Understanding the environmental drivers of forest transpiration is critical for improving global predictions of water availability and ecosystem health. Due to many competing controls on plant water stress and ecosystem transpiration, however, these drivers may vary widely across tree species which have adapted hydraulically to local climate conditions. Here, clustered machine learning models were used to analyze global drivers of forest transpiration rates using the SAPFLUXNET database. Sap flux data from a total of ninety-five sites spanning seven biomes were grouped using two clustering strategies: by biome and by plant functional type. Two supervised machine learning algorithms, a random forest algorithm and a neural network algorithm, were used to predict rates of sap flux for each cluster. The performance and feature importance in each model were analyzed and compared to evaluate the environmental variables that control each cluster's performance. By defining site clusters, these models are able to predict transpiration and its environmental drivers across a wide variety of geographical sites and tree species. Unlike models trained on the entire dataset, high-performing clustered models achieved R$^2$ values to measurement data in the range of 0.74 to 0.90, with the highest performance being achieved in mid-sized clusters of up to thirty-six sites. There was high variance in feature importance between clusters, indicating that key predictors of transpiration varied strongly across both plant functional type and biome. Overall, water-limited climates tended to be more controlled by soil moisture, whereas climates with high mean annual temperature tended to be more controlled by solar radiation and less dependent on air temperature. These findings provide insights into how forest transpiration responds to environmental factors across a wide range of climate types and tree species.

The phase of Quasi-Linkage Equilibrium (QLE) in evolutionary populations is analogous to the thermal equilibrium state in statistical mechanics, a concept pioneered by Kimura in 1965 for two-locus two-allele models. QLE describes a stationary state maintained by the interplay of selection, mutation, recombination and genetic drift. Here we extend QLE theory to populations connected by migration, a fundamental evolutionary force that couples the evolutionary dynamics of interacting subpopulations. Specifically, we examine two populations interacting via symmetric or asymmetric migration while evolving under multi-locus selection. Using whole-genome time-series data generated through FFPopSim, we demonstrate that the QLE phase persists under conditions of sufficiently low migration rates. In this regime, we derive analytical inference relations that allow for the accurate and quantitative estimation of both additive fitness and epistatic interactions.

Non-pharmaceutical interventions (NPIs), including mask-wearing, physical distancing, and hygiene measures, provide the primary means of reducing transmission in the early stages of an epidemic. Individuals adopt one of two strategies-adherence (A) or non-adherence (N) to NPIs. These strategies influence the transmission rate and thus the number of infections, but they also come with inherent costs and benefits. We propose a model coupling behavior and disease dynamics in adherers and non-adherers based on the SIR framework. This gives rise to six behavioral-epidemiological compartments. Using numerical simulations and analytical considerations, we first examine the case where strategies are fixed. Stronger NPIs and more initial adherers lead to fewer infections, and adherers consistently experience lower infection risk than non-adherers. We then introduce behavioral switching based on the benefits and costs of the two strategies. When NPIs are effective, higher transmission rates promote adherence, resulting in fewer infections. Strikingly, in high-severity outbreaks, even modestly effective NPIs can significantly reduce infections. These findings highlight the critical role of the coupling between behavior and disease dynamics, and underscore how individual choices can compromise or compensate public health interventions.

Processing sequential inputs is a fundamental brain function, underlying tasks such as sensory perception, language, and motor control. A challenge in sequence processing is to represent not only the order of events, but also their precise timing. While existing computational models can learn sequential structure, many lack biologically plausible mechanisms to encode element-specific timing and to flexibly control the speed of sequence replay. The spiking Temporal Memory (sTM) model, a biologically inspired network model, provides a framework for key aspects of sequence processing. In the sTM model, each sequence element is represented by a small set of neurons firing synchronously, where the set of active neurons encodes the element's identity in its sequential context. In its original version, however, the sTM model learns the order but not the timing of sequence elements. Further, it remains an open question in neuroscience how the speed of sequence replay can be flexibly modulated. We propose a mechanism where the duration of sequence elements is represented by a sequential activation of element specific neuronal populations, enabling the model to encode sequences across a wide range of timescales. This provides a biologically plausible basis for learning and replaying complex temporal patterns. Additionally, we show that oscillatory background inputs can serve as a clock signal and provide a robust and flexible mechanism for controlling the speed of sequence replay. Our findings suggest that elapsed time is encoded by unique and sparse spatiotemporal patterns of neural activity, and that the speed of sequence replay during wakefulness and sleep is correlated to the characteristics of global oscillatory activity observed in EEG or LFP recordings.

Does the relationship between learning rules and brain alignment generalize across species? We extend our prior finding that untrained CNNs match backpropagation at human V1 by testing the same five learning rules against macaque electrophysiology. The rules are backpropagation (BP), feedback alignment (FA), predictive coding (PC), spike-timing-dependent plasticity (STDP), and an untrained random-weights baseline. The macaque data come from two datasets: MajajHong2015 (V4/IT, 3,200 stimulus presentations, 88/168 neurons) and FreemanZiemba2013 (V1/V2, 135 stimuli, 102/103 neurons). Using RSA with identical model weights from our human study, we find: (1) all models achieve higher alignment with macaque early visual cortex (rho = 0.15-0.30 at V1/V2) than with human fMRI (rho = 0.01-0.08), consistent with the higher signal-to-noise ratio of electrophysiology; (2) STDP and PC produce the highest macaque V1/V2 alignment (rho ~ 0.30 and 0.28), consistent with their leading position among trained rules in human V1; (3) at IT, learning rule rankings show no detectable correlation across species (Kendall's tau = 0.00, p = 1.00), though this null result is expected given that n = 5 provides power only at tau = +/-1.0, and is further confounded by stimulus set differences; (4) a pretrained ResNet-50 (ImageNet) achieves rho = 0.25 at macaque IT, substantially above all custom CNN conditions (rho = 0.07-0.14), suggesting IT alignment is limited by model capacity and training data rather than by the learning rule. Noise ceilings, multi-seed variability (5 seeds), and a stimulus-control analysis are reported. These results demonstrate that early visual alignment is robust across species, while higher-area alignment is modulated by model capacity and stimulus domain.

Highly pathogenic avian influenza A(H5N1) has expanded geographically and ecologically, affecting wild birds, mammalian wildlife, domestic animals, and humans. Wildlife surveillance provides critical early warning for One Health preparedness, yet national-scale analyses integrating host ecology, spatial patterns, seasonality, viral lineage, and risk factors remain limited. This study analysed Canadian wildlife HPAI A(H5N1) surveillance records from 2022 to 2026 to characterise spatiotemporal dynamics and identify factors associated with detection counts. A retrospective analysis of 2,657 detections across 13 provinces and territories was conducted using descriptive epidemiology, spatial clustering methods, and Negative Binomial mixed models. Detections were predominantly avian, with waterfowl and raptors as the major host groups, while mammals accounted for a smaller but epidemiologically important proportion. Detection burden was highest in 2022, with increased activity in autumn and spring. Ontario, Alberta, and British Columbia were identified as major hotspots, with evidence of local clustering in parts of the Prairie region. Reassortant Eurasian-North American lineages dominated detections and were strongly associated with higher detection counts. Modelling results identified year, season, and lineage as key predictors. These findings support risk-based One Health surveillance prioritising high-burden regions, migration-associated periods, key avian host groups, reassortant viral lineages, and continued monitoring of mammalian wildlife.

We present an investigation of the applicability of topological data analysis (TDA) to the study of high-resolution confocal microscopy images of fibrin network structures from patients with oesophageal cancer undergoing intended curative surgery. Investigation of clot structure brings new knowledge about blood coagulation, risk of bleeding, and thrombosis in this group of patients. Images of fibrin network formation in the collected blood samples were captured by confocal microscopy and three-dimensional z-stacks were analysed. Each z-stack was cropped to a centre region for analysis, the validity of which is assessed in detail. Overall, we found no significant differences in fibrin network topology across the perioperative period, and no consistent differences in network structure between the standard and intervention groups.

The "binding problem" of how distributed neural activity unifies into conscious experience has remained an open challenge since its articulation in 1890. We present evidence that conscious integration relies on self-organized criticality maintained by brain-body resonance, placing human cognition within the universality class of critical systems. Using 64-channel EEG data, we demonstrate that conventional preprocessing inadvertently eliminates the very integrative dynamics it seeks to measure. Removing physiological signals conventionally treated as "artifacts" drastically reduces the shared variance between global phase synchronization and stimulus-evoked amplitude, an effect highly specific to physiological components. We trace this to a fundamental brain-body resonance at 78 milliseconds that establishes zero-lag synchronization driven by robust bidirectional causality. Crucially, raw data exhibits heavy-tailed avalanche dynamics indicative of a near-critical regime, whereas conventionally cleaned data definitively rejects power-law distributions, signaling an artificial shift to subcriticality. Finally, we show these critical dynamics enable holographic information encoding, evidenced by a significant emergence of spatial interference patterns post-resonance. Together, these findings indicate that physiological signals actively and selectively support the coupling between large-scale neural coordination and event-related processing.

Stride-to-stride fluctuations in human walking carry a fractal correlation structure that reverses sign under external cueing: self-paced gait is persistent, whereas metronomic or visually cued gait is anti-persistent. Three decades of detrended fluctuation analysis (DFA) have established this reversal as a scaling-exponent shift, but DFA cannot distinguish genuine long-memory dynamics from short-memory autoregressive moving-average (ARMA) processes that produce the same apparent exponent. We fit the full eight-model ARFIMA(1,d,1) family to stride interval and stride speed series from three datasets (N = 70 subjects) spanning overground walking, fixed-speed treadmill walking, metronomic and visual cueing, and graded positional constraint. Model evidence is aggregated through BIC-based Schwarz weights, and the fractional differencing parameter d together with the autoregressive and moving-average coefficients phi and theta are estimated by Bayesian model averaging. Three findings emerge. (i) Long-memory specifications decisively outweigh ARMA alternatives under both persistent and anti-persistent conditions, establishing cued gait anti-persistence as a genuine fractional phenomenon. (ii) DFA alpha overestimates d + 0.5 by 0.25 to 0.34 units, a discrepancy jointly attributable to short-memory components that DFA conflates with long-memory persistence and to a finite-sample negative bias inherent to exact ML-ARFIMA estimation. (iii) The estimated (d, phi, theta) parameters are consistent with a corrective sensorimotor model in which a fractal intrinsic generator, a reactive feedback correction, and a motor-delay component together shape stride-to-stride fluctuations. Whether a single mechanistic model can account quantitatively for the observed parameter ranges across rhythmic, spatial, and unconstrained conditions is a question that the present analysis motivates but cannot alone resolve.

Organelle patterning and its heritability remain central mysteries in cell biology, highlighting the fundamental tension between genetic inheritance and self-assembly. Here, we explore the nonequilibrium assembly and emdedded size control of the Golgi cisternae and endosomes, amid a continuous flux of membrane traffic, within a stochastic framework of mechanochemical fusion-fission cycles that violate detailed balance. Using a dynamical systems approach, we identify distinct, robust regimes, ranging from fixed points to limit cycles with definite phase relations between cisternae. We identify these dynamical regimes with diverse phenotypes, from stable cisternae to periodic, cell-cycle-dependent dissolution/reassembly of cisternae to cisternal progression. We analyse its dynamic response to systematic perturbations or driving protocols and make definite predictions that may be tested experimentally. Our analysis reveals that the two competing models of Golgi organization - vesicular transport and cisternal progression - are, in fact, two phases of the same underlying nonequilibrium process. We see that cisternal size homeostasis is brought about by a size-dependent embedded control system driven by fusion-fission kernels. Finally, our framework offers a strategy for controlling cisternal number and chemical identity by modulating the interplay between glycosylation enzymes and membrane fission-fusion dynamics.

While experiments and computer simulations to study biological phenomena are usually performed in diluted in vitro conditions, such phenomena happen inside the cell, an environment densely packed with diverse macromolecules. Here, we revise recent computational methods to investigate crowded and cellular environments. Protein crowders, inert crowders and small molecules were used to mimic crowding. Simulations were performed for models of the cytoplasm. New methods were developed to simulate crowded systems, reaching up to 200 microseconds of simulation time. Apart from the challenges, modeling and simulations to investigate biological phenomena inside cells is a growing field, and has a lot of potential to improve our understanding of how such phenomena happen in vivo.

Spatio-temporal epidemic forecasting is critical for public health management, yet existing methods often struggle with insensitivity to weak epidemic signals, over-simplified spatial relations, and unstable parameter estimation. To address these challenges, we propose the Spatio-Temporal priOr-aware Epidemic Predictor (STOEP), a novel hybrid framework that integrates implicit spatio-temporal priors and explicit expert priors. STOEP consists of three key components: (1) Case-aware Adjacency Learning (CAL), which dynamically adjusts mobility-based regional dependencies using historical infection patterns; (2) Space-informed Parameter Estimating (SPE), which employs learnable spatial priors to amplify weak epidemic signals; and (3) Filter-based Mechanistic Forecasting (FMF), which uses an expert-guided adaptive thresholding strategy to regularize epidemic parameters. Extensive experiments on real-world COVID-19 and influenza datasets demonstrate that STOEP outperforms the best baseline by 11.1% in RMSE. The system has been deployed at a provincial CDC in China to facilitate downstream applications.

Applications of machine learning in chemistry are often limited by the scarcity and expense of labeled data, restricting traditional supervised methods. In this work, we introduce a framework for molecular reasoning using general-purpose Large Language Models (LLMs) that operates without requiring task-specific model training. Our method anchors chain-of-thought reasoning to the molecular structure by using unique atomic identifiers. First, the LLM performs a zero-shot task to identify relevant fragments and their associated chemical labels or transformation classes. In an optional second step, this position-aware information is used in a few-shot task with provided class examples to predict the chemical transformation. We apply our framework to single-step retrosynthesis, a task where LLMs have previously underperformed. Across academic benchmarks and expert-validated drug discovery molecules, our work enables LLMs to achieve high success rates in identifying chemically plausible reaction sites ($\geq90\%$), named reaction classes ($\geq40\%$), and final reactants ($\geq74\%$). Ultimately, our work establishes a general blueprint for applying LLMs to challenges where molecular reasoning and molecular transformations are key, positioning atom-anchored LLMs as a powerful solution for data-scarce chemistry domains.

Attractor dynamics are a hallmark of many complex systems, including the brain. Understanding how such self-organizing dynamics emerge from first principles is crucial for advancing our understanding of neuronal computations and the design of artificial intelligence systems. Here we formalize how attractor networks emerge from the free energy principle applied to a universal partitioning of random dynamical systems. Our approach obviates the need for explicitly imposed learning and inference rules and identifies emergent, but efficient and biologically plausible inference and learning dynamics for such self-organizing systems. These result in a collective, multi-level Bayesian active inference process. Attractors on the free energy landscape encode prior beliefs; inference integrates sensory data into posterior beliefs; and learning fine-tunes couplings to minimize long-term surprise. Analytically and via simulations, we establish that the proposed networks favor approximately orthogonalized attractor representations, a consequence of simultaneously optimizing predictive accuracy and model complexity. These attractors efficiently span the input subspace, enhancing generalization and the mutual information between hidden causes and observable effects. Furthermore, while random data presentation leads to symmetric and sparse couplings, sequential data fosters asymmetric couplings and non-equilibrium steady-state dynamics, offering a natural generalization of conventional Boltzmann Machines. Our findings offer a unifying theory of self-organizing attractor networks, providing novel insights for AI and neuroscience.

Diabetes is a global health burden, and early detection is critical for timely intervention. This study explores a non-invasive, data-driven framework to identify individuals at risk of diabetes using Volatile Organic Compounds (VOCs) and lifestyle variables. We use causal inference techniques to estimate the impact of VOCs such as acetone, isopropanol, isoprene, and ethanol on blood glucose levels. Additionally, we designed a classifier to distinguish diabetics from non-diabetics using non-invasive markers. We created a risk-based ranking system for individuals in the "gray zone," and identified natural clusters in the population using Gaussian Mixture Model. Our results suggest that specific VOCs exhibit a strong causal influence on glucose levels and that machine learning models can reliably classify and stratify individuals at high risk. This integrated causal-explainable analysis can support the development of tool for non-invasive early screening of diabetes.

Stenting is among the most common transcatheter interventions for congenital heart disease (CHD). Patient-specific computational fluid dynamics (CFD) simulations can predict hemodynamic outcomes of intervention scenarios but require post-operative vascular geometries that reflect stent-induced shape changes, which existing tools either model inadequately or require extensive time or manual effort to generate. We present SDFStent, a signed distance function (SDF) based mesh deformation method for virtual stenting that operates in real time, maintains mesh integrity, and preserves junction geometry. The stent is modeled as a pipe surface composed of piecewise-capsule SDFs joined by a smooth-minimum operator. Mesh vertices near the expanding SDF surface are displaced along the SDF gradient with a compactly supported fall-off function and an alpha blending mask. SDFStent was benchmarked against three existing approaches and validated on three tetralogy of Fallot (ToF) patients and three coarctation of the aorta (CoA) patients using rigid-wall steady-state CFD simulations against clinical catheterization measurements. Against a prescribed diameter of 6.0 mm, the method produced a mean stented diameter of 5.92 $\pm$ 0.08 mm in 1.5 s, over 100$\times$ faster than the best stenting-specific comparator. All output meshes were watertight and self-intersection-free. CFD-simulated post-operative pressure drops agreed with clinical measurements within 4 mmHg (mean error 2 mmHg). SDFStent produces simulation-ready post-stent models that match prescribed stent dimensions at interactive speeds, from pre-operative anatomy and catheterization data alone. The implementation is open-source and available in 3D Slicer. Its scriptable architecture enables automated generation of large synthetic cohorts for data-driven surrogate modeling.

We give a Hasse-diagram characterization of when a clustering system $\mathcal C$ on a finite taxa set $X$ is the hardwired clustering system $C_N$ of a rooted level-$k$ network. For each non-trivial block $B$ of $H=\mathcal H[\mathcal C]$, we define a parameter $μ(B)$ using minimum families of clusters that generate all overlap-intersections inside $B$. The main theorem proves that there exists a rooted level-$k$ network $N$ with $C_N=\mathcal C$ if and only if $μ(B)\le k$ for every non-trivial block $B$ of $H$. The necessity proof shows that overlap-intersection pieces must be represented by non-root hybrid vertices in any realizing block. The sufficiency proof is constructive: starting from the Hasse diagram, it iteratively splits selected hybrid vertices, preserves the hardwired clustering system, and terminates with a realization whose level is bounded by the block-wise values of $μ$.

Phylogenetic trees are hybrid objects: branch lengths vary continuously, while topologies change discretely through edge contractions and expansions. Billera-Holmes-Vogtmann (BHV) tree space provides a canonical geometry for this structure, representing each resolved topology as a Euclidean orthant and topological changes as motion across shared lower-dimensional boundaries. We introduce PhylaFlow, a hybrid flow-matching model that learns posterior-basin transport in BHV tree space. PhylaFlow is trained on BHV geodesic paths from random starting trees to short-run posterior samples, coupling continuous branch-length motion within orthants with learned boundary events and discrete topology transitions. We evaluate the learned geometry operationally: if the flow reaches posterior-relevant regions, finite-budget Bayesian refinement initialized from, or guided by, its terminal trees should recover posterior-supported topologies more efficiently. Across DS1-DS8 phylogenetic posterior benchmarks, PhylaFlow substantially reduces initial Tree-KL relative to classical initializers. After finite-budget MrBayes refinement, direct PhylaFlow improves early and intermediate topology-recovery trajectories on most datasets, while split-guided PhylaFlow-MCMC obtains the strongest hard-case results. The best PhylaFlow variant outperforms short-warmup on seven of eight datasets and PhyloGFN on five of eight under the same refinement budget. In a joint sequence-conditioned experiment, sequence embeddings steer posterior split recovery, although exact posterior topology recovery remains preliminary. These results show that hybrid flow matching can learn actionable transport in BHV tree space and provide a geometry-aware proposal mechanism for Bayesian phylogenetic inference.

We investigate how key epidemiological parameters shape both seasonal epidemics and the persistence of dengue transmission. Our findings confirm known mechanistic drivers of epidemic variability and introduce a ranking of parameter importance in our dengue model, which in turn informs the prioritization of public health policies. We propose a stochastic vector-host model with waning immunity, exogenous infection, and vertical transmission. To assess parameter influence, we first qualitatively analyze the macroscopic model. We then perform a multivariate Sobol sensitivity analysis of epidemic summary statistics, and examine the variance of the endemic equilibrium as a function of model parameters. We show that the macroscopic model is well posed, vertical transmission lowers the threshold for persistence, and low spatial coupling increases infectious endemic equilibria. The vector-host population ratio and host recovery rate have the largest first-order and total sensitivity indices, surpassing the contact rates; this implies that control measures during seasonal dengue should prioritize protecting infectious hosts from mosquito bites. Finally, we show that the covariance of hosts and vectors at the endemic equilibrium is asynchronous in the contact-rate plane. This robust pattern has epidemiological, ecological and evolutive interpretations. A dengue strain has two niches to exploit during the endemic regime, and coexisting strain have two niches each. Moreover, large fluctuations in a given strain during the endemic regime provide a mechanistic explanation for high vertical transmission, enabling viral reservoirs that can hatch and trigger outbreaks in the following season. We argue that our model and results can be adapted to address specific public health questions to guide dengue control using field data.

Phylogenetic networks and, more generally, directed acyclic graphs (DAGs) represent hierarchical structure beyond trees, for instance in the presence of reticulate evolutionary events such as hybridization or horizontal gene transfer. A central question is which parts of such graphs are essential with respect to leaf-observable information, and which parts can be removed without changing this information. Resolving this question can lead to principled simplification methods for phylogenetic networks, such as the recent normalization approach of Francis et al. In this paper, we study this question from three related perspectives: clusters displayed by a DAG $G$, least common ancestors (LCAs) of subsets of its leaf set, and visibility, a path-based property of vertices. We first introduce an LCA-based simplification procedure called $i$-regularization. For a DAG $G$ and $i\geq 1$, the DAG $\reg_i(G)$ retains precisely those vertices that occur as unique LCAs of leaf subsets of size at most $i$, removes the remaining non-leaf vertices by a graph-editing operation $\ominus$, and then deletes shortcuts. We show that $\reg_i(G)$ preserves all such LCAs, is $i$-lca-relevant, and admits a cluster-level description: it is regular, i.e., isomorphic to the Hasse diagram of the corresponding lca-clusters. We then compare LCA-based regularization with normalization. Using the same $\ominus$-operator, we describe the cover construction underlying normalization, identify visible vertices that are nevertheless removed, and characterize when regularization and normalization coincide. Together, these results provide a unified framework for cluster-based, LCA-based, and visibility-based simplifications of DAGs and phylogenetic networks.

Polygenic risk scores (PRSs) aggregate genetic effect estimates to predict disease susceptibility, yet clinical deployment often exposes raw genotype data to third-party compute infrastructure. Prior homomorphic-encryption approaches, still require trust in a designated evaluator. We present bioETH-PRS, a protocol that replaces that evaluator role with immutable smart contracts on a blockchain supporting Fully Homomorphic Encryption (fhEVM). Using the integer-exact TFHE scheme, bioETH-PRS computes the PRS dot product entirely within the encrypted domain, keeping both genotype dosage vectors and GWAS weight vectors hidden from external parties throughout execution. We introduce a three-step fixed-point quantisation scheme for representing signed GWAS weights as unsigned 64-bit integers, achieving machine-epsilon reconstruction accuracy on validated fixtures. A four-contract architecture separates data custody, model publication, computation, and output release, and supports both a classic chunked path and a streaming path, with the latter reducing mock-measured gas by 37%. An on-chain noisy output oracle emits an encrypted noisy-score handle and a publicly decryptable ternary category, reducing raw score exposure and probing risk. Prototype evaluation on real GWAS fixtures confirms linear gas scaling and suggests that the approach may be cost-competitive in low-gas deployment environments.

Protein-protein interactions (PPIs) govern nearly all cellular processes, yet computational methods for identifying binding partners typically produce ranked predictions without mechanistic justification. This creates a fundamental barrier to adoption because biologists cannot assess whether predictions reflect genuine biochemical insight or spurious correlations. We present \textbf{Protein Thoughts}, a framework that reformulates PPI discovery as an interpretable search problem with explicit reasoning. The system decomposes binding evidence into four biologically meaningful signals: sequence similarity reflecting evolutionary relationships, structural complementarity capturing geometric fit, interface balance, and chemical compatibility encoding residue-level interactions. Rather than collapsing these signals into an opaque score, we preserve their individual contributions through a transparent value function that enables both ranking and auditing. To navigate large candidate spaces efficiently, we introduce hypothesis-guided entropy-regularized Tree-of-Thoughts search. A fine-tuned language model generates search directives from embedding-derived features, classifying candidates as high-priority, exploratory, or skippable. These directives condition a Boltzmann policy that balances exploitation with entropy-driven exploration, while hypothesis-aware pruning prevents premature abandonment of promising candidates. For candidates exhibiting score disagreement, hypothesis-conditioned embedding-space flow matching transports protein embeddings toward the binder manifold. On the SHS148k benchmark, Protein Thoughts achieves mean best-binder rank of 11.2 versus 47.7 for an entropic tree search baseline, a 76% improvement, and for binding prediction the trained value function achieves $91.08 \pm 0.19$ Micro-F1, outperforming existing PPI methods on the same dataset.

Computational functionalism about consciousness is often criticized for relying on observer-relative interpretations of physical systems. This paper proposes a mathematical refinement of functionalism that avoids this problem. The central idea is that consciousness-relevant functional organization should be identified not with arbitrary input-output mappings, semantic labels, or externally imposed computational descriptions, but with a system's canonical functional structure: the minimal state-transition structure obtained by identifying internal states that have identical future behavior under all possible continuations. On this view, a state is functionally defined by its complete counterfactual role: how the system would evolve and respond from that state under possible future interactions. We call this position canonical functionalism. The framework does not claim to identify which systems are conscious, nor to show that functional organization is sufficient for consciousness. Rather, it identifies the canonical object over which functionalist theories of consciousness should be formulated: the task is to specify consciousness-relevant invariants, measures, or structural conditions over canonical functional structures, rather than over arbitrary semantic interpretations or superficial behavioral profiles. This reframes familiar objections about lookup tables, simulations, unfolding, and observer-relative computation: such cases do not by themselves refute functionalism, but force the functionalist to specify whether the relevant canonical structure is preserved, and if not, which additional structural features are missing.

Graph neural networks (GNNs) are increasingly used to model biological systems, yet the reliability of post-hoc explanation methods for recovering meaningful molecular mechanisms remains unclear. Here, we systematically evaluate four widely used approaches: Saliency Attribution (SA), Integrated Gradients (IG), GNNExplainer, and Layer-wise Relevance Propagation (LRP) for identifying disease-relevant structure in breast cancer RNA-seq data projected onto a protein-protein interaction network. Using synthetic benchmarks with known ground-truth motifs, we show that explanation methods recover distinct signal organizations: SA performs best for sparse single-node drivers, whereas IG and LRP preferentially recover distributed pathway-like and cascade-like signals. In TCGA BRCA data, we identify a consistent topological signature of disease-associated hubs in which attribution peaks in the immediate 1-hop neighborhood and decays across successive network shells, a pattern most pronounced for IG and LRP and associated with strong enrichment of known cancer hubs. We further observe a trade-off between local hub enrichment and global gene ranking performance, with IG optimizing local enrichment and SA achieving superior global discrimination. Motivated by these complementary behaviors, we introduce a framework combining a shell-based hub score with consensus ranking across explainers. Consensus scores improve prioritization of canonical cancer genes (TP53, BRCA1, ESR1, MYC), reduce dependence on node degree, and, especially when tuned, outperform individual methods. Pathway enrichment further reveals improved recovery of biologically coherent cancer programs, including ERBB2, RTK, MAPK, immune, and cytokine signaling. Together, these results demonstrate that topology-aware integration of graph explanations can improve biological interpretability and biologically relevant molecular recovery.

Informative cluster size (ICS) and informative subgroup size (ISS) can distort marginal association estimates when the number of observed units, or their distribution across outcome-defined categories, is related to the outcomes under study. This issue is especially relevant for paired outcomes, where the observed association can depend on cluster size, paired-category composition, and the process by which units become available for analysis. We propose three weighted estimating approaches for marginal association between paired outcomes in clustered data. The weights are derived from within-cluster resampling arguments and extend inverse cluster-size and subgroup-size weighting to paired outcome categories. We also modify an existing ISS testing procedure by utilizing Stouffer's method to reduce computational burden. To evaluate the methods, we develop a simulator for clustered paired outcomes that separates unit-level association, latent cluster-level association, and outcome-dependent retention. Simulations show that pair-based weighting can reduce bias when association arises through unit-level dependence and subgroup composition is informative, but can attenuate association carried by latent cluster-level structure. Typical inverse-cluster weighting remains more stable when the association is primarily cluster-level. Application to NHANES oral-health data shows small positive periodontal and caries associations overall, with filled-surface outcomes showing stronger ISS evidence and greater sensitivity to pair-based weighting than decayed-surface outcomes. These results indicate that marginal association under ICS and ISS should be interpreted in relation to the source of association, observed-unit structure, and assumptions used to choose the weighting scheme.

Building virtual cells with generative models to simulate cellular behavior in silico is emerging as a promising paradigm for accelerating drug discovery. However, prior image-based generative approaches can produce implausible cell images that violate basic physical and biological constraints. To address this, we propose to post-train virtual cell models with reinforcement learning (RL), leveraging biologically meaningful evaluators as reward functions. We design seven rewards spanning three categories-biological function, structural validity, and morphological correctness-and optimize the state-of-the-art CellFlux model to yield CellFluxRL. CellFluxRL consistently improves over CellFlux across all rewards, with further performance boosts from test-time scaling. Overall, our results present a virtual cell modeling framework that enforces physically-based constraints through RL, advancing beyond "visually realistic" generations towards "biologically meaningful" ones.

Cellular metabolism is globally regulated by various currency metabolites such as ATP, GTP, and NAD(P)H. These metabolites cycle between charged (high-energy) and uncharged (low-energy) states to mediate energy transfer. While distinct currency metabolites are associated with different metabolic functions, their charged and uncharged forms are generally interchangeable via biochemical reactions such as ${\rm ATP{\,+\,}GDP{\,\rightleftharpoons\,}ADP{\,+\,}GTP}$ and $\rm NADP^+{\,+\,}NADH{\,\rightleftharpoons\,}NADPH{\,+\,}NAD^+ $. Thus, their energetic states are generally coupled and influence each other, which would hinder the independent regulation of different currency metabolites. Despite the extensive knowledge of the molecular biology of individual currency metabolites, it remains poorly understood how the coordination of various coupled currency metabolites shapes metabolic regulation, efficiency, and ultimately the evolution of organisms. Here, we present a minimal theoretical model of metabolic currency coupling and reveal a fundamental tradeoff relationship between metabolic controllability and thermodynamic cost: increasing the capacity to independently regulate multiple currency metabolites generally requires comparable abundances of those metabolites, which in turn incurs a higher entropy production rate. The tradeoff suggests that in complex environments, organisms evolutionarily favor an equal abundance of currency metabolites to enhance metabolic controllability at the expense of a higher thermodynamic cost; conversely, in simple environments, organisms evolve to have imbalanced amounts of them to reduce heat dissipation. These considerations also offer a hypothesis regarding evolutionary trends in nucleotide-pool balance and genomic GC content.

Constraint-based metabolic modeling is the predominant framework for simulating cellular metabolism. The central assumption of these models is that metabolism operates at a steady state, meaning that the production and consumption rates of each metabolite are balanced. This assumption imposes linear constraints on the fluxes of biochemical reactions. Flux Balance Analysis (FBA), a fundamental method in the field, is formulated as an optimization problem maximizing a cellular objective (e.g., growth) over the resulting linear subspace of steady state fluxes. Many other methods in the field are expressed either as a modification to FBA, or use FBA as a black box within an algorithm. Here, we propose a general alternative to optimization called FBApro. For any given vector of reference fluxes, FBApro finds the closest flux vector within the steady-state subspace, and accounts for both partially given reference fluxes and exact constraints on reactions. While FBApro is the solution to a quadratic program, we show that it can be implemented as a single linear operation using orthogonal projections to corresponding affine spaces and sets of linear equations. The overall approach is computationally efficient, does not require a cellular objective, and is easy to implement. We formally derive the closed-form expressions for FBApro and simpler variants, and validate it on both synthetic and real cancer cell line data.

Molecular property regression struggles with cases in chemically relevant target ranges that are underrepresented in datasets. Standard average error minimization approaches underperform in these highly relevant cases, and oversampling approaches lead to meaningless molecular representations. In this paper, we propose SPECTRA, a spectral, domain-aware graph generation method designed to improve the prediction of underrepresented but relevant molecular property values. It combines a rarity-aware budgeting scheme to focus generation where data are scarce, target-neighbors graph alignment to establish structural correspondence, and interpolation of Laplacian spectra, node features, and targets. Coupled with spectral GNN using edge-aware Chebyshev convolutions, SPECTRA shows its effectiveness in property prediction benchmarks with competitive performance over leading state-of-the-art methods in relevant target ranges, while requiring ~4x less computational time.

The cycle of scientific discovery is frequently bottlenecked by the slow, manual creation of software to support computational experiments\cite{hannay2009how}. To address this, we present Empirical Research Assistance (ERA), an AI system that creates expert-level scientific software whose goal is to maximize a quality metric. The system uses a Large Language Model (LLM) and Tree Search (TS)\cite{silver2016mastering} to systematically improve the quality metric and intelligently navigate the large space of possible solutions. ERA achieves expert-level results when it explores and integrates complex research ideas from external sources. The effectiveness of tree search is demonstrated across a diverse range of tasks. In bioinformatics, ERA discovered 40 novel methods for single-cell data analysis that outperformed the top human-developed methods on a public leaderboard. In epidemiology, ERA generated 14 models that outperformed the CDC ensemble and all other individual models for forecasting COVID-19 hospitalizations. ERA also produced expert-level software for geospatial analysis, neural activity prediction in zebrafish, and numerical solution of integrals, and a novel rule-based construction for time series forecasting. By devising and implementing novel solutions to diverse tasks, ERA represents a significant step towards accelerating scientific progress.