Accurate interpretation of panoramic dental radiographs requires the integration of multiple reasoning capabilities: detection, spatial localization, and quantitative assessment. Despite recent advances in multimodal learning, existing medical visual question answering (VQA) benchmarks do not fully capture this complexity, often reducing the task to simplified classification or templated queries. As a result, they provide limited coverage of the diverse reasoning processes required for clinically meaningful interpretation. We introduce DentiAsk, a large-scale dental VQA benchmark that pairs high-resolution panoramic dental radiographs with clinician-validated question-answer pairs spanning three reasoning tiers: descriptive recognition, spatial localization, and numerical quantification across three high-prevalence pathologies: periapical radiolucency (PARL), impacted teeth, and dental caries. DentiAsk comprises 1,000 high-resolution radiographs annotated with 10,000 expert-curated QA pairs. To our knowledge, it is the first dental VQA benchmark to unify categorical, spatial, and quantitative reasoning as separately scored tasks within a single evaluation framework. We benchmark 10 state-of-the-art vision-language models, including LLaVA-v1.5, LLaVA-v1.6, Qwen-VL, InternVL2, and LLaVA-Med, and find that models achieve stronger performance on descriptive queries, whereas they degrade sharply on spatial localization and counting, exposing limitations in compositional, multi-step reasoning. These findings reveal a gap between visual recognition and clinically meaningful reasoning, establishing DentiAsk as a challenging benchmark for advancing multimodal reasoning in medical imaging.

Perfusion imaging guides clinical evaluation of stroke and brain tumors by characterizing tissue-level hemodynamics. Routine quantification relies on manual arterial input function (AIF) selection followed by deconvolution, producing summary maps without an executable temporal model for simulation or mechanistic insight. Tracer-dynamics-based models infer transport or compartmental parameters from perfusion time series, but do not yield clinically actionable perfusion indices (e.g., CBF, CBV, MTT) that inform diagnosis and treatment decisions. In this work, we propose HemoPIC, a physics-informed cerebral hemodynamics digital twin that explains perfusion time series through tracer mass conservation and a lumped parameter hemodynamic model. Specifically, HemoPIC solves a constrained inverse problem that jointly estimates digital twin parameters and latent states from perfusion imaging, eliminating manual AIF selection and deconvolution from routine perfusion quantification while directly producing clinically actionable perfusion summary maps. Experiments demonstrate that HemoPIC reconstructs tracer dynamics, generates physiologically consistent perfusion maps with lesion hypoperfusion patterns, satisfies central volume consistency, and yields a mechanistic hemodynamic digital twin that enables forward simulation and counterfactual intervention analysis. Code is publicly available at this https URL.

The push toward large language models for biology (BioLM) has created a need for training corpora that can endow models with a genuine understanding of biology. However, existing biological resources, such as molecular databases, protein repositories, genomic annotations, single-cell atlases, and pathway databases, are scattered across heterogeneous formats and remain unorganized into a cohesive corpus for language model training. We present TheBioCollection, a 52.6B-token pre-training-scale corpus that converts these disparate resources into a unified, training-ready form spanning small molecules, proteins, genomic sequences, cells, and pathways. Beyond consolidating existing data, TheBioCollection enriches each record with tool-computed biological properties and introduces new instruction tasks for capabilities that current corpora barely cover. We pair the corpus with TheBioCollection-Eval, a matched suite probing recognition, generation, and prediction across molecular, protein, genomic, cellular, and cross-domain settings. Holding the base Gravity-16B-A3B architecture fixed, training on TheBioCollection more than doubles its overall score on TheBioCollection-Eval with gains in every domain, while leaving general linguistic ability nearly intact.

The functional organization of the brain relies on coordinated activity across spatially distributed regions, making the analysis of inter-regional dependencies fundamental. Existing connectivity measures address this predominantly through phase synchronization, which is vulnerable to volume conduction artifacts and discards amplitude-domain coupling. This study introduces the Spatial Neighboring Scattering Transform, which extends the wavelet scattering transform to the multichannel setting, yielding two descriptors that jointly capture amplitude-envelope coupling between channels and its modulation across frequency scales. SNST was evaluated on the BCI Competition IV-2a motor imagery dataset using a bias-corrected, false-discovery-rate-controlled statistical pipeline, with the validation criterion defined as spatial consistency of significant coupling across subjects. The first-order descriptor identified statistically significant amplitude coupling within a central-parietal electrode neighborhood, reproduced consistently across all subjects and both imagery conditions. The second-order descriptor revealed that this coupling is periodically gated by slow rhythms, indicating a cross-frequency amplitude-modulation structure absent from single-frequency connectivity measures. Phase lag index and weighted phase lag index, computed under an identical correction procedure and verified robust to volume conduction, identified negligible significant coupling with zero overlap with SNST findings, demonstrating that amplitude envelope coupling constitutes a largely distinct connectivity signal. These results establish SNST as a cross-channel scattering-based connectivity descriptor that recovers amplitude-envelope and cross-frequency coupling structure systematically, applicable to any multichannel EEG analysis where amplitude-domain inter-regional dependence is of interest.

Biological measurements are commonly assumed to approximate Gaussian distributions, and normality is routinely assessed as a prerequisite for statistical analysis. However, whether the degree of Gaussianity itself contains biological information remains largely unexplored. Here, we quantified the Gaussianity of biological measurements using the root mean square error of normal quantile-quantile plots (QQ-RMSE). A reference distribution was constructed from 10,249 biological measurements from the National Health and Nutrition Examination Survey (NHANES) 1999-2023, enabling direct comparison of the Gaussian phenotype, defined as the degree to which a biological measurement approximates a Gaussian distribution. Biological measurements exhibited characteristic Gaussian phenotypes. Structural and capacity-related traits, including body measurements, grip strength, spirometry, and red blood cell count, consistently showed low QQ-RMSE values. Homeostatically regulated variables, such as total cholesterol, also exhibited high Gaussianity. In contrast, biomarkers associated with physiological responses or pathology, including triglycerides, C-reactive protein, liver enzymes, serum creatinine, and urinary albumin, showed progressively larger deviations from Gaussianity. Biological normalization further improved Gaussianity: the albumin-to-creatinine ratio consistently exhibited lower QQ-RMSE values than urinary albumin alone across all NHANES survey cycles. These findings indicate that Gaussianity is not merely a statistical assumption but a measurable biological property. We propose the concept of the Gaussian phenotype, in which the degree of Gaussianity reflects biological mechanisms governing variability. This study establishes the first reference atlas of Gaussianity for interpreting biological measurements.

A model predictive controller (MPC) is used to construct a virtual assistant to aid a physician in prescribing cortisol replacement therapy for patients with adrenal insufficiency (AI). AI, also known as hypocortisolism, is a condition that occurs due to a low concentration of cortisol. This hormonal imbalance significantly impacts the individual's ability to regulate stress, metabolism, and immune responses. Thus, it is essential to maintain cortisol levels within a healthy range. The production of cortisol is governed by the hypothalamus-pituitary-adrenal (HPA) axis, a part of the endocrine system. In this paper, a novel mathematical model of the HPA axis is proposed that incorporates the endogenous circadian rhythm. This model simulates two conditions of hypocortisolism: primary and secondary AI. Adrenal insufficiency cannot be cured, but it can be treated with cortisol replacement therapy. The standard practice is to prescribe a therapeutic dose of hydrocortisone (HC). To evaluate the accuracy of the proposed HPA axis model, an open-loop cortisol replacement strategy with a fixed dosage is used to simulate both primary and secondary AI. The simulation results show that, analytically, it is possible to arrive at a fixed working cortisol replacement strategy. However, this strategy, though effective, is not optimal. To obtain optimal cortisol replacement strategies, an MPC is proposed. An important feature of MPC is that constraints on allowable cortisol replacement dosages can be rigorously addressed. This controller can serve as a virtual assistant to physicians in prescribing daily cortisol replacement therapy.

Dynamic allocation of attention across the visual field, quantified as a visuospatial attention gradient, is essential for maintaining perceptual breadth. Disruptions to this flexibility may contribute to altered spatial attentional bias and may be influenced by trait anxiety. We investigated whether individual differences in structural brain morphology predict spatial attentional deployment as a function of trait anxiety. Sixty participants, recruited based on an a priori sample size calculation, completed a visuospatial attention gradient task incorporating brief partial facial emotion cues. Although discrete emotional cues did not significantly modulate attention gradients, structural neuroimaging analyses revealed that greater grey matter volume in bilateral cerebellar lobule VI and increased cortical thickness in the left precentral gyrus and paracentral lobule were associated with reduced interaction between the magnitude of the spatial attention gradient (averaged across emotions) and trait anxiety. Machine-learning models further predicted individual attention-anxiety profiles from these neuroanatomical features. These findings suggest that greater structural integrity in cerebellar and sensorimotor regions is associated with more flexible spatial attentional deployment in individuals with lower trait anxiety. Together, the results highlight the contribution of cerebellar and sensorimotor regions, beyond their traditional motor functions, to individual differences in visual spatial attention and cognitive-affective interactions, while demonstrating the predictive utility of structural brain markers.

Chemotaxis underlies the collective migration of cell populations in developmental processes and immune responses. While the theoretical investigation of single-cell-type collective chemotaxis has received considerable attention, heterogeneous chemotaxis involving multiple interacting cell types remains poorly understood. Here, we generalise a model of heterogeneous self-generated chemotaxis and analyse the resulting collective migration patterns. We show that coupled migration between two cell types gives rise to \emph{propagating terraces} -- coupled travelling fronts moving at different speeds. While a sensor-only population leads the migrating collective, a slower mixed sensor-consumer population follows. Our analysis reveals that these fronts are coupled via the dynamics of the self-generated chemoattractant gradients. We derive analytical expressions for the migration speeds of the two fronts just in term of model parameters and experimentally measurable quantities. Our analytical results reveal that heterogeneity can enhance long-range migration via self-generated chemotaxis for sensor cells. While sensor cells can leverage benefit from mixing with consumer cells, the latter migrate more efficiently when mixing with cells of the same type. Together, our results provide a comprehensive theoretical framework for understanding heterogeneous self-generated chemotaxis.

Species-specific pest and vector control strategies, including the sterile insect technique, Wolbachia-based interventions, and genetic control technologies, offer powerful alternatives to broad-spectrum chemical control, with applications ranging from targeted crop protection to large-scale disease control. Among these, genetic control technologies are advancing rapidly, but the pace of technological development is outstripping the modelling tools needed to predict outcomes, guide technology design and its implementation, compare alternative strategies across different use settings, and support regulatory and operational decision-making. Here we present PesTwin, an agent-based modelling framework for simulating genetic control technologies across species, ecological settings, and deployment strategies within a common computational environment. PesTwin captures stochastic demographic effects, species-specific life-history traits, heterogeneous dispersal, and temporal variation in resource availability and infestation pressure. We validate PesTwin against published laboratory cage data from four genetic control systems, drawn from three studies, in two insect species, showing close agreement between predicted and observed population trajectories, including their replicate-to-replicate variability. We then illustrate how the same validated models extend beyond the cage to spatially explicit, field-scale scenarios, using PesTwin to explore how the timing, density and spatial placement of releases shape suppression and spread across heterogeneous landscapes. By making genetic control systems testable in silico before they are built or released, PesTwin can shorten the path from laboratory construct to field intervention: informing which constructs to prioritise, how to design the experiments that test them, where and when to release, and what evidence is needed to evaluate them.

Current electroencephalography (EEG)-based dream detection relies on power spectral density (PSD) and statistical moment features, achieving a state-of-the-art area under the receiver operating characteristic curve (AUC) of approximately 0.70 on the DREAM database (Wong et al., 2025, Nature Communications). We introduce PHINN-EEG (Persistent Homology Inspired Neural Network for EEG), the first topological time-series framework for dream mentation analysis. Using sliding-window Takens delay embeddings and Vietoris-Rips filtrations on multichannel pre-awakening EEG epochs, we extract Dynamic Betti Curves that characterize the geometric architecture of neural activity, not merely its energy. These topological invariants, combined with topology-conditioned flow matching, are analytically projected to outperform existing PSD and catch22 benchmarks, targeting AUC = 0.82-0.90 on the 1,462-awakening open-access subset of the DREAM database (drawn from a full registry of 3,191 total awakenings from 263 participants across 20 independent laboratories). We further introduce a topology-conditioned rectified flow model for dream-state EEG synthesis-with a spectral-conditioned flow model of comparable feature dimensionality as an additional ablation baseline to isolate the value of topological conditioning specifically-and propose a set of candidate Betti transition archetypes linking topology to phenomenological dream report categories, presented as an exploratory hypothesis space pending empirical validation. If validated, this work represents a paradigm shift from spectral energy to phase-space geometry in neural rare-event detection, with potential future implications for wearable BCI dream monitoring.

Decision making often exhibits context dependence that challenges classical probability theory. This paper develops a quantum-like extension of the Tug-of-War (QTOW) decision-making model to clarify when such context dependence can be represented by a single minimal internal state. The QTOW construction uses a qutrit internal state, conservation-preserving updates, and measurement-induced disturbance to model decision, learning, and probing operations within one coherent state space. Within this minimal representation, KCBS-type probing contexts can be constructed, yielding a witness of non-contextual classical non-embeddability. The main claim is not that quantum theory is uniquely or assumption-freely derived from decision making. Rather, a classical reconstruction of the same operation family requires additional contextual memory, history dependence, or an enlarged hidden-state representation. Thus, contextual probability appears as a resource signature of minimal decision dynamics, while quantum probability provides a compact, memory-efficient realization of this structure.

The coupling in flow matching -- the rule pairing noise vectors with data points -- is typically treated as a computational choice. We show that this coupling can instead serve as an alignment interface: by matching noise and data according to a target molecular property, it embeds controllable structure directly into the learned flow field. Building on this view, we introduce Reward Transport, which uses optimal transport coupling at training time to align a scalar noise-space coordinate with molecular rewards; at inference, varying this coordinate steers the generated distribution without requiring an oracle, reward model, gradient guidance, or additional computation. In the coupling-preserving limit, thresholding this coordinate recovers the Cross-Entropy Method's truncated reward distribution, providing a principled, continuously adjustable distribution-level control knob. Empirically, on ZINC-250K and GuacaMol, sweeping the scalar induces monotone control of logP and consistent QED control over its operating range; most tellingly, the same knob produces opposite structural responses for different targets, growing molecules for logP but shrinking them for QED, which rules out a generic size bias. The interface is complementary to classifier-free guidance and conditional flow matching, while a negative result under epsilon-prediction diffusion clarifies where coupling-level alignment is structurally absent. Code: this https URL

Quantum logic is usually presented as a non-classical departure from ordinary reasoning forced on us by quantum mechanics, with classical logic kept as the secure starting point. We argue for the opposite order of explanation in a finite and fully computable setting. The free orthomodular lattice on two generators has ninety-six elements, the direct product of a six-element non-distributive factor and a sixteen-element Boolean factor. Reading the first factor as a register of contexts and the second as Boolean content, we obtain a calculus whose elements are context--bit-vector pairs and whose operations act component by component. With this calculus we establish three results. First, we classify the six layers by commutativity, identifying the central kernel of context-neutral propositions together with a dual central layer in which all complementary contexts are present. Second, we show that orthocomplementation rearranges the layers exactly as the complementation of the small factor rearranges its elements, which makes the duality among the layers rigid rather than accidental. Third, we prove that the operation forgetting the context is a surjective homomorphism of orthocomplemented lattices whose quotient is the classical Boolean algebra, so that classical logic is a six-to-one, information-losing image of the contextual calculus.

The ability to generate variable-length proteins is crucial in protein design, where the optimal length is often unknown and tightly coupled to designability. Current diffusion- and flow-based generative models typically require the protein length to be specified before sampling, limiting their flexibility in exploring the feasible design space. To address this limitation, we introduce Generalized Poisson Flow (GPFlow), a variable-length generative framework that learns the rate function of an inhomogeneous generalized Poisson process by minimizing its negative log-likelihood. We establish population-level guarantees for recovering the joint multimodal distribution and derive an upper bound on the KL divergence between the data and generated distributions. We comprehensively evaluate GPFlow across structure and sequence design, motif scaffolding, and peptide co-design, spanning Euclidean, categorical, and Riemannian modalities to fully validate its variable-length generation quality. In unconditional design, GPFlow improves structural designability and achieves the best distributional fitness for sequence design compared to their corresponding fixed-length baselines, while perfectly recovering the length distribution. In conditional motif scaffolding, GPFlow ranks first on 10 of 16 structure-based design tasks with significantly more unique successes and also achieves more passed tasks in sequence-based design. In peptide co-design, GPFlow remains competitive even without access to a native-length oracle.

The Susceptible-Exposed-Infectious-Removed (SEIR) model is a fundamental model in epidemiology. Model parameters such as the reciprocal transmission, incubation, and infectious rates are often difficult to measure directly, and they are estimated by solving an optimisation problem aiming to minimise the difference between the observed data and the model solution. However, the parameters of the standard SEIR system are not globally identifiable, causing optimisation algorithms to frequently converge to incorrect local optima and suffer from numerical stiffness. Here we show a comprehensive structural identifiability analysis of the SEIR framework, and present a globally identifiable and computationally stable reparameterisation of the model derived via an observational system approach. We fully characterise the multiple locally identifiable parameters, and by transforming the system into a globally identifiable structure, we eliminate the non-uniqueness issues in the parameter estimation approaches. Our numerical experiments demonstrate that this reformulation significantly improves convergence frequency, avoids runtime errors caused by numerical overflow, and consistently recovers the correct parameters. Furthermore, incorporating first-order sensitivity equations into the optimiser enhances the robustness and execution speed of the estimation process. Numerically well-conditioned methods for parameter identification, together with a comprehensive understanding of the identifiability of the parameters, ensure that the model yields reliable, rigorous insights for infectious disease forecasting and theoretical epidemiology.

Understanding the dynamical properties of coupled phase oscillator systems with heterogeneous oscillator frequencies has been a long-standing challenge of complex systems theory. While the seminal work of Ott and Antonsen dramatically improved our theoretical understanding of coupled phase oscillators for a small family of oscillator frequency distributions, we here present a mean-field reduction method for arbitrary frequency distributions. Our method leverages the drastic dimensionality reduction obtained for Lorentzian frequency distributions, and combines it with a data-driven multi-ensemble approach. As such, the method renders the Ott-Antonsen equations directly applicable to empirical distributions of phase oscillator frequencies, often achieving a drastic dimensionality reduction and allowing to study real-world physical and biological systems by means of stability, sensitivity, and bifurcation analyses.

Self-supervised pretrained foundation models (FM) have shown early promise for non-invasive electroencephalogram (EEG) decoding applications. Many recent large-scale models converged on the approach of tokenizing raw EEG followed by masked reconstruction pretraining. However, this recipe has been shown to be suboptimal for data, like EEG, with high noise amplitude and information confined to limited dimensions such as narrow frequency bands. Building on this insight, we develop a novel contrastive-pretrained EEG model with multiscale temporal convolution input layers and Transformer encoder blocks (CoCoT). CoCoT matches or beats state-of-the-art reconstruction-pretrained EEG models on extensive benchmark decoding tasks with heterogeneous electrode configurations. Furthermore, CoCoT trained from scratch outperforms previous single-task decoding models and even rivals pretrained models, showcasing the architecture's flexibility and data efficiency. Through systematic ablations, including model architecture and pretraining objective, we demonstrate the viability of contrastive learning for building EEG FMs while suggesting key architectural design considerations, prompting further investigations in alternative large-scale pretraining strategies.

The extracellular polymeric substances (EPS) matrix of microbial biofilms exhibits a complex structural heterogeneity that profoundly influences mass transport and metabolic activity. Conventional biofilm models typically assume a homogeneous matrix, thereby neglecting the localized transport resistance introduced by the bacterial capsule, a distinct, low-diffusivity polysaccharide layer surrounding individual cells. In this theoretical study, we develop a multiscale "cell-capsule" continuum model that represents the capsule as a concentric shell enveloping each microbial cell core within the bulk EPS. Utilizing a one-dimensional reaction-diffusion framework coupled with a geometric characterization of capsule spacing and thickness, we quantify how microscale architecture modulates oxygen transport in developing biofilms. Model simulations demonstrate that incorporating a discrete capsular phase introduces a pronounced "resistance-in-series" effect, reducing local oxygen availability by up to 70% compared to conventional homogeneous models. Furthermore, our analysis indicates that capsule thickness and matrix compaction jointly control the effective diffusivity and oxygen effectiveness factor within the biofilm. These results provide critical mechanistic insights into how microscale organization governs macroscale biofilm function, offering a new framework for integrating structural heterogeneity into multiscale biofilm simulations.

The statistics of correlations are central quantities characterizing the collective dynamics of recurrent neural networks. We derive exact expressions for the statistics of correlations of nonlinear recurrent networks in the limit of a large number N of neurons, including systematic 1/N corrections, in the regime of Gaussian quenched disorder. Our approach uses a path-integral representation of the network stochastic dynamics, which reduces the description to a few collective variables and enables efficient computation. This generalizes previous results on linear networks to include a wide family of nonlinear activation functions, which enter as interaction terms in the path integral. These interactions can resolve the instability of the linear theory and yield a strictly positive participation dimension. We present explicit results for power-law activations, revealing scaling behavior controlled by the network coupling. In addition, we introduce a class of activation functions based on Pade approximants and provide analytic predictions for their correlation statistics. Numerical simulations confirm our theoretical results with excellent agreement. We also compare with previous works that have studied the complementary case with annealed disorder, and based on this we propose a new self-consistent equation for the more general case of colored noise.

We formalize human choice behavior in a probabilistic hide-and-seek task. In our geometric construction, vectors represent participant choice frequencies as well as probability matching and maximizing strategies. We measured choice behavior not just in the well-studied scenario of pursuing an objective (seeking), but also the rarely studied scenario of avoiding consequences (hiding). We used our geometric construction to define the avoidance counterpart of probability matching, probability antimatching, as a vector reflection across the uniform distribution. Decomposing the behavior of participants when they were seeking into matching and maximizing components, we could mathematically derive the analogous antimatching and minimizing strategies for hiding. Participants did change their choice frequencies between hiding and seeking conditions. In both cases, we found that a linear combination of just two vectors did an excellent job of fitting participant choice frequencies: matching + maximizing for seeking, antimatching + minimizing for hiding. We could account for diversity in participant strategy usage by varying the coefficients of the two relevant basis strategy vectors. We successfully applied this model in scenarios of up to 7 rooms. We conclude that an apparent diversity of human conduct in stochastic environments can, in some cases, be explained by varying the weighting of two principle strategies: whether to match/antimatch or maximize/minimize.

Growth and decay are system-level properties of chemical reaction networks (CRNs) relevant from prebiotic chemistry to cellular metabolism. Their properties are typically analyzed through the kinetics of particular models, which requires specification of the full set of kinetic laws and parameters. In this work, assuming a steady balanced-growth regime, we derive stoichiometry-based constraints on the growth (or shrinkage) rate. The resulting bounds are controlled by a topological quantity, the maximum amplification factor, defined via a von Neumann max-min problem over feasible fluxes as illustrated by numerical tests on random-network ensembles of CRNs. We argue for the relevance of our results in the context of origins of life studies and the design of synthetic chemical reaction networks.

Identifying dynamically influential nodes in biological networks is a central problem in systems biology, particularly for prioritizing intervention targets in gene regulatory networks. In this paper, we propose a Shapley-value-based framework for assessing the importance of nodes in a Boolean network with respect to a given target node. The framework comprises two complementary measures: the Knock-out and the Knock-in Shapley values. Moreover, we present a propagation-based method that enables their efficient computation. By exploiting the logical structure of the network, the method avoids exhaustive simulations. The approach is exact for acyclic networks and provides good approximations for cyclic networks. Evaluation on benchmark models from the Cell Collective database shows that the propagation method accurately recovers node importance rankings while achieving substantial speed-ups.

Precise sound localization relies on microsecond sensitivity to interaural time differences (ITDs), yet binaural perception exhibits sluggish tracking of dynamic acoustic cues. How microsecond-level ITD sensitivity arises despite such slow responses remains unresolved. This study proposes that ITD is represented as a stable equilibrium of neural population dynamics rather than through the classical place-coding framework based on delay-line coincidence detection. In this framework, excitatory and inhibitory interactions across frequency channels drive the system toward an equilibrium corresponding to the estimated ITD. Despite relying on relatively slow temporal dynamics, the model achieves microsecond-level precision and reproduces key physiological observations, including frequency-dependent best-delay distributions, without requiring explicit delay lines or precisely timed inhibition. These results challenge the classical place-coding framework and suggest a fundamentally different principle for binaural computation. More generally, the findings indicate that precise temporal information can emerge from stable dynamical states rather than from equally precise neural timing mechanisms, providing a potential resolution to a long-standing paradox in auditory neuroscience.

Single-cell sequencing is revolutionizing biology by enabling detailed investigations of cell-state transitions. Many biological processes unfold along continuous trajectories, yet it remains challenging to extract smooth, low-dimensional representations from inherently noisy, high-dimensional single-cell data. Neighbor embedding (NE) algorithms, such as t-SNE and UMAP, are widely used to embed high-dimensional single-cell data into low dimensions. But they often introduce undesirable distortions, resulting in misleading interpretations. Existing evaluation methods for NE algorithms primarily focus on separating discrete cell types rather than capturing continuous cell-state transitions, while dynamic modeling approaches rely on strong assumptions about cellular processes and specialized data. To address these challenges, we build on the Predictability-Computability-Stability (PCS) framework for reliable and reproducible data-driven discoveries. First, we systematically evaluate popular NE algorithms through empirical analysis, simulation, and theory, and reveal their key shortcomings, such as artifacts and instability. We then introduce NESS, a principled and interpretable machine learning approach to improve NE representations by leveraging algorithmic stability and to enable robust inference of smooth biological structures. NESS offers useful concepts, quantitative stability metrics, and efficient computational workflows to uncover developmental trajectories and cell-state transitions in single-cell data. Finally, we apply NESS to six single-cell datasets, spanning pluripotent stem cell differentiation, organoid development, and multiple tissue-specific lineage trajectories. Across these diverse contexts, NESS consistently yields useful biological insights, such as identification of transitional and stable cell states and quantification of transcriptional dynamics during development.

Self-supervised learning in fluorescence microscopy often relies on 2D projections, despite the inherently three-dimensional nature of cells. We present a systematic comparison of 2D and 3D masked autoencoders (MAE-2D vs. MAE-3D) on volumetric microscopy data. Under matched architectures and training protocols, MAE-3D consistently outperforms 2D max-projection and slice-based variants on downstream single-cell tasks. We further align visual representations with a pretrained protein language model (ESM2) and show that cross-modal supervision yields larger gains for volumetric models. Channel cross-attention and frequency-domain regularization are critical for leveraging 3D spatial context. On protein--protein interaction prediction, our best model achieves a ROC--AUC of 0.86, while on protein localization it reaches an AUC$_{\text{micro}}$ of 0.95 and an F1$_{\text{micro}}$ of 0.74, demonstrating competitive performance on both tasks. Overall, our findings highlight the potential of volumetric modeling and multimodal alignment for representation learning in single-cell microscopy.