Metastasis is the leading cause of cancer-related mortality, yet most predictive models rely on shallow architectures and neglect patient-specific regulatory mechanisms. Here, we integrate classical machine learning and deep learning to predict metastatic potential across multiple cancer types. Gene expression profiles from the Cancer Cell Line Encyclopedia were combined with a transcription factor-target prior from DoRothEA, focusing on nine metastasis-associated regulators. After selecting differential genes using the Kruskal-Wallis test, ElasticNet, Random Forest, and XGBoost models were trained for benchmarking. Personalized gene regulatory networks were then constructed using PANDA and LIONESS and analyzed through a graph attention neural network (GATv2) to learn topological and expression-based representations. While XGBoost achieved the highest AUROC (0.7051), the GNN captured non-linear regulatory dependencies at the patient level. These results demonstrate that combining traditional machine learning with graph-based deep learning enables a scalable and interpretable framework for metastasis risk prediction in precision oncology.

The identification of homologous gene families across multiple genomes is a central task in bacterial pangenomics traditionally requiring computationally demanding all-against-all comparisons. PanDelos addresses this challenge with an alignment-free and parameter-free approach based on k-mer profiles, combining high speed, ease of use, and competitive accuracy with state-of-the-art methods. However, the increasing availability of genomic data requires tools that can scale efficiently to larger datasets. To address this need, we present PanDelos-plus, a fully parallel, gene-centric redesign of PanDelos. The algorithm parallelizes the most computationally intensive phases (Best Hit detection and Bidirectional Best Hit extraction) through data decomposition and a thread pool strategy, while employing lightweight data structures to reduce memory usage. Benchmarks on synthetic datasets show that PanDelos-plus achieves up to 14x faster execution and reduces memory usage by up to 96%, while maintaining accuracy. These improvements enable population-scale comparative genomics to be performed on standard multicore workstations, making large-scale bacterial pangenome analysis accessible for routine use in everyday research.

Non-invasive colorectal cancer (CRC) screening represents a key opportunity to improve colonoscopy participation rates and reduce CRC mortality. This study explores the potential of the gut-liver axis for predicting colorectal neoplasia through liver-derived radiomic features extracted from routine CT images as a novel opportunistic screening approach. In this retrospective study, we analyzed data from 1,997 patients who underwent colonoscopy and abdominal CT. Patients either had no colorectal neoplasia (n=1,189) or colorectal neoplasia (n_total=808; adenomas n=423, CRC n=385). Radiomics features were extracted from 3D liver segmentations using the Radiomics Processing ToolKit (RPTK), which performed feature extraction, filtering, and classification. The dataset was split into training (n=1,397) and test (n=600) cohorts. Five machine learning models were trained with 5-fold cross-validation on the 20 most informative features, and the best model ensemble was selected based on the validation AUROC. The best radiomics-based XGBoost model achieved a test AUROC of 0.810, clearly outperforming the best clinical-only model (test AUROC: 0.457). Subclassification between colorectal cancer and adenoma showed lower accuracy (test AUROC: 0.674). Our findings establish proof-of-concept that liver-derived radiomics from routine abdominal CT can predict colorectal neoplasia. Beyond offering a pragmatic, widely accessible adjunct to CRC screening, this approach highlights the gut-liver axis as a novel biomarker source for opportunistic screening and sparks new mechanistic hypotheses for future translational research.

DNA cytosine methylation is a critical epigenetic mark regulating gene expression and thus playing an important role in development and differentiation across eukaryotes. Existing tools for high-throughput methylation analysis often lack cross-species flexibility or require command-line expertise. We present Methylator, a novel, end-to-end DNA methylation analysis framework integrated into the Galaxy platform, enabling accessible DNA methylation analysis for mammals and plants. Methylator supports analyses from data obtained using diverse protocols like WGBS, RRBS, and PBAT and handles all contexts of DNA methylation (CpG, CHG, and CHH). The Methylator framework includes quality control, alignment, methylation calling, differential analysis, and functional analysis through reproducible, user-friendly workflows. Its unique Dirty-Harry alignment method enhances mapping efficiency, while a Shiny-based interface allows for interactive, publication-ready visualizations. Methylator is freely available, offering researchers a versatile, user-friendly solution for epigenomic studies.

Since early 2020, several theories have suggested that a distribution of restriction endonuclease recognition sites in the SARS-CoV-2 genome indicates a synthetic origin. The most influential of these, a 2022 preprint by Bruttel et al. claimed: "The BsaI/BsmBI restriction map of SARS-CoV-2 is unlike any wild-type coronavirus, and it is unlikely to evolve from its closest relatives." To test this, I reanalyzed the same 11 contested sites using an expanded set of sarbecovirus genomes, including bat coronaviruses published after the Bruttel et al. preprint. For each site, I identified the bat coronaviruses most closely matching SARS-CoV-2 in the surrounding sequences, excluding the sites themselves. The Bruttel et al. hypothesis predicts that these closely related viruses should differ from SARS-CoV-2 at many of the contested sites if restriction sites had been artificially introduced or removed. Contrary to this prediction, one or more of the most closely related bat coronaviruses are identical to SARS-CoV-2 at all 11 sites. Equivalent "synthetic fingerprints" were identified in natural pangolin sarbecoviruses. Finally, I conducted a re-analysis of the dataset that Bruttel et al. used to test where the SARS-CoV-2 BsaI/BsmBI restriction map was significantly more "evenly spaced" than expected in a natural genome. I found technical and conceptual errors that resulted in Bruttel et al. reporting that their chosen metric was 0.07% likely to occur by chance rather than 4.2%, reducing the apparent rarity 60-fold. Using a more informative metric, I tested whether restriction sites in SARS-CoV-2 or two pangolin sarbecoviruses were significantly more evenly spaced than expected and found they were not. These results show that the restriction maps of SARS-CoV-2 and related pangolin viruses are unremarkable in the context of related bat coronaviruses.

Understanding the relationship between antibody sequence, structure and function is essential for the design of antibody-based therapeutics and research tools. Recently, machine learning (ML) models mostly based on the application of large language models to sequence information have been developed to predict antibody properties. Yet there are open directions to incorporate structural information, not only to enhance prediction but also to offer insights into the underlying molecular mechanisms. This chapter provides an overview of these approaches and describes two ML frameworks that integrate structural data (via graph representations) with neural networks to predict properties of antibodies: ANTIPASTI predicts binding affinity (a global property) whereas INFUSSE predicts residue flexibility (a local property). We survey the principles underpinning these models; the ways in which they encode structural knowledge; and the strategies that can be used to extract biologically relevant statistical signals that can help discover and disentangle molecular determinants of the properties of interest.

Evolutionary systems must learn to generalize, often extrapolating from a limited set of selective conditions to anticipate future environmental changes. The mechanisms enabling such generalization remain poorly understood, despite their importance to predict ecological robustness, drug resistance, or design future-proof vaccination strategies. Here, we demonstrate that annealed population heterogeneity, wherein distinct individuals in the population experience different instances of a complex environment over time, can act as a form of implicit regularization and facilitate evolutionary generalization. Mathematically, annealed heterogeneity introduces a variance-weighted demographic noise term that penalizes across-environment fitness variance and effectively rescales the population size, thereby biasing evolution toward generalist solutions. This process is indeed analogous to a variant of the mini-batching strategy employed in stochastic gradient descent, where an effective multiplicative noise produces an inductive bias by triggering noise-induced transitions. Through numerical simulations and theoretical analysis we discuss the conditions under which variation in how individuals experience environmental selection can naturally promote evolutionary strategies that generalize across environments and anticipate novel challenges.

Hydrogen peroxide oxidises cysteine residues to control protein function, yet bulk rate constants predict hours for changes that occur in cells in seconds. Here, this work shows that local electromagnetic fields (EMFs), ubiquitous in proteins, membranes and nanodomains, can lawfully modulate the Eyring barrier and orientate reactants, accelerating cysteine oxidation without changing the underlying chemistry. Embedding a field term into the Eyring expression, demonstrated that plausible local EMFs with realistic dipole changes accelerate rate constants by orders of magnitude. This local acceleration reconciles the discrepancy between predicted vs. observed rates of H2O2-mediated cysteine oxidation. The framework generates falsifiable predictions, such as vibrational Stark readouts in thiolate peroxide complexes should fall within predicted ranges, and reframes rate-constants as mutable, field conditioned parameters. Cysteine redox sensing is fast not because the chemistry is exotic, but because the physics is local.

This paper introduces an innovative Electronic Health Record (EHR) foundation model that integrates Polygenic Risk Scores (PRS) as a foundational data modality, moving beyond traditional EHR-only approaches to build more holistic health profiles. Leveraging the extensive and diverse data from the All of Us (AoU) Research Program, this multimodal framework aims to learn complex relationships between clinical data and genetic predispositions. The methodology extends advancements in generative AI to the EHR foundation model space, enhancing predictive capabilities and interpretability. Evaluation on AoU data demonstrates the model's predictive value for the onset of various conditions, particularly Type 2 Diabetes (T2D), and illustrates the interplay between PRS and EHR data. The work also explores transfer learning for custom classification tasks, showcasing the architecture's versatility and efficiency. This approach is pivotal for unlocking new insights into disease prediction, proactive health management, risk stratification, and personalized treatment strategies, laying the groundwork for more personalized, equitable, and actionable real-world evidence generation in healthcare.

Peristalsis is the driving mechanism behind a broad array of biological and engineered flows. In peristaltic pumping, a wave-like contraction of the tube wall produces local changes in volume which induce flow. Net flow arises due to geometric nonlinearities in the momentum equation, which must be properly captured to compute the flow accurately. While most previous models focus on radius-imposed peristalsis, they often neglect longitudinal length changes - a natural consequence of radial contraction in elastic materials. In this paper, to capture a more accurate picture of peristaltic pumping, we calculate the flow in an elastic vessel undergoing contractions in the transverse and longitudinal directions simultaneously, keeping the geometric nonlinearities arising in the strain. A careful analysis requires us to study our fluid using the Lagrangian coordinates of the elastic tube. We perform analytic calculations of the flow characteristics by studying the fluid inside a fixed boundary with time-dependent metric. This mathematical manipulation works even for large-amplitude contractions, as we confirm by comparing our analytical results to COMSOL simulations. We demonstrate that transverse and longitudinal contractions induce instantaneous flows at the same order in wall strain, but in opposite directions. We investigate the influence of the wall's Poisson ratio on the flow profile. Incompressible walls suppress flow by minimizing local volume changes, whereas auxetic walls enhance flow. For radius-imposed peristaltic waves, wall incompressibility reduces both reflux and particle trapping. In contrast, length-imposed waves typically generate backflow, although trapping can still occur at large amplitudes for some Poisson ratios. These results yield a more complete description of peristalsis in elastic media and offer a framework for studying contraction-induced flows more broadly.

Boundary Vector Cells (BVCs) are a class of neurons in the brains of vertebrates that encode environmental boundaries at specific distances and allocentric directions, playing a central role in forming place fields in the hippocampus. Most computational BVC models are restricted to two-dimensional (2D) environments, making them prone to spatial ambiguities in the presence of horizontal symmetries in the environment. To address this limitation, we incorporate vertical angular sensitivity into the BVC framework, thereby enabling robust boundary detection in three dimensions, and leading to significantly more accurate spatial localization in a biologically-inspired robot model. The proposed model processes LiDAR data to capture vertical contours, thereby disambiguating locations that would be indistinguishable under a purely 2D representation. Experimental results show that in environments with minimal vertical variation, the proposed 3D model matches the performance of a 2D baseline; yet, as 3D complexity increases, it yields substantially more distinct place fields and markedly reduces spatial aliasing. These findings show that adding a vertical dimension to BVC-based localization can significantly enhance navigation and mapping in real-world 3D spaces while retaining performance parity in simpler, near-planar scenarios.

Proteins are traditionally optimized through the costly construction and measurement of many mutants. Active Learning-assisted Directed Evolution (ALDE) alleviates that cost by predicting the best improvements and iteratively testing mutants to inform predictions. However, existing ALDE methods face a critical limitation: selecting the highest-predicted mutants in each round yields homogeneous training data insufficient for accurate prediction models in subsequent rounds. Here we present FolDE, an ALDE method designed to maximize end-of-campaign success. In simulations across 20 protein targets, FolDE discovers 23% more top 10% mutants than the best baseline ALDE method (p=0.005) and is 55% more likely to find top 1% mutants. FolDE achieves this primarily through naturalness-based warm-starting, which augments limited activity measurements with protein language model outputs to improve activity prediction. We also introduce a constant-liar batch selector, which improves batch diversity; this is important in multi-mutation campaigns but had limited effect in our benchmarks. The complete workflow is freely available as open-source software, making efficient protein optimization accessible to any laboratory.

During development, embryonic tissues experience mechanical stresses ranging from cellular to supracellular length scales. In response, cells generate active forces that drive rearrangements, allowing the tissue to relax accumulated stresses. The nature of these responses depends strongly on the magnitude and duration of the deformation, giving rise to the tissue's characteristic viscoelastic behavior. Although experiments have characterized tissue rheology in various contexts, simpler theoretical approaches that directly connect cellular activity to emergent rheological behavior are still limited. In this study, we employ a vertex-based model of epithelial tissue incorporating active force fluctuations in cell vertices to represent cell motility. We capture distinct rounding dynamics and motility-dependent timescales by benchmarking against experimental observations such as the bulging of presomitic mesoderm (PSM) explants driven by Fibroblast Growth Factor(FGF) gradients. Stress relaxation tests reveal rapid short-timescale relaxation alongside persistent long-timescale residual stresses that decrease from anterior to posterior (AP) region of the PSM. By applying oscillatory shear, we analyzed the resulting elastic and viscous responses, revealing motility dependence of storage and loss modulus. Finally, we introduce spatially patterned cues applied in a temporally pulsed manner, mimicking dynamic biochemical or mechanical signals during development. Our results show that while higher motility promotes tissue remodeling in response to these cues, this response is constrained by spatial scale; cellular-scale perturbations are relaxed irrespective of motility strength, preventing complete morphological adaptation.

Artificial intelligence in medicine is built to serve the average patient. By minimizing error across large datasets, most systems deliver strong aggregate accuracy yet falter at the margins: patients with rare variants, multimorbidity, or underrepresented demographics. This average patient fallacy erodes both equity and trust. We propose a different design: a multi-agent ecosystem for N-of-1 decision support. In this environment, agents clustered by organ systems, patient populations, and analytic modalities draw on a shared library of models and evidence synthesis tools. Their results converge in a coordination layer that weighs reliability, uncertainty, and data density before presenting the clinician with a decision-support packet: risk estimates bounded by confidence ranges, outlier flags, and linked evidence. Validation shifts from population averages to individual reliability, measured by error in low-density regions, calibration in the small, and risk--coverage trade-offs. Anticipated challenges include computational demands, automation bias, and regulatory fit, addressed through caching strategies, consensus checks, and adaptive trial frameworks. By moving from monolithic models to orchestrated intelligence, this approach seeks to align medical AI with the first principle of medicine: care that is transparent, equitable, and centered on the individual.

We ask where, and under what conditions, dyslexic reading costs arise in a large-scale naturalistic reading dataset. Using eye-tracking aligned to word-level features (word length, frequency, and predictability), we model how each feature influences dyslexic time costs. We find that all three features robustly change reading times in both typical and dyslexic readers, and that dyslexic readers show stronger sensitivities to each, especially predictability. Counterfactual manipulations of these features substantially narrow the dyslexic-control gap by about one third, with predictability showing the strongest effect, followed by length and frequency. These patterns align with dyslexia theories that posit heightened demands on linguistic working memory and phonological encoding, and they motivate further work on lexical complexity and parafoveal preview benefits to explain the remaining gap. In short, we quantify when extra dyslexic costs arise, how large they are, and offer actionable guidance for interventions and computational models for dyslexics.

Accurately predicting the three-dimensional structures of protein-ligand complexes remains a fundamental challenge in computational drug discovery that limits the pace and success of therapeutic design. Deep learning methods have recently shown strong potential as structural prediction tools, achieving promising accuracy across diverse biomolecular systems. However, their performance and utility are constrained by scarce experimental data, inefficient architectures, physically invalid poses, and the limited ability to exploit auxiliary information available at inference. To address these issues, we introduce Pearl (Placing Every Atom in the Right Location), a foundation model for protein-ligand cofolding at scale. Pearl addresses these challenges with three key innovations: (1) training recipes that include large-scale synthetic data to overcome data scarcity; (2) architectures that incorporate an SO(3)-equivariant diffusion module to inherently respect 3D rotational symmetries, improving generalization and sample efficiency, and (3) controllable inference, including a generalized multi-chain templating system supporting both protein and non-polymeric components as well as dual unconditional/conditional modes. Pearl establishes a new state-of-the-art performance in protein-ligand cofolding. On the key metric of generating accurate (RMSD < 2 Å) and physically valid poses, Pearl surpasses AlphaFold 3 and other open source baselines on the public Runs N' Poses and PoseBusters benchmarks, delivering 14.5% and 14.2% improvements, respectively, over the next best model. In the pocket-conditional cofolding regime, Pearl delivers $3.6\times$ improvement on a proprietary set of challenging, real-world drug targets at the more rigorous RMSD < 1 Å threshold. Finally, we demonstrate that model performance correlates directly with synthetic dataset size used in training.

Object binding, the brain's ability to bind the many features that collectively represent an object into a coherent whole, is central to human cognition. It groups low-level perceptual features into high-level object representations, stores those objects efficiently and compositionally in memory, and supports human reasoning about individual object instances. While prior work often imposes object-centric attention (e.g., Slot Attention) explicitly to probe these benefits, it remains unclear whether this ability naturally emerges in pre-trained Vision Transformers (ViTs). Intuitively, they could: recognizing which patches belong to the same object should be useful for downstream prediction and thus guide attention. Motivated by the quadratic nature of self-attention, we hypothesize that ViTs represent whether two patches belong to the same object, a property we term IsSameObject. We decode IsSameObject from patch embeddings across ViT layers using a similarity probe, which reaches over 90% accuracy. Crucially, this object-binding capability emerges reliably in self-supervised ViTs (DINO, MAE, CLIP), but markedly weaker in ImageNet-supervised models, suggesting that binding is not a trivial architectural artifact, but an ability acquired through specific pretraining objectives. We further discover that IsSameObject is encoded in a low-dimensional subspace on top of object features, and that this signal actively guides attention. Ablating IsSameObject from model activations degrades downstream performance and works against the learning objective, implying that emergent object binding naturally serves the pretraining objective. Our findings challenge the view that ViTs lack object binding and highlight how symbolic knowledge of "which parts belong together" emerges naturally in a connectionist system.

There are two approaches to describing DNA-ions interactions. The physical approach is an analysis of electrostatic interactions between ions and charges on the DNA molecule. The coordination chemistry approach is a search for modes of direct binding of ions to ionophores of DNA. We study both the inner and outer sphere coordination of ions by ionophores of the A and C forms of DNA in molecular dynamics simulations in two low-polarity solvents: in ethanol-water and methanol-water mixtures. We show that the counterion-DNA outer sphere coordination plays a key role in the experimentally observed conformational polymorphism of the DNA molecule: a transition to the A form in ethanol and to the C form in methanol. We identify the ionophores responsible for the existence of the A- and C-complexes. In both the complexes, the ions' inner sphere ligands are mostly water molecules, the ions reside in water clusters. In ethanol-water mixture, the water clusters are large, the major groove of the A-DNA is filled with water, and all ionophores are accessible to ions. In methanol-water mixture, the water clusters are small, and a large number of methanol clusters are present near DNA surface. They interfere with the coordination of ions in one of the ionophores of the major groove, and also with other ionophores near phosphates. Therefore, in methanol, the interaction energy of counterions with A-DNA cannot compensate for the repulsion between closely located phosphates. Therefore, the ions fill more accessible ionophores of the C-complex, converting DNA into the C form.

Understanding gene perturbation effects across diverse cellular contexts is a central challenge in functional genomics, with important implications for therapeutic discovery and precision medicine. Single-cell technologies enable high-resolution measurement of transcriptional responses, but collecting such data is costly and time-consuming, especially when repeated for each cell type. Existing computational methods often require separate models per cell type, limiting scalability and generalization. We present CFM-GP, a method for cell type-agnostic gene perturbation prediction. CFM-GP learns a continuous, time-dependent transformation between unperturbed and perturbed gene expression distributions, conditioned on cell type, allowing a single model to predict across all cell types. Unlike prior approaches that use discrete modeling, CFM-GP employs a flow matching objective to capture perturbation dynamics in a scalable manner. We evaluate on five datasets: SARS-CoV-2 infection, IFN-beta stimulated PBMCs, glioblastoma treated with Panobinostat, lupus under IFN-beta stimulation, and Statefate progenitor fate mapping. CFM-GP consistently outperforms state-of-the-art baselines in R-squared and Spearman correlation, and pathway enrichment analysis confirms recovery of key biological pathways. These results demonstrate the robustness and biological fidelity of CFM-GP as a scalable solution for cross-cell type gene perturbation prediction.

Biomolecular interactions underpin almost all biological processes, and their rational design is central to programming new biological functions. Generative AI models have emerged as powerful tools for molecular design, yet most remain specialized for individual molecular types and lack fine-grained control over interaction details. Here we present ODesign, an all-atom generative world model for all-to-all biomolecular interaction design. ODesign allows scientists to specify epitopes on arbitrary targets and generate diverse classes of binding partners with fine-grained control. Across entity-, token-, and atom-level benchmarks in the protein modality, ODesign demonstrates superior controllability and performance to modality-specific baselines. Extending beyond proteins, it generalizes to nucleic acid and small-molecule design, enabling interaction types such as protein-binding RNA/DNA and RNA/DNA-binding ligands that were previously inaccessible. By unifying multimodal biomolecular interactions within a single generative framework, ODesign moves toward a general-purpose molecular world model capable of programmable design. ODesign is available at this https URL ,

We consider a population with two types of individuals, distinguished by the resources required for reproduction: type-$0$ (small) individuals need a fractional resource unit of size $\vartheta \in (0,1)$, while type-$1$ (large) individuals require $1$ unit. The total available resource per generation is $R$. To form a new generation, individuals are sampled one by one, and if enough resources remain, they reproduce, adding their offspring to the next generation. The probability of sampling an individual whose offspring is small is $\rho_{R}(x)$, where $x$ is the proportion of small individuals in the current generation. We call this discrete-time stochastic model a two-size Wright--Fisher model, where the function $\rho_{R}$ can represent mutation and/or frequency-dependent selection. We show that on the evolutionary time scale, i.e. accelerating time by a factor $R$, the frequency process of type-$0$ individuals converges to the solution of a Wright--Fisher-type SDE. The drift term of that SDE accounts for the bias introduced by the function $\rho_{R}$ and the consumption strategy, the latter also inducing an additional multiplicative factor in the diffusion term. To prove this, the dynamics within each generation are viewed as a renewal process, with the population size corresponding to the first passage time $\tau(R)$ above level $R$. The proof relies on methods from renewal theory, in particular a uniform version of Blackwell's renewal theorem for binary, non-arithmetic random variables, established via $\varepsilon$-coupling.

Humans leverage rich internal models of the world to reason about the future, imagine counterfactuals, and adapt flexibly to new situations. In Reinforcement Learning (RL), world models aim to capture how the environment evolves in response to the agent's actions, facilitating planning and generalization. However, typical world models directly operate on the environment variables (e.g. pixels, physical attributes), which can make their training slow and cumbersome; instead, it may be advantageous to rely on high-level latent dimensions that capture relevant multimodal variables. Global Workspace (GW) Theory offers a cognitive framework for multimodal integration and information broadcasting in the brain, and recent studies have begun to introduce efficient deep learning implementations of GW. Here, we evaluate the capabilities of an RL system combining GW with a world model. We compare our GW-Dreamer with various versions of the standard PPO and the original Dreamer algorithms. We show that performing the dreaming process (i.e., mental simulation) inside the GW latent space allows for training with fewer environment steps. As an additional emergent property, the resulting model (but not its comparison baselines) displays strong robustness to the absence of one of its observation modalities (images or simulation attributes). We conclude that the combination of GW with World Models holds great potential for improving decision-making in RL agents.

Deep learning techniques have driven significant progress in various analytical tasks within 3D genomics in computational biology. However, a holistic understanding of 3D genomics knowledge remains underexplored. Here, we propose MIX-HIC, the first multimodal foundation model of 3D genome that integrates both 3D genome structure and epigenomic tracks, which obtains unified and comprehensive semantics. For accurate heterogeneous semantic fusion, we design the cross-modal interaction and mapping blocks for robust unified representation, yielding the accurate aggregation of 3D genome knowledge. Besides, we introduce the first large-scale dataset comprising over 1 million pairwise samples of Hi-C contact maps and epigenomic tracks for high-quality pre-training, enabling the exploration of functional implications in 3D genomics. Extensive experiments show that MIX-HIC can significantly surpass existing state-of-the-art methods in diverse downstream tasks. This work provides a valuable resource for advancing 3D genomics research.

We study a fast-slow version of the Bazykin-Berezovskaya predator-prey model with Allee effect evolving on two timescales, through the lenses of Geometric Singular Perturbation Theory (GSPT). The system we consider is in non-standard form. We completely characterize its dynamics, providing explicit threshold quantities to distinguish between a rich variety of possible asymptotic behaviors. Moreover, we propose numerical results to illustrate our findings. Lastly, we comment on the real-world interpretation of these results, in an economic framework and in the context of predator-prey models.

Large language models (LLMs) have revolutionized natural language processing and are increasingly applied to other sequential data types, including genetic sequences. However, adapting LLMs to genomics presents significant challenges. Capturing complex genomic interactions requires modeling long-range dependencies within DNA sequences, where interactions often span over 10,000 base pairs, even within a single gene, posing substantial computational burdens under conventional model architectures and training paradigms. Moreover, standard LLM training approaches are suboptimal for DNA: autoregressive training, while efficient, supports only unidirectional understanding. However, DNA is inherently bidirectional, e.g., bidirectional promoters regulate transcription in both directions and account for nearly 11% of human gene expression. Masked language models (MLMs) allow bidirectional understanding but are inefficient, as only masked tokens contribute to the loss per step. To address these limitations, we introduce JanusDNA, the first bidirectional DNA foundation model built upon a novel pretraining paradigm that combines the optimization efficiency of autoregressive modeling with the bidirectional comprehension of masked modeling. JanusDNA adopts a hybrid Mamba, Attention and Mixture of Experts (MoE) architecture, combining long-range modeling of Attention with efficient sequential learning of Mamba. MoE layers further scale model capacity via sparse activation while keeping computational cost low. Notably, JanusDNA processes up to 1 million base pairs at single nucleotide resolution on a single 80GB GPU. Extensive experiments and ablations show JanusDNA achieves new SOTA results on three genomic representation benchmarks, outperforming models with 250x more activated parameters. Code: this https URL

Characterizing the cellular properties of neurons is fundamental to understanding their function in the brain. In this quest, the generation of bio-realistic models is central towards integrating multimodal cellular data sets and establishing causal relationships. However, current modeling approaches remain constrained by the limited availability and intrinsic variability of experimental neuronal data. The deterministic formalism of bio-realistic models currently precludes accounting for the natural variability observed experimentally. While deep learning is becoming increasingly relevant in this space, it fails to capture the full biophysical complexity of neurons, their nonlinear voltage dynamics, and variability. To address these shortcomings, we introduce NOBLE, a neural operator framework that learns a mapping from a continuous frequency-modulated embedding of interpretable neuron features to the somatic voltage response induced by current injection. Trained on synthetic data generated from bio-realistic neuron models, NOBLE predicts distributions of neural dynamics accounting for the intrinsic experimental variability. Unlike conventional bio-realistic neuron models, interpolating within the embedding space offers models whose dynamics are consistent with experimentally observed responses. NOBLE enables the efficient generation of synthetic neurons that closely resemble experimental data and exhibit trial-to-trial variability, offering a $4200\times$ speedup over the numerical solver. NOBLE is the first scaled-up deep learning framework that validates its generalization with real experimental data. To this end, NOBLE captures fundamental neural properties in a unique and emergent manner that opens the door to a better understanding of cellular composition and computations, neuromorphic architectures, large-scale brain circuits, and general neuroAI applications.

Conventional colorimetric sensing methods typically rely on signal intensity at a single wavelength, often selected heuristically based on peak visual modulation. This approach overlooks the structured information embedded in full-spectrum transmission profiles, particularly in intensity-based systems where linear models may be highly effective. In this study, we experimentally demonstrate that applying a forward feature selection strategy to normalized transmission spectra, combined with linear regression and ten-fold cross-validation, yields significant improvements in predictive accuracy. Using food dye dilutions as a model system, the mean squared error was reduced from over 22,000 with a single wavelength to 3.87 using twelve selected features, corresponding to a more than 5,700-fold enhancement. These results validate that full-spectrum modeling enables precise concentration prediction without requiring changes to the sensing hardware. The approach is broadly applicable to colorimetric assays used in medical diagnostics, environmental monitoring, and industrial analysis, offering a scalable pathway to improve sensitivity and reliability in existing platforms.