliberate. Rather than competing with the many frameworks that build, train and host neural networks, tangermeme concentrates exclusively on what happens after a model has been trained: making predictions efficiently, perturbing sequences, attributing importance to individual nucleotides, calling regulatory patterns and designing new DNA. The result, according to the paper, is a comprehensive, flexible and efficient Swiss Army knife for cis-regulatory pattern identification and analysis.

The rationale for this division of labor is grounded in how the field has actually evolved. Model architectures and training strategies churn constantly, with convolutions giving way to transformers and new optimizers arriving every year. Downstream analyses, by contrast, are remarkably stable and largely agnostic to architecture. Estimating the effect of a noncoding variant on a model’s prediction works essentially the same way whether the underlying network uses convolutional filters or self-attention, and whether it was trained with one optimizer or another. This modularity means that a well-engineered library of analysis operations can serve any model, old or new. Yet until now, no optimized repository of such methods existed that generalized across model types. Instead, researchers typically ship bespoke analysis code bundled with each released model, and existing packages either implement only a few related algorithms or focus on training and fine-tuning with limited post-training support.

Tangermeme’s central technical trick is the strict separation of sequence manipulations from model operations, which can then be freely stacked to create analyses. In silico marginalization compares predictions before and after substituting a short sequence into a template, quickly identifying which motifs from a database the model responds to. Ablation is the conceptual opposite, altering a region the model may be using and measuring the drop in prediction. Variant effect estimation compares predictions before and after one or a few noncontiguous substitutions, allowing fine-mapping of candidate regulatory variants. Crucially, the operation performed after a sequence manipulation need not be a simple prediction; it can be DeepLIFT/SHAP attribution, in silico saturation mutagenesis, or any custom operation a researcher writes. This composability lets scientists ask questions such as how surrounding sequence context influences transcription factor binding, whether a regulatory element is suitable for a synthetic construct, or how two variants at the same locus interact, all in a handful of commands.

The package also embraces DNA design, implementing several methods catalogued in a recent review of enhancer modeling. The simplest is screening: generate random sequences, score them against a user-defined objective, and keep the best. Greedy substitution, sometimes called directed evolution, iteratively improves a sequence by testing every possible single-nucleotide change and keeping the best one at each step. Motif implantation works similarly but swaps in entire motifs from a database rather than single bases. Most novel is a construct marginalization method that merges design with marginalization, creating short DNA inserts that reliably shift model predictions, such as patterns of predicted chromatin accessibility, averaged across many background sequences, without requiring prior knowledge of what the construct should contain. Such libraries of designed inserts could prove valuable for synthetic biology and therapeutic applications.

Engineering choices under the hood are where tangermeme distinguishes itself from prior efforts. Operations come with built-in batching, support every data type and device available in PyTorch, and work out of the box on models with multiple inputs and outputs, arbitrary internal units including convolutions, long short-term memory blocks and transformers, and any output format from a single binary value to a full base-pair-resolution profile. Speed benchmarks illustrate the payoff: tangermeme one-hot encodes the entirety of human chromosome 1 in under two seconds, roughly three times faster than the next fastest implementation tested. That may sound trivial, but fast encoding enables on-the-fly batch generation during analysis, and the same optimization philosophy runs throughout the codebase. Batching also solves a practical bottleneck in DeepLIFT/SHAP, which in many implementations requires all background sequences to fit in GPU memory simultaneously, a serious constraint for massive models.

Perhaps the most consequential finding to emerge from building the toolkit carefully is a subtle failure mode in widely used attribution code. DeepLIFT/SHAP works by overriding the backward pass of nonlinear operations to incorporate a reference sequence, and identifying which operations are nonlinear requires checking each computational layer against an internal dictionary. Existing implementations can silently fail when they encounter a nonlinear operation outside that dictionary or when an activation is reused, often without documentation or warnings. The telltale sign is convergence deltas, which should be zero in theory and near machine precision in practice, ballooning to indicate that the attributions are simply wrong. Tangermeme never fails silently: it raises warnings when convergence deltas are too high, can return those deltas for monitoring, and allows users to register custom nonlinear functions to avoid the problem entirely.

Beyond wrapping established algorithms, tangermeme introduces genuinely new methods for distilling learned cis-regulatory logic, centered on objects called seqlets, the contiguous spans of high-attribution nucleotides that attribution methods such as DeepLIFT/SHAP, in silico saturation mutagenesis, PISA or saliency produce. The package’s novel recursive seqlet caller is, according to the paper, the first to call variable-length seqlets directly, using a principled statistical definition: a seqlet is a span whose attribution sum is statistically significant against a null distribution, with the requirement that all subspans within it also pass the test. The algorithm bins attribution values, builds null distributions for each candidate length, converts them to P values, and decodes the longest non-overlapping significant spans. Compared against a reimplementation of TF-MoDISco’s caller on synthetic sequences with an inserted MYC motif, the recursive approach produced shorter, more precise calls and ran faster on modest-to-large numbers of sequences, while TF-MoDISco’s wider, more permissive calls reflect design choices made for its original role inside a larger pipeline.

The payoff of this machinery is demonstrated on real models. Applying the automatic seqlet-calling and annotation pipeline, which maps called seqlets to motifs from the JASPAR database and then counts motif occurrences, pairwise co-occurrences and spacing relationships, Schreiber compared two models ostensibly trained to predict the same thing: MYC binding. At the PLD6 promoter, attribution analysis revealed that BPNet’s predictions are driven almost exclusively by a MYC motif, whereas the much larger Beluga model relies on a broader lexicon of motifs also associated with chromatin accessibility and transcription initiation. Counting annotated seqlets across all MYC peaks and their pairwise occurrences made these differences in learned regulatory logic immediately visible, despite both models achieving similar predictive goals.

Tangermeme is released under the MIT license, installable with a single pip command and accompanied by extensive documentation and tutorials that reproduce the paper’s figures. Although the demonstrations focus on transcription factor binding and chromatin accessibility, the authors emphasize that the toolkit applies to models of any genomic modality, including alternative splicing, transcription, enhancer activity, RNA stability and RNA binding. Schreiber also points toward the future: as large language model-based coding agents become common in research, well-documented packages with tested, composable building blocks will let such systems reuse reliable code rather than reinventing analyses, reducing the burden of auditing generated code. Future development will focus on expanding functionality and deepening documentation for exactly that purpose, positioning tangermeme as infrastructure for an era in which interpreting the genome’s regulatory code may depend as much on trustworthy software as on the models themselves.

Subject of Research: A deep learning interpretation toolkit for cis-regulatory genomics

Article Title: Tangermeme: a toolkit for understanding cis-regulatory logic using deep learning models

Article References: Schreiber, J. (2026). Tangermeme: a toolkit for understanding cis-regulatory logic using deep learning models. Nature Methods. https://doi.org/10.1038/s41592-026-03254-z

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03254-z

Keywords: deep learning, genomics, cis-regulatory logic, software toolkit, feature attribution, seqlets, transcription factor binding, variant effect prediction, DNA design, interpretability, Nature Methods, open source

Liquid chromatography is a standard technique used to separate mixtures of substances. As a mixture passes through a separation column, individual molecules are retained to varying degrees based on their chemical properties. Consequently, they elute from the column at different times. This elution time, known as the retention time, serves as a key indicator for identifying which substance is present in a sample. However, accurately predicting this time has historically been difficult because it depends heavily on specific experimental conditions, including the type of column, solvent, gradient, pH, temperature, and even minor technical variations in the apparatus.

Prof. Dr. Sebastian Böcker, a bioinformatician at the University of Jena, explained that previous predictive models often reached their limits due to this sensitivity to experimental setup. For instance, replacing a tube in the apparatus with one that is slightly longer can significantly shift the measured times. As a result, earlier models typically required extensive training or fine-tuning using data from the exact measurement system on which they were to make predictions. This meant researchers had to measure numerous standard substances before the model could be used effectively, a process that is time-consuming, expensive, and often impractical for many applications.

The new approach, developed by Böcker’s team, overcomes these limitations by focusing on the most commonly used type of liquid chromatography, known as reversed-phase mode. Unlike previous methods that are tied to specific hardware, this new method can make predictions for new systems and unknown molecules without prior system-specific training. The method operates in two distinct steps, allowing it to generalize across different experimental setups while maintaining high predictive accuracy.

In the first step of the process, a machine learning model calculates a retention order index for a given molecule. This index does not directly represent a time value but instead describes the molecule’s position within the expected sequence of retention order. By focusing on the relative order rather than absolute times, the model can account for the inherent variability in experimental conditions. This step establishes a framework for how molecules are likely to separate from one another, independent of the specific hardware used.

In the second step, the calculated retention order index is converted into specific retention times using a small number of known reference points. This conversion allows the method to translate the relative order into practical, measurable times for a specific experimental setup. Fleming Kretschmer, a co-first author of the paper who worked on the method during his doctorate, emphasized that this two-step process is a key innovation. He noted that their method outperforms other approaches that require extensive training on the target system, enabling precise predictions out-of-the-box even for new systems.

The development of this tool, named “2-step,” represents a significant advancement in computational modeling for analytical chemistry. The method is available as a software package and a web application, making it accessible to researchers who may not have extensive programming expertise. According to the researchers, the tool is designed to be integrated into existing analytical programs, which could enable laboratories working with liquid chromatography and mass spectrometry to use it automatically in their workflows. This integration could streamline the process of identifying unknown molecules in complex samples.

The implications of this new method extend across several fields of research, including drug discovery, natural product research, and environmental analysis. In natural product research, scientists often seek substances from bacteria, fungi, or plants that could serve as new antibiotics, cancer drugs, or active compounds against other illnesses. The ability to predict retention times more reliably helps researchers confirm whether a measured retention time matches a presumed chemical structure, thereby aiding in the identification of potential therapeutic agents.

Similarly, in environmental analysis, food chemistry, and pharmaceutical research, the question of what is contained in a complex sample is paramount. The new method provides a robust way to address this question by offering reliable predictions that do not depend on extensive prior calibration for each new system. This could reduce the time and cost associated with sample analysis, allowing researchers to focus more on interpreting the biological or chemical significance of the findings rather than on the technical aspects of sample preparation and calibration.

The study highlights the potential of artificial intelligence to solve complex problems in analytical chemistry by leveraging machine learning to handle the variability inherent in experimental setups. By decoupling the prediction of retention order from the specific experimental conditions, the researchers have created a tool that is both flexible and accurate. As the field of metabolomics continues to grow, with an increasing number of small molecules being analyzed, methods like “2-step” will likely play a crucial role in enabling the rapid and accurate identification of these compounds in diverse biological and environmental contexts.

Subject of Research: Biology

Article Title: When does the molecule emerge from the column?

Article References: When does the molecule emerge from the column?. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: Artificial Intelligence, Analytical Chemistry, Liquid Chromatography, Metabolomics, Machine Learning, University of Jena, Nature Methods, does, molecule, emerge, column, scientific research

Conventional Hi-C, first described more than a decade and a half ago, revolutionized genomics by revealing that chromosomes fold into compartments, topologically associating domains, and loops that bring distant regulatory elements into physical proximity. But the technique has always carried a fundamental trade-off. To generate enough material for sequencing, researchers typically dissociate tissues into cell suspensions or grind them into homogenates, destroying the spatial context that makes a brain a brain. Single-cell Hi-C and imaging-based chromatin tracing recovered some of that information, but at the cost of throughput, resolution, or genomic coverage. Spatial transcriptomics, meanwhile, maps gene expression across tissue but says nothing about the physical folding of the DNA that produces that expression. Spatial Hi-C was designed to close precisely this gap.

The technical core of the method is a microfluidic device that performs the Hi-C chemistry in situ on thin tissue sections. The chip contains arrays of parallel channels, with widths of 10, 20, or 50 micrometers depending on the desired resolution, through which barcoded reagents are delivered to spatially defined strips of the section. By running the barcoding in two perpendicular passes, each microscopic spot on the tissue acquires a unique combination of two barcodes, effectively assigning a spatial address to every chromatin contact captured there. The researchers verified that the channels do not leak or diffuse into one another using fluorescent dyes, and demonstrated that valid contact counts concentrate within expected barcode combinations rather than spreading across the chip. The result is a contact map, the raw currency of Hi-C analysis, attached to a coordinate on the tissue.

To validate the approach, the team applied spatial Hi-C to sections of mouse embryos at embryonic day 13. Pseudobulk contact matrices, generated by pooling data across an entire section, showed the expected high-quality features of Hi-C data, including clear compartmentalization and domain structure at resolutions down to 10 kilobases. Unsupervised clustering of the spatial data reproduced the anatomy of the embryo, separating the central nervous system from surrounding mesenchyme and other tissues without any anatomical annotation being provided. When the researchers compared their spatial Hi-C clusters with bulk Hi-C data generated from manually dissected embryonic organs, principal component analysis showed tight agreement, and aggregate compartment and TAD signals were highly consistent. Correlations between replicate sections exceeded 0.95, indicating that the method is both faithful and reproducible.

Integrating spatial Hi-C with spatial transcriptomics on adjacent sections proved particularly powerful. In the E13 embryo, the team identified differential chromatin interactions between the central nervous system and mesenchymal regions, including changes in A/B compartment scores, reorganized TAD boundaries, and distinct chromatin loops. Genes such as Nova1, a neuronal splicing regulator, and Tbx3, a transcription factor involved in developmental patterning, showed spatial expression patterns that aligned with spatially resolved changes in their local chromatin contact landscapes. Differential loop anchors were enriched for genes involved in neural and developmental functions, and examples such as Dner in the nervous system and Gja1 in mesenchyme illustrated how enhancer-promoter wiring differs across tissue regions. The compartment scores derived from spatial Hi-C tracked the expression of marker genes for neural progenitor cells, premature neurons, heart, and liver clusters identified by the transcriptomic data.

The adult mouse cerebellum provided the most striking demonstration of the method’s spatial resolving power. At 20-micrometer resolution, spatial Hi-C distinguished the molecular layer, Purkinje layer, granular layer, and white matter by their 3D genome architectures. Within the granular layer, the analysis revealed subclusters with distinct 3D genomes that aligned with the anatomical lobules of the cerebellum, the repeating folds that organize cerebellar function. Genes such as Dpp6, Gprin3, and Galntl6 showed promoter-region compartment scores that varied systematically across these granule cell subclusters, and the corresponding spatial expression patterns in the transcriptomic data mirrored the chromatin differences. The researchers also documented B-compartment fusion events and differential TAD reorganization between the granular and Purkinje layers, including loops at the Etv1 locus, a regulator of granule cell maturation, whose associated enhancer was validated in an independent in vivo reporter assay.

Pushing resolution further, the team achieved single-spot 3D chromatin reconstruction in the adult cortex at 10 micrometers, a scale comparable to single-cell data. Using the Tensor-FLAMINGO algorithm, they reconstructed distance matrices and 3D structures for individual nuclear spots that closely matched those derived from Dip-C single-cell datasets of the same brain regions. Spatial Hi-C clusters in the cortex corresponded to cortical layers and to cell types identified by single-nucleus transcriptomic references, and spots overlapping segmented nuclei by more than 90 percent provided near-single-cell chromatin contact information. In the hippocampus, the method resolved clusters corresponding to the dentate gyrus, CA1 pyramidal layer, and even the ependymal cell layer lining the ventricles, with compartment scores at loci such as Npas3 and Vit1a varying across these microdomains.

Perhaps the most conceptually significant finding came from the developing brain. Across embryonic days 14.5, 16.5, and 18.5, the researchers measured the ratio of short-range to long-range chromatin interactions on coronal sections and found an increasing radial gradient of this ratio from the germinal zone toward the cortical plate. This gradient tracked neuronal maturity: when compared with CytoTRACE2 developmental potency scores computed from spatial transcriptomic data, the short-versus-long interaction ratio correlated significantly across clusters at 10-micrometer resolution. In other words, the physical compactness of local chromatin packing, readable directly from tissue, serves as a marker of how far a neuron has progressed along its developmental trajectory. Genes such as Sox2 and Notch2, markers of progenitors, and Celf2 and Ank3, associated with maturing neurons, showed spatial expression patterns consistent with the chromatin gradient.

The implications extend well beyond neurobiology. Disease-associated non-coding variants often act through long-range enhancer-promoter contacts that are specific to cell type and tissue context, and a method that preserves both the contact information and the anatomical address could pinpoint where in a tissue a regulatory variant exerts its effect. The data and code from the study have been deposited in public repositories, including the Genome Sequence Archive and Zenodo, with BEDPE contact files ready for standard Hi-C visualization tools, lowering the barrier for other laboratories to adopt the framework. The protocol itself is available through protocols.io, and the underlying microfluidic chip design is included with the supplementary materials.

Spatial Hi-C establishes what its developers call a paradigm for investigating spatially resolved chromatin structures within complex tissue microenvironments. By fusing the contact-mapping power of Hi-C with the positional awareness of spatial omics, it transforms the 3D genome from an abstraction studied in dissociated cells into an anatomical object that can be read layer by layer, lobule by lobule, and spot by spot. As the technology matures and resolution improves further, researchers anticipate applying it to human tissue, pathological specimens, and models of neurological disease, where the folding of the genome in its native spatial context may hold keys to understanding how regulatory architecture goes awry.

Subject of Research: Spatially resolved 3D chromatin architecture mapping in mammalian brain tissue using spatial Hi-C

Article Title: Spatially resolved chromatin architectures in mammalian brain tissues

Article References: Chen, Z., Guo, M., Zhang, L., Yu, H., Wang, X., Yu, X., Chen, M., Lv, J., Chen, Z., Peng, C., Gong, Q., Zhang, Q., Guo, R., Huang, Y., Gao, L., Jiang, S., Wang, J., Zhang, Z., Qian, J., … Chen, X. (2026). Spatially resolved chromatin architectures in mammalian brain tissues. Nature Methods. https://doi.org/10.1038/s41592-026-03218-3

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03218-3

Keywords: spatial Hi-C, 3D genome, chromatin architecture, mouse brain, cerebellum, cortex, spatial transcriptomics, topologically associating domains, A/B compartments, neuronal development, microfluidics, Nature Methods

To appreciate why this matters, it helps to consider the two measurements the method combines. The first, chromatin accessibility, is conventionally assayed by a technique known as ATAC-seq, or assay for transposase-accessible chromatin using sequencing. ATAC-seq exploits an engineered enzyme, a hyperactive Tn5 transposase, that preferentially inserts sequencing adapters into regions of the genome that are physically open. Because open chromatin tends to correspond to active promoters, enhancers, and other regulatory elements, the density of transposase insertion at any given locus serves as a proxy for that locus’s regulatory activity. ATAC-seq has become one of the most widely used assays in genomics precisely because it is comparatively simple, requires modest amounts of material, and can be applied to single cells, yet on its own it says nothing about where in a tissue the measured cell resides or how its DNA is folded.

The second measurement, chromatin organization, is conventionally assayed by Hi-C, a method that captures the three-dimensional contacts between genomic regions. In Hi-C, DNA that is physically close together in the nucleus, even if far apart along the linear chromosome, is crosslinked, cut, ligated together, and then sequenced; the resulting contact maps reveal loops, topologically associating domains, and larger-scale compartments that shape which enhancers can reach which genes. Hi-C has been enormously influential in showing that genome folding is not a passive consequence of packaging but an active regulatory layer, with disruptions to loops and domains implicated in developmental disorders and cancer. Like ATAC-seq, however, conventional Hi-C is performed on populations of dissociated cells, which erases the spatial context in which those folding patterns occur.

Spatial-ATAC-Hi-C unites these two assays on the same thin tissue section. The general strategy follows the logic of spatial genomics platforms that have emerged over the past several years: a tissue section is placed on a substrate patterned with an array of indexed capture spots or otherwise barcoded in space, enzymatic reactions are carried out in situ, and the material released at each position is tagged with a spatial barcode before sequencing. In this case, the workflow is designed so that both the transposase-tagged fragments reporting on open chromatin and the ligation products reporting on long-range genomic contacts are captured and barcoded from the same tissue coordinates. After sequencing, computational processing separates the two modalities, producing for each spatial pixel or spot both an accessibility profile and a contact profile, alongside histological information from the underlying tissue.

The dual readout is what gives the method its distinctive power. Accessibility data alone can identify which regulatory elements are active in a given tissue region, and contact data alone can show how the genome is folded there, but the combination allows researchers to ask whether the two are coupled at the same location. A region of open chromatin sitting in a tissue microenvironment where the relevant loop is present may behave very differently from an identically open region where the loop is absent. Conversely, cell types that look similar in their accessibility profiles may differ in their higher-order folding, or vice versa. Having both measurements from the same spatial coordinates makes it possible to detect such discordances rather than inferring them by overlaying separately generated datasets, each with its own artifacts and alignment uncertainties.

The authors demonstrated the technique on mouse brain, a tissue whose elaborate cellular architecture makes it a natural testing ground for spatial genomics. The brain contains densely intermingled populations of excitatory and inhibitory neurons, astrocytes, oligodendrocytes, microglia, and other cell types arranged in layered and regionally specialized structures. Applying Spatial-ATAC-Hi-C to mouse brain sections allowed the researchers to profile chromatin accessibility and organization in a spatially resolved manner across these structures, illustrating how the method can recover spatially patterned regulatory information from a complex mammalian organ. The demonstration establishes that the chemistry and barcoding steps are compatible with intact, structurally preserved tissue rather than dissociated cells, which is the central technical hurdle for any spatial assay of this kind.

The second demonstration moved from normal tissue to disease, with applications to human glioblastoma and astrocytoma samples. Gliomas are tumors arising from glial or glial progenitor lineages, and glioblastoma in particular is among the most aggressive of human cancers, characterized by infiltrative growth, pronounced intratumoral heterogeneity, and poor responses to therapy. Spatially resolved measurements are especially valuable in this setting because the microenvironment of a tumor varies dramatically from one region to another: proliferative cores, invasive edges, hypoxic niches, and regions infiltrated by immune cells can all coexist within a single section. By co-profiling accessibility and chromatin architecture across such regions, the method offers a way to examine whether regulatory programs and genome folding differ between tumor compartments, and how tumor cells’ epigenetic states relate to their immediate surroundings.

For cancer biology, the ability to measure chromatin architecture in situ is particularly intriguing. Genome folding alterations, including changes in topologically associating domains and enhancer-promoter loops, have been implicated in oncogene activation and tumor suppressor silencing, but these observations have mostly come from cell lines or dissociated tumors, where the relationship between folding and tissue context is lost. A spatial assay that preserves histological landmarks means that contact maps can be interpreted alongside the pathology visible in the section, connecting molecular events to morphological features such as tumor margins or regions of distinct grade. The same logic applies to accessibility: enhancer activity in an invasive edge cell can be compared directly with that in a cell at the tumor core, without the confounding of bulk averaging.

Like all emerging technologies, Spatial-ATAC-Hi-C comes with considerations that will shape its adoption. Spatial assays of chromatin generally trade off resolution, throughput, and sensitivity: capturing both short fragments from transposase tagmentation and long-range ligation products from the same section is chemically demanding, and the amount of informative data recoverable per spatial unit depends on the size of the spots, the thickness of the section, and the depth of sequencing. Computational analysis is also more involved than for either assay alone, since contact maps are sparse at single-spot resolution and typically require aggregation, imputation, or specialized visualization to interpret. The authors’ demonstrations on mouse brain and human glioma samples suggest these challenges are tractable, but as with any new platform, broader adoption will depend on reproducibility across laboratories, tissue types, and sample qualities, including archived or clinically derived specimens.

Even so, the arrival of a method that co-profiles chromatin accessibility and spatial genome organization marks a meaningful expansion of the spatial genomics toolkit. Until now, researchers studying gene regulation in tissues have had to choose between knowing where a regulatory state occurs and knowing how the genome is physically configured, or else stitch together separate experiments. Spatial-ATAC-Hi-C closes that gap, opening the possibility of systematically mapping how the folded genome varies across development, physiology, and disease within the native architecture of organs. For fields ranging from neurobiology to cancer research, where epigenetic state and tissue context are inseparable determinants of cell behavior, the ability to read both layers of regulation from the same tissue section is likely to find rapid and wide application.

Subject of Research: Spatially resolved co-profiling of chromatin accessibility and three-dimensional chromatin architecture in mammalian tissues

Article Title: Spatial chromatin architecture and accessibility co-profiling of mammalian tissues

Article References: Wang, P., Wang, J., Wang, Q., Youngblood, M. W., Cheng, Y., Tao, B., Wong, J. H.-Y., Luan, Y., Yu, S., Swaroop, A., Nandoliya, K. R., Najem, H., Fu, Y., McCortney, K., Zhang, D., Diao, Y., Heimberger, A. B., Sonabend, A. M., Horbinski, C. M., … Yue, F. (2026). Spatial chromatin architecture and accessibility co-profiling of mammalian tissues. Nature Methods. https://doi.org/10.1038/s41592-026-03217-4

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03217-4

Keywords: Spatial-ATAC-Hi-C, chromatin accessibility, chromatin architecture, spatial genomics, ATAC-seq, Hi-C, mouse brain, glioblastoma, astrocytoma, epigenomics, gene regulation, Nature Methods

The core idea behind SVPG is straightforward in concept but demanding in execution. Instead of aligning long sequencing reads against a single reference genome, SVPG uses a pangenome graph — a network of genomic sequences that together represent many individuals — as the backdrop against which structural variants are identified. This design choice matters because variants that are common in some populations are often absent from linear reference genomes. When such variants are forced into alignment against a reference that lacks them, reads spanning those regions can map poorly or not at all, causing true variants to be missed and spurious ones to be called. A pangenome graph, by contrast, provides a natural home for alternative alleles, allowing reads to align to whichever path in the graph best explains them.

Long-read sequencing technologies, which produce reads spanning thousands to tens of thousands of DNA bases, have transformed structural variant discovery because most structural variants are large enough that a single read can bridge the entire event. Long reads, however, bring their own computational complications: base-level errors, uneven coverage and repetitive regions of the genome can all confound alignment and variant interpretation. SVPG is built to work directly with these long-read data, combining the sensitivity of read-based analysis with the representational completeness of a pangenome. The result, according to the study, is a method that performs accurate pangenome-facilitated structural variant detection while simultaneously accelerating a second, related task that has become a growing bottleneck for the field.

That second task is pangenome graph augmentation. As sequencing projects continue to enroll participants from increasingly diverse populations, curators of pangenome resources face a persistent dilemma: when a new sample arrives, should the entire pangenome be rebuilt from scratch, an enormously expensive computational undertaking, or should the new sample simply be analyzed independently, forgoing the benefit of the collective reference? Neither option is ideal. Rebuilding everything from the ground up can take days or weeks of computation on large cohorts, while analyzing new samples in isolation discards precisely the advantage that pangenomes were designed to provide. SVPG tackles this problem head-on by providing a rapid procedure for augmenting an existing pangenome graph with new samples, updating the graph to incorporate novel variants without requiring a complete reconstruction.

The practical implications of this dual capability extend well beyond methodological elegance. Large-scale genomics initiatives, such as the Human Pangenome Reference Consortium and national genome sequencing programs around the world, are generating hundreds of thousands of genomes and continually adding to their reference structures. Every improvement in the speed and accuracy of graph augmentation translates directly into faster turnaround for researchers who depend on these resources. A laboratory that can incorporate a newly sequenced individual into the pangenome within hours rather than weeks can provide variant interpretations to clinical collaborators, population geneticists and evolutionary biologists far more quickly. In a clinical context, where structural variants can underlie developmental disorders, cancer and other conditions, timely and accurate detection has direct consequences for patients.

Underlying the method’s performance is a careful treatment of how reads interact with graph structures. Graph-based alignment is computationally more demanding than alignment to a linear reference because each read may map to multiple possible paths through the network. SVPG navigates this complexity by identifying regions of the graph where a new sample’s reads diverge from existing paths and by distinguishing genuinely novel variation from artifacts of sequencing and alignment. Variants confirmed in this way can then be added as new paths in the graph, enriching the reference with each additional sample while preserving the integrity of previously incorporated data. This incremental philosophy stands in contrast to the rebuild-from-scratch workflows that have historically dominated pangenome construction.

Accuracy in structural variant calling is typically measured against sets of variants that have been independently validated, often through a combination of multiple sequencing technologies, optical mapping and painstaking manual curation. Structural variant discovery has historically suffered from high false discovery rates in part because the signatures of these events — split reads, discordant read pairs, changes in read depth — can be mimicked by other genomic features such as segmental duplications and tandem repeats. By anchoring detection in a pangenome that already accounts for much of the common variation, SVPG reduces the ambiguity that produces such false positives, while its long-read foundation preserves the sensitivity needed to detect rare and sample-specific events that no reference graph could anticipate.

The study’s emphasis on pangenome-facilitated detection also reflects a broader conceptual shift in genomics. For two decades, essentially all genome analysis has been conducted against the same reference framework, originally derived from a small number of donors. That framework has been enormously productive, but it embeds biases: populations and individuals whose genomes differ most from the reference tend to be analyzed least accurately. Pangenome graphs are the field’s principal remedy, and tools like SVPG represent the software layer needed to make that remedy practical at scale. A pangenome is only as useful as the analytical ecosystem built around it, and each new tool that treats the graph as a first-class object rather than a curiosity expands the range of questions that can be asked of diverse genomic data.

For researchers working on nonhuman species, the significance is equally clear. Pangenomes are being constructed for crop plants, livestock, model organisms and pathogens, in many cases with even more dramatic structural diversity than found in humans. Polyploid crops, in particular, present alignment challenges that make structural variant detection exceptionally difficult. A method that can rapidly augment a pangenome graph with new samples and call structural variants accurately against long-read data offers a template that can be adapted across these domains, supporting applications from crop breeding to the surveillance of structural variation in evolving pathogen populations.

The publication of SVPG in Nature Methods signals that graph-based structural variant detection has moved from an aspirational goal to a maturing set of practical techniques. As long-read sequencing costs continue to fall and pangenome resources grow, the bottleneck in genomic analysis is increasingly shifting from data generation to interpretation and integration. Methods that combine sensitivity, specificity and computational efficiency — and that keep reference structures current without prohibitive recomputation — address precisely that bottleneck. If the trajectory of the field holds, approaches like SVPG are likely to become standard components of the genomic toolkit, quietly reshaping how structural variation is discovered, catalogued and understood across the tree of life.

Subject of Research: Pangenome-based structural variant detection and rapid pangenome graph augmentation using long-read sequencing data.

Article Title: SVPG: a pangenome-based structural variant detection approach and rapid augmentation of pangenome graphs with new samples

Article References: Jiang, T., Hu, H., Gao, R., Cao, S., Jiang, Z., Zhou, M., Gao, W., Zhou, S., & Wang, G. (2026). SVPG: a pangenome-based structural variant detection approach and rapid augmentation of pangenome graphs with new samples. Nature Methods. https://doi.org/10.1038/s41592-026-03219-2

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03219-2

Keywords: pangenome, structural variants, long-read sequencing, genome graphs, variant calling, graph augmentation, pangenome reference, genomics, reference bias, Nature Methods, SVPG, pangenome-based

Nanopore sequencing has transformed genomics by allowing DNA and RNA molecules to be read in their entirety, producing long reads that span entire transcripts and reveal complex splicing patterns that short-read technologies often miss. Oxford Nanopore platforms thread individual molecules through protein nanopores and infer their sequence from disruptions in an electrical current. This approach delivers reads that can stretch tens of thousands of bases, capturing full-length transcript isoforms and preserving modifications that are lost when RNA is fragmented and amplified. Yet the technology has long been hampered by a higher raw error rate compared with Illumina short-read sequencing, and this limitation has been particularly consequential for SNP calling, where a single misread base can be mistaken for a genuine variant.

Conventional approaches to correcting these errors typically rely on consensus building, aligning many reads to a reference genome and counting the frequency of each base at every position. Such methods work reasonably well for DNA sequencing at high coverage, but transcriptome data presents unique complications. Expression levels vary enormously across genes, meaning some transcripts are covered by thousands of reads while others yield only a handful. Low-coverage transcripts offer too few observations for statistical consensus methods to separate signal from noise, and highly expressed genes can harbor natural RNA editing events that complicate variant interpretation further. The result has been a landscape in which SNP detection from nanopore transcriptome data has demanded heavy computational filtering and often produced inconsistent results across tools and datasets.

NanoTS takes a fundamentally different approach by framing SNP calling as a pattern recognition problem that deep learning is exceptionally well suited to solve. Rather than relying solely on base-level alignments, the tool learns from the rich features embedded in nanopore signal data, the raw electrical current measurements that underlie every base call. Neural networks can detect subtle signatures that distinguish systematic sequencing errors, which tend to follow repeatable patterns tied to sequence context and pore dynamics, from genuine genetic variation, which manifests consistently across independent molecules. This capacity to model complex, high-dimensional relationships in the data allows NanoTS to make confident genotype calls even in regions where traditional statistical methods falter.

The implications of accurate SNP calling from direct RNA sequencing extend across a wide range of biological and biomedical applications. Variant detection at the transcript level reveals which alleles are actively expressed, a dimension invisible to DNA-based genotyping. Allele-specific expression, in which one copy of a gene is transcribed more abundantly than the other, plays a critical role in imprinted genes, autoimmune disease, cancer biology and X-chromosome inactivation. Researchers studying these phenomena have historically needed to combine DNA genotyping with RNA expression analysis from separate experiments, introducing uncertainty about whether observed expression patterns truly reflect the same cells and conditions. A tool that can call genotypes directly from transcriptome data collapses these two measurements into a single, internally consistent dataset.

The deep learning architecture at the heart of NanoTS represents part of a broader movement in genomics toward learned models that outperform hand-crafted algorithms. Basecalling, the process of converting raw nanopore current signals into nucleotide sequences, was revolutionized years ago by neural networks that replaced earlier hidden Markov model approaches, dramatically improving read accuracy. Variant calling is now following a similar trajectory. DeepVariant, developed by Google, demonstrated that convolutional neural networks could outperform classical statistical variant callers on Illumina data by treating genomic alignment piles as images. NanoTS extends this paradigm to the specific and demanding context of long-read transcriptome sequencing, where error profiles, coverage distributions and biological variation patterns differ substantially from genomic DNA sequencing.

Training such a model requires data of exceptional quality, since the network must learn to associate signal and alignment patterns with known ground-truth genotypes. The developers of NanoTS tackled this challenge by curating training examples in which true variants could be established with high confidence, allowing the network to internalize the distinctions between authentic polymorphism and sequencing artifact. Once trained, the model generalizes across genes, transcripts and coverage regimes, producing variant calls that remain reliable even for transcripts sequenced at low depth, a scenario that has historically been the Achilles heel of long-read transcriptome analysis.

The clinical dimension of this work is particularly compelling. Personalized medicine increasingly depends on knowing both which variants a patient carries and how those variants are expressed in relevant tissues. RNA sequencing of patient samples, whether from tumor biopsies or blood cells, is already routine in many clinical and research settings. If those same RNA reads can yield accurate genotype information without a parallel DNA sequencing run, the cost and complexity of comprehensive molecular profiling drops considerably. Cancer genomics stands to benefit substantially, as tumor transcripts carry both expressed mutations and the allele-specific expression patterns that reveal loss of heterozygosity and other genome alterations relevant to treatment decisions.

Beyond clinical applications, NanoTS opens doors for basic research into population genetics and evolutionary biology at the transcript level. Understanding how genetic variation shapes transcript diversity across individuals and species requires tools that can reliably measure both. Long-read transcriptome sequencing captures the full complexity of isoform expression, and with accurate SNP calling layered on top, researchers can begin to map how specific variants influence splicing, expression levels and transcript structure in ways that short-read studies have only approximated. The combination of full-length isoform resolution and direct genotype calling provides a uniquely integrated view of the relationship between genome and transcriptome.

The publication of NanoTS in Nature Methods underscores the growing recognition that computational innovation is as essential to unlocking the potential of new sequencing technologies as the hardware itself. As nanopore sequencing continues its rapid improvements in accuracy and throughput, tools like NanoTS ensure that the analytical pipeline keeps pace, converting noisy raw signals into biologically meaningful and clinically actionable insights. For laboratories worldwide working with direct RNA sequencing, the arrival of a robust deep learning solution for SNP calling marks a significant step toward making long-read transcriptomics a complete, standalone platform for both discovery and diagnosis.

Subject of Research: A deep learning tool for accurate SNP calling and genotype detection in nanopore long-read transcriptome sequencing data.

Article Title: NanoTS: a deep learning tool for accurate SNP calling in nanopore long-read transcriptome data

Article References: NanoTS: a deep learning tool for accurate SNP calling in nanopore long-read transcriptome data. (n.d.). https://doi.org/10.1038/s41592-026-03225-4

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03225-4

Keywords: NanoTS, deep learning, SNP calling, nanopore sequencing, long-read transcriptomics, genotyping, direct RNA sequencing, variant detection, neural networks, Nature Methods, allele-specific expression, computational genomics

Traction force microscopy, the foundational method behind CFM, has been a workhorse of mechanobiology for nearly two decades. In its classical form, cells are plated on a flat, elastic gel embedded with fluorescent beads. As a cell pulls and pushes on the gel, the beads shift, and by tracking those displacements researchers can computationally reconstruct the traction stresses the cell applies to its substrate. The approach has revealed how cancer cells turn mechanically aggressive, how stem cells sense the stiffness of their niche, and how fibroblasts remodel wound tissue. But its central limitation has always been geometry: real cells rarely live on open flat surfaces, and their force-generating machinery behaves differently when confined on multiple sides.

Confinement force microscopy addresses this gap with a microconfiner that can be dynamically tuned, positioned with precision and held stable over long imaging periods. The device brings a controlled, deformable ceiling down over cells resting on a soft elastic substrate, effectively sandwiching them in a defined three-dimensional space that mimics the physical confinement of a living tissue. Because the confining element is itself compliant and instrumented, the forces cells exert against it can be measured alongside the traction stresses they generate on the bottom surface. The result is a fuller, more faithful accounting of a single cell’s mechanical output under realistic compressive boundary conditions.

The dynamic character of the platform is one of its most significant advances. Earlier confinement approaches tended to be static, locking cells into a fixed geometry and leaving researchers unable to probe how cells respond as their mechanical environment changes. With CFM, the degree of confinement can be adjusted during an experiment. A cell can be observed in a relatively open state, then progressively constrained, then released, allowing researchers to map how traction stresses evolve as confinement tightens or relaxes in real time. This turns the microconfiner from a passive chamber into an active perturbation tool, capable of asking causal questions about how spatial restriction drives mechanotransduction.

Precision and stability, the other two pillars highlighted by the developers, matter because traction measurements are exquisitely sensitive to boundary conditions. Any drift in the position or compliance of a confining surface contaminates the displacement field on the underlying gel, corrupting the reconstructed force map. The CFM architecture maintains a stable confining geometry over the timescales of live-cell imaging, from minutes to hours, so that force reconstructions remain quantitatively reliable. The precision of the confinement also allows single cells and multicellular aggregates alike to be studied under well-defined conditions, which is essential for comparing mechanical phenotypes across genetic backgrounds, drug treatments or differentiation states.

The applications span an unusually broad swath of biology. In cancer research, one of the central puzzles is how tumor cells invading through dense stiffer tissue generate and reorient forces in tight, crowded spaces. Conventional two-dimensional traction assays cannot reproduce this compressive regime, and invasion assays in organoids or tissue slices typically sacrifice quantitative force readouts for biological realism. CFM offers a bridge between the two: cells or small aggregates can be confined in tissue-like geometries while their complete traction signature is measured with subcellular resolution. Similar logic applies to immune cells squeezing through endothelial barriers, epithelial sheets closing wounds, and stem cells differentiating under the compressive loads of developing organs.

Cell aggregates add a further layer of biological complexity that the platform is specifically designed to handle. Multicellular clusters, such as spheroids used to model early tumors or embryonic tissues, generate forces collectively, coordinating contractility across many cells through junctions and extracellular matrix. Measuring aggregate-scale traction under confinement reveals how mechanical load is distributed across the cluster, where regions of high stress concentrate, and how the collective output changes as the group is compressed. These are questions that neither single-cell assays on flat gels nor bulk rheology of tissue can answer, positioning CFM as a tool for probing emergent mechanics in multicellular systems.

Technically, the method builds on the mature computational machinery of traction force microscopy while adding a second force channel from the confining interface. Displacements of fluorescent markers in the deformable substrate are converted into surface traction fields through regularized inverse algorithms, a process well established in the field. The confining element contributes complementary measurements of the forces transmitted vertically and laterally onto the cell. Combining these datasets requires careful mechanical calibration and modeling of the microconfiner’s own elasticity, but the payoff is a genuinely three-dimensional force balance for a living cell, something the field has pursued for years with more complex and less accessible systems.

The broader significance of the work lies in what it says about the maturation of mechanobiology as a discipline. The field has moved decisively from observing that cells are mechanical objects to quantifying precisely how they sense, generate and transmit force in contexts that resemble the body. Tools such as atomic force microscopy, micropillar arrays, optical tweezers and droplet microfluidics each probe one corner of the mechanical landscape, and CFM fills a conspicuous gap: the confined, compressive, three-dimensional regime that characterizes nearly every tissue. By making that regime dynamic and measurable, the platform opens experimental doors that were previously accessible only through indirect inference.

Looking ahead, the researchers and the field at large anticipate that confinement-resolved traction mapping will become a standard component of mechanobiological phenotyping, much as flow cytometry became standard for molecular phenotyping. Because the technique is compatible with live-cell imaging, it can be paired with fluorescent reporters of cytoskeletal dynamics, focal adhesion turnover and mechanosensitive signaling pathways, linking the force maps it produces directly to the molecular machinery that generates them. For cancer biologists, stem cell engineers and developmental biologists alike, the ability to watch and measure how cells push back against a crowded world represents a meaningful step toward understanding mechanics where it actually matters: inside tissue.

Subject of Research: A tunable microconfinement technique for measuring traction forces of single cells and aggregates under tissue-mimicking spatial confinement.

Article Title: CFM: confinement force microscopy—a dynamic, precise and stable microconfiner for traction force microscopy in spatial confinement

Article References: CFM: confinement force microscopy—a dynamic, precise and stable microconfiner for traction force microscopy in spatial confinement. (n.d.). https://doi.org/10.1038/s41592-026-03216-5

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03216-5

Keywords: confinement force microscopy, traction force microscopy, mechanobiology, cell mechanics, spatial confinement, cell aggregates, 3D cell culture, traction stresses, microconfiner, Nature Methods, cellular forces, tissue mechanics

The core difficulty that SPLENDID addresses stems from how human genetic variation is actually structured. Genetic ancestry does not come in neat, mutually exclusive boxes. Decades of population genetics have shown that human variation varies smoothly across geography and history, and many people, particularly those with recent admixture from multiple continental sources, sit at points in ancestry space that no pre-specified category can capture. When existing multi-ancestry prediction methods force every individual into a labeled group, they make two kinds of errors at once. Individuals whose ancestry blends several sources are assigned awkwardly to one group or another, discarding information, and the boundaries between groups themselves are arbitrary, changing with the reference panels and clustering algorithms used to define them. As the SPLENDID authors note, clinical decisions are rarely, if ever, made on the basis of an ancestry label, so a prediction framework that depends on such labels is poorly matched to real-world practice.

SPLENDID, whose name reflects its penalized-regression foundation, works directly with individual-level genotype data at biobank scale and treats ancestry as a continuum throughout the modeling pipeline. Rather than building separate ancestry-specific scores and then deciding which one to hand to each patient, the method fits a unified penalized regression model in which genetic effects are allowed to vary smoothly as a function of continuous ancestry information, typically captured through principal components of the genotype matrix. The framework uses modern sparse-regression machinery, including L0L1-type penalties that encourage the selection of a parsimonious set of variants, together with ensemble learning strategies that combine multiple candidate models tuned across the ancestry continuum. This produces one model, not many, and that model can be applied to any incoming individual without first asking which population bucket they belong to.

The mathematical machinery behind this idea is demanding, because biobank-scale genotyping data routinely contain hundreds of thousands of individuals and millions of variants. Fitting penalized models in which variant effects interact smoothly with ancestry dimensions requires efficient coordinate-descent and discrete-optimization algorithms that can operate on ultrahigh-dimensional data. The authors, including Tony Chen and Xihong Lin of Harvard T.H. Chan School of Public Health, Haoyu Zhang of the National Cancer Institute and Rahul Mazumder of MIT’s Sloan School of Management, built SPLENDID on computational tools of the kind developed for fast best-subset selection and large-scale sparse regression, making the method practical for datasets such as the UK Biobank and the All of Us Research Program. The software is implemented for R and released openly with tutorials, lowering the barrier for other groups to adopt the approach.

The empirical evaluation is extensive. In simulation studies designed to mimic the genetic architecture and demographic history of real human populations, SPLENDID outperformed a battery of existing polygenic prediction methods, including GWAS-based approaches and multi-ancestry tools such as PRS-CSx, CT-SLEB, PROSPER and iPGS. Notably, when the team forced other methods to produce a single pooled-ancestry score rather than ancestry-specific ones, prediction accuracy dropped sharply, particularly for the GWAS-based approaches. This comparison underscores the central argument of the paper: existing frameworks implicitly assume that ancestry labels are available and reliable, and when that assumption fails, their performance collapses in ways that SPLENDID’s continuum-based design avoids by construction.

The real-world tests were conducted on two of the largest and most demographically diverse genetic resources in existence. In the All of Us Research Program, whose more than 224,000 genotyped participants include substantial representation of African, Hispanic, Asian and Indigenous American ancestries, and in the UK Biobank, with roughly 340,000 analyzed participants, SPLENDID delivered significantly higher prediction accuracy than competing methods across nine continuous traits examined. The improvements were most pronounced for non-European and admixed individuals, exactly the groups for whom conventional scores perform worst. In admixed UK Biobank samples, identified through probabilistic classification against the 1000 Genomes Project reference panel, SPLENDID maintained accuracy where label-dependent methods faltered, validating the premise that treating ancestry as a continuum captures genuine genetic heterogeneity that discrete labels erase.

One striking illustration of why continuous modeling matters comes from the biology of blood cell traits. The paper examines individual-level effects of a variant, rs1213375, on mean corpuscular hemoglobin across the genetic ancestry spectrum, an example that connects to well-known ancestry-linked variation in hematological parameters such as the influence of Duffy status on neutrophil counts in people of African ancestry. Effect sizes of genetic variants are known from recent work to be broadly conserved across human groups, but fine-scale population structure and local ancestry still modulate how variants express themselves. A model that can represent these gradual shifts, rather than jumping between group-specific estimates at arbitrary boundaries, is better positioned to translate genomic discovery into accurate predictions for everyone.

The broader significance of this work lies in health equity. Polygenic risk scores are beginning to enter clinical workflows, with validated assays for conditions ranging from cardiovascular disease to breast cancer and with large consortia such as PRIMED dedicated to reducing disparities in polygenic risk assessment. If these tools systematically underperform in non-European populations, the benefits of genomic medicine will flow disproportionately to people already advantaged by existing research infrastructure, widening health disparities rather than narrowing them. Analyses spanning the Global Biobank Meta-analysis Initiative and studies of portability across nine ancestry groups have repeatedly documented the scope of this problem. SPLENDID’s contribution is to show that part of the solution does not require waiting for larger non-European datasets alone; it also requires statistical frameworks that use the full richness of ancestry information already present in biobanks.

There are practical implications as well. Because SPLENDID produces a single model, health systems deploying it do not need to maintain separate score pipelines for separate populations, nor do they need to assign patients an ancestry label before applying a score, a step that raises both logistical and ethical concerns given the fraught relationship between genetic ancestry and social identity. Scholars have cautioned repeatedly that genetic ancestry must be handled carefully in science and society, and a method that renders explicit labeling unnecessary sidesteps many of those pitfalls. The method’s demonstrated scalability also means it can be retrained as biobanks grow and diversify, and its dependence on individual-level data makes it well suited to trusted-access environments such as All of Us and the UK Biobank, where controlled analyses are the norm.

The path from a Nature Methods paper to routine clinical use is never short, and the authors are careful to frame SPLENDID as a tool for robust risk prediction across diverse populations rather than a finished clinical product. Validation for additional traits, disease endpoints and health systems will be needed, and continued investment in diversifying genomic datasets remains essential, because no statistical method can fully compensate for reference data that under-represent much of humanity. Still, the results reported across more than half a million participants in two flagship biobanks represent a meaningful advance: evidence that acknowledging the continuous, blended nature of human genetic ancestry, rather than forcing it into categories, yields measurably better predictions for the people who need them most. In a field where each incremental gain in accuracy can translate into earlier screening and better prevention, SPLENDID’s continuum-based approach offers a template for building genomic risk tools that work equitably across the full breadth of human diversity.

Subject of Research: A penalized regression framework, SPLENDID, that treats genetic ancestry as a continuum to improve polygenic risk prediction across diverse biobank populations.

Article Title: SPLENDID incorporates continuous genetic ancestry in biobank-scale data to improve polygenic risk prediction across diverse populations

Article References: Chen, T., Zhang, H., Mazumder, R., & Lin, X. (2026). SPLENDID incorporates continuous genetic ancestry in biobank-scale data to improve polygenic risk prediction across diverse populations. Nature Methods. https://doi.org/10.1038/s41592-026-03235-2

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03235-2

Keywords: polygenic risk scores, genetic ancestry, SPLENDID, biobank-scale data, All of Us Research Program, UK Biobank, penalized regression, health disparities, admixed populations, Nature Methods, precision medicine, statistical genetics

The research introduces an integrated set of computational and experimental tools—Spl-ISO-Seq2, Spl-IsoQuant-2 and Spl-IsoFind—that together enable spatial isoform sequencing at single-cell resolution. The methods were demonstrated on mouse brain slices, where they mapped isoform expression across the intricate architecture of neural tissue. Rather than treating tissue as a homogenized soup, the approach preserves the geographic context of each transcript, allowing researchers to ask not just which isoforms exist, but where in the tissue they are produced and by which cells.

Single-cell transcriptomics has transformed biology over the past decade, but most existing platforms read only the ends of RNA molecules, capturing short fragments that are often insufficient to distinguish one isoform from another. Full-length isoform sequencing, by contrast, reads entire transcript molecules, revealing the complete exon composition of each RNA. Combining that depth with spatial information has been a formidable technical challenge: full-length reads are harder to obtain, and preserving spatial coordinates while sequencing long molecules demands sophisticated molecular barcoding and computational reconstruction.

The new pipeline addresses these challenges at several stages. Spl-ISO-Seq2 performs the isoform sequencing itself, generating long-read transcript data from spatially preserved samples. Spl-IsoQuant-2 then processes the resulting data, calling molecular barcodes at a scale the researchers describe as exceeding 450 million barcodes. This barcode-calling step is what allows individual transcript molecules to be assigned to their original cellular locations, effectively giving each sequenced RNA a return address within the tissue. Finally, Spl-IsoFind scans the spatially resolved isoform data to detect isoforms whose abundance varies significantly across locations, flagging candidates for biological follow-up.

The scale of the barcode calling deserves emphasis. In spatial transcriptomics experiments, each RNA molecule is tagged with a barcode that identifies its position in the tissue. With hundreds of millions of barcodes to process, the computational burden is enormous, and errors in barcode assignment can blur the spatial picture. Spl-IsoQuant-2 was designed to handle this volume with high accuracy, ensuring that the spatial resolution promised by the experimental design is actually delivered in the final data. Without robust barcode calling, the entire enterprise would collapse into noisy averages that obscure the very cell-to-cell differences the method is meant to reveal.

Applied to mouse brain slices, the workflow uncovered cell-type-specific spatial isoform variability across multiple brain cell types. In other words, the isoform repertoire of a given cell type was not uniform throughout the tissue: neurons of the same class, situated in different anatomical regions, showed measurable differences in how they spliced their transcripts. This finding suggests that local environmental cues—signals from neighboring cells, positional gradients, or regional circuit context—may influence RNA processing, adding a spatial dimension to the well-known cell-type specificity of splicing.

The implications reach into several areas of neuroscience. Many neurological and psychiatric conditions, including autism spectrum disorders, amyotrophic lateral sclerosis and certain dementias, involve disruptions of RNA splicing. If isoform usage varies with position even in healthy tissue, then disease models and gene expression atlases may need to account for spatial context when interpreting splicing changes. A therapy or biomarker developed from bulk measurements could miss isoform shifts confined to specific brain regions or rare cell populations.

The technical achievement also complements other frontiers in spatial biology. Spatial proteomics maps where proteins reside, while spatial genomics locates chromosomal regions within the nucleus. Isoform-resolved spatial transcriptomics fills a gap between these layers, connecting genomic sequence to the precise protein-coding messages that cells actually deploy at particular sites. Because alternative splicing frequently changes protein domains—altering binding sites, localization signals or interaction partners—knowing which isoform is expressed where provides a more direct link between genome and function than gene-level expression alone.

The researchers position the three tools as a modular pipeline, each component addressing a distinct bottleneck. Spl-ISO-Seq2 handles the wet-lab and sequencing stage, Spl-IsoQuant-2 manages the massive barcode processing and quantification, and Spl-IsoFind performs the statistical search for spatially variable isoforms. This modularity means that laboratories already invested in spatial transcriptomics could potentially adopt individual components, and that future improvements in long-read sequencing or tissue preparation could be slotted into the pipeline without redesigning the whole system.

As with any emerging method, questions remain about how broadly the findings will generalize. The current demonstration focuses on mouse brain, a tissue whose cellular diversity and anatomical organization make it a demanding but rewarding test case. Extending the approach to human tissue, developmental stages, or disease models will be the natural next steps, and the authors’ demonstration of high spatial resolution on brain slices suggests the framework is built to scale. What the study establishes now is a proof of principle: that the isoform landscape of the brain can be read molecule by molecule, cell by cell, and place by place. In doing so, it opens a new window onto the molecular geography of the nervous system, one in which the same gene tells different stories in different neighborhoods of the brain.

Subject of Research: Spatial isoform sequencing at single-cell resolution to map cell-type-specific RNA isoform variability in mouse brain

Article Title: Spatial isoform sequencing at single-cell resolution reveals cell-type-specific spatial isoform variability in multiple brain cell types

Article References: Michielsen, L., Prjibelski, A. D., Foord, C., Spiegelman, Y., Kim, T., Hu, W., Jarroux, J., Hsu, J., Pfeil, R., Zhang, X., Gan, L., Tomescu, A. I., Hajirasouliha, I., & Tilgner, H. U. (2026). Spatial isoform sequencing at single-cell resolution reveals cell-type-specific spatial isoform variability in multiple brain cell types. Nature Methods, 23(9), 1869-1881. https://doi.org/10.1038/s41592-026-03211-w

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03211-w

Keywords: spatial transcriptomics, isoform sequencing, alternative splicing, single-cell resolution, mouse brain, RNA isoforms, Spl-ISO-Seq2, Spl-IsoQuant-2, Spl-IsoFind, molecular barcoding, neuroscience, Nature Methods