Associating multiple sensory cues with a single experience or object is a fundamental process that improves object recognition and memory performance. However, neural mechanisms that bind sensory features during learning and augment memory expression are unknown. Here we demonstrate multisensory appetitive and aversive memory in Drosophila. Combining colours and odours improved memory performance, even when each sensory modality was tested alone. Temporal control of neuronal function revealed visually-selective mushroom body Kenyon Cells (KCs) to be required for enhancement of visual and olfactory memory recall after multisensory training. Synapse-level connectomics suggests that valence-relevant dopaminergic reinforcement could permit the KC-spanning serotonergic DPM neurons to bridge between previously modality-selective KC streams. Consistent with this model, DPM transmission is uniquely required during multisensory memory formation and for enhanced expression of olfactory memory afterwards. In addition, signalling via the DopR1 dopamine receptor is required in APL neurons, suggesting that reinforcing dopamine could locally release GABA-ergic inhibition to permit bridging microcircuits to function. Cross-modal binding thereby expands the KCs representing the olfactory memory engram into those representing the colour. We propose that broadening of the engram improves memory performance after multisensory learning and permits a single sensory feature to retrieve the memory of the multimodal experience.

Salts are an integral part of the environment for living systems and, therefore, understanding their effects on proteins and other biomolecules is of fundamental interest. Small-angle X-ray scattering (SAXS) of protein solutions can provide valuable information on salt effects, but extracting this information has been a significant challenge. For example, SAXS data of bovine serum albumin (BSA) at various salt concentrations were fit to three different spherical models. Here we combined the newly developed FMAPIq approach with explicit-solvent all-atom molecular dynamics simulations to show that the complex effects of salt on the SAXS of BSA originate from the interplay of ions and hydration water, leading to a general picture of protein-ion-water interactions.

Integrating theoretical neuroscience, decision theory, and probabilistic inference offers a promising route to understanding human cognition, yet concrete methodological bridges between agentic AI models and behavioral data analysis remain formally underdeveloped. We advance this synthesis under the framework of agentic behavioral modeling (ABM), which treats artificial agents as latent, generative hypotheses about cognitive mechanisms and evaluates them by their statistical adequacy in explaining human behavior. After outlining its conceptual foundations, we apply the framework to two minimal laboratory paradigms: a binary perceptual contrast-discrimination task and a symmetric two-armed bandit learning task. We formalize each task-agent-data system as a joint probability model, derive explicit conditional log-likelihoods for behavioral inference, validate different model variants using model and parameter recovery simulations, and evaluate them in light of empirical data. Using these minimal examples, we provide an agent-centric interpretation of the psychometric function, derive optimal policies for both tasks, and show the equivalence between Rescorla-Wagner learning and Bayesian inference in symmetric bandits. More broadly, this work may serve as a conceptual and practical foundation for applying ABM to cognitive behavioral science.

Within many bacterial colonies, persister cells exist as a subpopulation that is tolerant to antibiotics and other stressors, yet not genetically distinct from the rest of the colony. A recent study has proposed epigenetic inheritance as a mechanism that leads to the presence of persister cells. We analyze a nonlocal PDE--ODE model introduced in that study to describe the epigenetic inheritance process and establish its mathematical well-posedness, including existence, uniqueness, and nonnegativity of solutions. We identify a sharp parameter threshold delineating extinction from persistence of the colony: below this threshold the washout equilibrium is globally asymptotically stable, while above it a unique positive equilibrium exists and the population is weakly persistent. Notably, this threshold is independent of the internal community structure.

Early detection of Alzheimer's disease from spontaneous speech has emerged as a promising non-invasive screening approach. However, the influence of automatic speech recognition (ASR) quality on downstream clinical language modeling remains insufficiently understood. In this study, we investigate Alzheimer's disease detection using lexical features derived from Whisper ASR transcripts on the ADReSSo 2021 diagnosis dataset. We evaluate interpretable machine-learning models, including Logistic Regression and Linear Support Vector Machines, using TF-IDF text representations under repeated 5x5 stratified cross-validation. Our results demonstrate that transcript quality has a statistically significant impact on classification performance. Models trained on Whisper-small transcripts consistently outperform those using Whisper-base transcripts, achieving balanced accuracy above 0.7850 with Linear SVM. Paired statistical testing confirms that the observed improvements are significant. Importantly, classifier complexity contributes less to performance variation than ASR transcription quality. Feature analysis reveals that cognitively normal speakers produce more semantically precise object- and scene-descriptive language, whereas Alzheimer's speech is characterized by vagueness, discourse markers, and increased hesitation patterns. These findings suggest that high-quality ASR can enable simple, interpretable lexical models to achieve competitive Alzheimer's detection performance without explicit acoustic modeling. The study provides a reproducible benchmark pipeline and highlights ASR selection as a critical modeling decision in clinical speech-based artificial intelligence systems.

Automated radiology report generation has gained increasing attention with the rise of deep learning and large language models. However, fully generative approaches often suffer from hallucinations and lack clinical grounding, limiting their reliability in real-world workflows. In this study, we propose a multimodal retrieval-augmented generation (RAG) system for grounded drafting of chest radiograph impressions. The system combines contrastive image-text embeddings, case-based similarity retrieval, and citation-constrained draft generation to ensure factual alignment with historical radiology reports. A curated subset of the MIMIC-CXR dataset was used to construct a multimodal retrieval database. Image embeddings were generated using CLIP encoders, while textual embeddings were derived from structured impression sections. A fusion similarity framework was implemented using FAISS indexing for scalable nearest-neighbor retrieval. Retrieved cases were used to construct grounded prompts for draft impression generation, with safety mechanisms enforcing citation coverage and confidence-based refusal. Experimental results demonstrate that multimodal fusion significantly improves retrieval performance compared to image-only retrieval, achieving Recall@5 above 0.95 on clinically relevant findings. The grounded drafting pipeline produces interpretable outputs with explicit citation traceability, enabling improved trustworthiness compared to conventional generative approaches. This work highlights the potential of retrieval-augmented multimodal systems for reliable clinical decision support and radiology workflow augmentation

Cell extrusion is an essential mechanism for controlling cell density in epithelial tissues. Another essential element of epithelia is curvature, which is required to achieve complex shapes, like in the lung or intestine. Here we introduce a three-dimensional bubbly vertex model to study the interplay between extrusion and curvature. We find a generic cellular bulging instability at topological defects which is much stronger than for standard vertex models. Analyzing cell shapes in three-dimensional imaging data of spherical mouse colon organoids, we infer that pentagonal cells have an increased basal interfacial tension, suggesting that cells at topological defects react to the different force conditions. Using the bubbly vertex model, we show that such basal tensions stabilize against the predicted instability and result in better cell shape control than tissue-scale mechanisms such as lumen pressure and spontaneous curvature. Our theory suggests that epithelial curvature naturally leads to bulged and extrusion-like cell shapes because the interfacial curvature of individual cells at the defects strongly amplifies buckling effected by tissue-scale topological defects in elastic sheets. Our results highlight the complex interplay of forces across scales in three-dimensional tissue organization.

Inverse reinforcement learning (IRL) is a powerful paradigm for uncovering the incentive structure that drives agent behavior, by inferring an unknown reward function from observed trajectories within a Markov decision process (MDP). However, most existing IRL methods require access to the transition function, either prescribed or estimated \textit{a priori}, which poses significant challenges when the underlying dynamics are unknown, unobservable, or not easily sampled. We propose Fokker--Planck inverse reinforcement learning (FP-IRL), a novel physics-constrained IRL framework tailored for systems that can be described by Fokker--Planck (FP) dynamics. FP-IRL simultaneously infers both the reward and transition functions directly from trajectory data, without requiring access to sampled transitions. Our method leverages a correspondence between MDPs and the FP equation, linking reward maximization in MDPs with free energy minimization in FP dynamics. This connection enables inference of the FP potential function using our inference approach of variational system identification, from which the full set of MDP components -- reward, transition, and policy -- can be recovered using analytic expressions. We demonstrate the effectiveness of FP-IRL through experiments on synthetic benchmarks and a modified version of the Mountain Car problem. Our results show that FP-IRL achieves accurate recovery of agent incentives while preserving computational efficiency and physical interpretability.

Fabricating brightly fluorescent layers with nanometric thickness and digitally controlled lateral structuration remains a challenge for next-generation photonic devices, optical calibration standards, and biocompatible interfaces. Here, we introduce Nano-Bead Emitters (NBEs), hydrogel nanoparticles covalently functionalized with fluorophores, as a universal, water-processable ink platform for fabricating programmable nanometric fluorescent architectures. By immobilizing fluorophores within a charged nanohydrogel scaffold, the platform entirely decouples film morphology from dye solubility. This molecule-independent strategy enables spectrally distinct, inherently water-insoluble dyes to be processed using a single, standardized aqueous ink formulation. Combined with laser-induced forward transfer (LIFT) printing, this additive approach yields highly uniform fluorescent layers (~7 nm thickness, sub-nanometric roughness). This structural invariance produces complex multicolor patterns sharing identical thickness and surface morphology across all spectral channels, a critical requirement for quantitative optical calibration. Furthermore, LIFT printing provides programmable, layer-by-layer control over fluorescence intensity via successive deposition cycles, yielding precisely tunable brightness without aggregation-caused quenching. This maskless technique enables rapid, high-fidelity printing of both monochromatic and multicolor patterns over macroscopic areas with absolute spatial resolution. Finally, these universally compatible NBE inks stably deposit onto diverse substrates (glass, polymers, semiconductors, metasurfaces), effectively bridging scalable manufacturing with high-performance integrated photonic systems.

Virtual cell (VC) models aim to predict cellular responses to any perturbations in silico and have emerged as a promising approach for drug discovery and precision medicine. Yet, a clear gap still remains: while models routinely reported impressive results on standard benchmarks, it is unclear whether their predictions are truly meaningful in practice. This is mainly due to limitations in current evaluation setups, which are often overly simplified or inconsistent, and do not reflect the complexity and variability of real biological systems. Here, we introduce a standardized and modular benchmarking framework for virtual cell prediction. Our framework evaluates diverse models under in-the-wild challenging scenarios, including unseen cell contexts, unseen perturbations, and cross-dataset generalization, which better reflect practical applications. Our analysis shows that model performance is highly context-dependent and shaped by task design and evaluation criteria. In commonly used setups, performance is often overestimated, and naive dataset aggregation can even reduce performance. When evaluated under more strict conditions, model performance drops markedly, indicating limited robustness to shifts across cellular contexts. In unseen perturbation settings, models including simple linear approaches capture global transcriptional trends but fail to recover fine-grained perturbation-specific effects. In addition, different evaluation metrics focus on different biological properties, leading to substantially different model rankings. Together, our framework provides a more reliable and biologically grounded evaluation, offering clearer guidance for applying virtual cell models in real scenarios.

Motion retargeting from humans to human-like artificial agents is becoming increasingly important as humanoid robots grow more capable. However, most existing approaches focus only on reproducing kinematics and ignore the rich sensorimotor experience associated with human movement. In this work, we present a framework for simulating the multimodal sensorimotor experiences of infants using physical and virtual humanoids. From a single video, our method reconstructs the infant's body configuration by extracting its skeletal structure and estimating the full 3D pose from each frame. Then we map the reconstructed motion onto several developmental platforms: the physical iCub robot and the virtual simulators pyCub, EMFANT and MIMo. Replaying the retargeted motions on these embodiments produces simulated multisensory streams including proprioception (joints and muscles), touch, and vision. For the best-matching embodiment, the retargeting achieves sub-centimeter accuracy and enables a rich multimodal analysis of infant development as well as enhanced automated annotation of behaviors. This framework provides a unique window into the infant's sensorimotor experience, offering new tools for robotics, developmental science, and early detection of neurodevelopmental disorders. The code is available at https://github.com/ctu-vras/motion-retargeting/.

The tragedy of the commons has traditionally been framed as a problem of resource overuse driven by self-interested exploitation. In contrast, growing empirical evidence shows that insufficient use or abandonment of natural resources, known as underuse, can also lead to ecological degradation and loss of ecosystem services. Despite its relevance, underuse has rarely been examined within evolutionary theories of resource use. Here, we develop a simple eco-evolutionary model that integrates both provisioning and non-provisioning ecosystem services to analyze the evolution of resource-use strategies. Using adaptive dynamics, we investigate how individual resource use evolves while altering resource abundance. The model shows that overuse and underuse arise naturally as alternative evolutionary outcomes of the same underlying process, alongside intermediate use and evolutionary branching. We derive analytical conditions for the existence, number, and stability of evolutionarily singular strategies, and show that the qualitative evolutionary fate is primarily determined by the shape of provisioning benefits. Only when provisioning benefits increase in a concave manner does evolutionary dynamics converge to a unique intermediate strategy that is continuously stable. In contrast, convex increasing benefits generate a broader range of outcomes: overuse, underuse, bi-stability, and evolutionary branching. By explicitly comparing the continuously stable strategy with the socially optimal strategy, we further quantify how their deviations depend on the valuation of non-provisioning services. Our results provide a theoretical framework for viewing the common-pool resource dilemmas as intrinsically two-sided evolutionary problems, and offer a baseline for future studies exploring interventions to address overuse and underuse simultaneously.

Personalized cancer modeling for clinical applications requires robust and efficient parameter calibration, particularly in settings with limited patient data. This need is especially critical for medical digital twins (MDTs), which are virtual representations of disease continuously updated using longitudinal patient measurements. In this work, we propose a novel parameter personalization framework for dynamical cancer models under data scarcity. Our approach decomposes selected model parameters into a common component, shared across patients, and a personalized component, which is patient-specific and can be updated as new data become available. The common component captures population-level structure and is estimated once, providing an informed prior that enables rapid and accurate personalization. We demonstrate the effectiveness of this framework using synthetic data generated from canonical dynamical systems, such as logistic growth models with optimized treatment interventions. Our results show that parameter decomposition significantly improves calibration performance in limited-data regimes, facilitating fast and reliable personalization and supporting the development of patient-specific cancer models and MDTs.

Epidemics have shaped human history, often with devastating consequences, motivating the development of mathematical models to understand and control their dynamics. Among the many aspects of epidemic behavior, the conditions that lead to epidemic extinction stand out as a central-if not the fundamental-question in epidemic modeling. In this work, we study epidemic extinction in a continuous SIRS (Susceptible-Infected-Recovered-Susceptible) model governed by a system of ordinary differential equations (ODEs). The model includes vaccination as a time-dependent process and considers the reinfection of recovered individuals through waning immunity. We analyze how different parameter regimes -- particularly infection, recovery, and immunity loss rates -- affect the persistence or extinction of the epidemic. Special attention is given to the limitations of continuous population models, in which the infected fraction can fall below the equivalent of a single individual, leading to nonphysical outcomes such as unrealistically long persistence or artificial secondary peaks. By comparing the continuous SIRS dynamics with expected real-world thresholds for extinction, we highlight the importance of incorporating stochasticity or discrete effects to accurately describe epidemic fade-out.

Training stable biological foundation models requires rethinking attention mechanisms: we find that using sigmoid attention as a drop in replacement for softmax attention a) produces better learned representations: on six diverse single-cell datasets, sigmoid achieves 25% higher cell-type separation, better cell-type cohesion metrics, and lower validation loss, b) faster training, models with sigmoid attention train up to 10% faster than their softmax counterparts, and c) more stable training by eliminating inherent sources of instability in softmax attention. We establish that sigmoid attention has globally bounded derivatives ($\leq 0.25$) as opposed to softmax, and a diagonal Jacobian structure in contrast with softmax's dense coupling, which together help alleviate training instabilities. In stress tests on 160M-parameter bidirectional attention models trained without gradient clipping on 8K-token sequences, softmax diverges catastrophically, with gradients exploding by four orders of magnitude, while sigmoid remains stable. Finally, we implement and open-source TritonSigmoid, an efficient GPU kernel that achieves 515 TFLOPS on H100 GPUs, outperforming both FlashAttention-2 and FlashSigmoid, with native padding support, which is essential for biological sequences. Our results establish sigmoid attention as both theoretically grounded and empirically superior for biological foundation models. Code is available at https://github.com/MSDLLCpapers/triton-sigmoid

Vector-borne diseases often infect multiple host species, increasing the likelihood of disease persistence due to the presence of multiple reservoirs. Vector biting patterns and feeding preferences can shift in response to selective pressures introduced by disease control interventions, altering the dynamics of transmission. In this paper, we develop a mathematical model that couples host diversity and adaptive vector behavior with vector-borne disease transmission dynamics, focusing on a system with two hosts and a vector population exhibiting preference for one host. We derive the basic reproduction number, $R_0$, a threshold that determines the existence of two equilibria in our model, and obtain conditions that can lead to the long-term persistence of the disease. Our analysis suggests that shortening the infectious period of the vector's preferred host is an effective control strategy. We also identified a threshold that determines whether shifting vector preference toward a non-preferred host amplifies or reduces the disease burden on the primary preferred host. Our results show that protective measures for the preferred host can trigger adaptive shifts in vector preferences, reducing disease prevalence in that host. However, this shift may lead to an increase in overall host prevalence.

Automated depression detection often relies on static aggregation of conversational signals, potentially obscuring clinically meaningful behavioral dynamics. We investigated whether entropy-driven temporal biomarkers improve depression detection beyond standard pooled features using the DAIC-WOZ corpus. Using 142 labeled participants, we reconstructed utterance-level acoustic trajectories and compared pooled temporal baselines, trajectory dynamics, Shannon entropy biomarkers, recurrence quantification, sample entropy, fractal complexity, and coupling biomarkers under leakage-aware validation. Static pooling achieved an AUC of 0.593, trajectory dynamics improved performance to 0.637, and entropy biomarkers produced the strongest statistically significant improvement over pooled baselines (AUC 0.646; nested cross-validated AUC 0.615; permutation p = 0.017). Entropy biomarkers outperformed recurrence, coupling, sample entropy, and fractalbased features, with several biomarkers stable across folds. These findings suggest depression-related signal may lie less in average acoustic levels than in entropy of conversational dynamics, supporting temporally informed digital phenotypes for mental-health assessment.

T-cells are central to SARS-CoV-2 clearance and immunological memory, yet their contribution to the persistence of post-acute sequelae of COVID-19 (PASC) remains poorly understood. The immunological features that distinguish individuals who develop PASC from those who recover fully are unresolved, in part due to the phenotypic heterogeneity of the condition and the likely multiplicity of its underlying mechanisms. Here, we profiled longitudinal bulk TCR$β$ repertoires from 120 individuals in the INCOV cohort--71 with PASC and 49 without--sampled at two to three time points spanning the acute and post-acute phases of infection. Using robust statistical modeling of repertoire composition and clonal dynamics, we found that global statistics such as V, J gene usage and CDR3 length do not differ between groups, but that locally enriched sequence motifs and differentially dynamic clones reveal distinct T-cell signatures associated with PASC status. Clones contracting following the peak of the acute response were significantly enriched for SARS-CoV-2 specificity in both groups. Interestingly, Influenza A-specific TCRs were disproportionately enriched among contracting clones in PASC{$^+$} repertoires, implicating viral co-infection as a potential contributor to early disease severity and, possibly, PASC pathogenesis. Rare public TCR clones were markedly enriched for SARS-CoV-2 specificity, with PASC{$^+$} individuals harboring a modestly but significantly higher proportion than PASC{$^-$} individuals. Together, we identified over 1,000 candidate TCR$β$ receptors potentially discriminating PASC{$^+$} from PASC{$^-$} immune responses, opening a path toward the identification of disease-relevant T-cell specificities and the development of T-cell-based immunological biomarkers for long COVID.

Knowledge graphs used for retrieval treat all facts as equally current. Existing temporal approaches apply uniform decay, using a single forgetting curve regardless of knowledge type. We show this is fundamentally misspecified: different knowledge types exhibit different temporal dynamics, and the core retrieval problem is not latency or throughput but identifying what is important at query time. We propose a hierarchical framework that replaces uniform decay with a continuous decay surface parameterized by two orthogonal signals: velocity (how frequently a concept is observed) and volatility (how much the value changes between observations, measured via embedding distance). The decay surface is decomposed into three learnable levels: domain-level parameters capture universal patterns (some predicates are inherently permanent, others inherently transient), context-level parameters capture setting-dependent variation, and entity-level adaptation personalizes decay to specific subjects. All parameters emerge from data through survival analysis on observed value lifetimes, requiring no predefined taxonomies or domain expertise. We formulate edge lifetime as a survival problem where the event is value supersession (a meaningfully different value replacing the current one), distinct from mere re-observation. Experiments on synthetic temporal knowledge graphs demonstrate recovery of planted hierarchical parameters (HDBSCAN ARI = 1.0). Validation on 107 Wikipedia articles and 1,163 patient records from the Synthea clinical EHR simulator shows that velocity-volatility clusters emerge naturally, align with observable persistence patterns, and near-universally exhibit the Lindy effect (Weibull shape k < 1). Uniform decay performs 18x worse than no temporal weighting. Heterogeneous decay recovers from this, with each hierarchy level contributing measurable improvement.

Modern machine learning optimizes for accuracy without explicit treatment of internal computational cost, even though physical and biological systems operate under intrinsic energy constraints. We evaluate energy-aware learning across 2,203 experiments spanning vision, text, neuromorphic, and physiological datasets with 10 seeds per configuration and factorial statistical analysis. Three findings emerge. First, architecture alone explains negligible variance in accuracy (partial eta^2 = 0.001), while the architecture x dataset interaction is large (partial eta^2 = 0.44, p < 0.001), demonstrating that optimal architecture depends critically on task modality and rejecting the assumption of a universal best architecture. Second, a controlled lambda-sweep across lambda in {0, 1e-5, 1e-4, 1e-3, 1e-2} validates a single-parameter energy-regularized objective L = L_CE + lambda * E(theta, x): across this range, internal activation energy decreases by approximately three orders of magnitude relative to the unregularized lambda=0 baseline, with negligible accuracy change (<0.5 percentage points) on both MNIST and Fashion-MNIST. Third, energy-first architectures inspired by an action-principle framework yield 5-33% within-modality training-efficiency gains over conventional baselines. These results emerge from a research program that interprets learning through a structural correspondence between the action functional in classical mechanics, free energy in statistical physics, and KL-regularized objectives in variational inference. We frame this correspondence as a design hypothesis, not a derivation.

Block-sequential continual learning demands that a single model both protect prior solutions from catastrophic forgetting and efficiently infer at inference time which prior solution matches the current input without task labels. We present Functional Task Networks (FTN), a parameter-isolation method inspired by structural and dynamical motifs found in the mammalian neocortex. Similar to mixture-of-experts, this method uses a high dimensional, self-organizing binary mask over a large population of small but deep networks, inspired by dendritic models of pyramidal neurons. The mask is produced by a three-stage procedure: (1) gradient descent on a continuous mask identifies task-relevant neurons, (2) a smoothing kernel biases the result toward spatial contiguity, (3) and k-winner-take-all binarizes the resulting group at a fixed capacity budget. Like mixture-of-experts, each neuron is an independent deep network, so disjoint masks give exactly disjoint gradient updates, providing structural guarantees against catastrophic forgetting. This three-stage procedure recovers the sub-network of a previously-trained task in a single gradient step, providing unsupervised task segmentation at inference time. We test it on three continual-learning benchmarks: (1) a synthetic multi-task classification/regression generator, (2) MNIST with shuffled class labels (pure concept shift), and (3) Permuted MNIST (domain shift). On all three, FTN with fine grained smoothing (FTN-Slow) results in nearly zero forgetting. FTN with a large kernel and only 2 iterations of smoothing (FTN-Fast) trades off some retention for increased speed. We show that the spatial organization mechanism reduces the effective mask search from the combinatorial top-k subset problem in O(C(H,K)) to the complexity of a near-linear scan in O(H) over compact cortical neighborhoods, which is parallelized by the gradient-based update.

We introduce RosettaSearch, an inference-time multi-objective optimization approach for backbone conditioned protein sequence design. We use large language models (LLMs) as a generative optimizer within a search algorithm capable of controlled exploration and exploitation, using rewards computed from RosettaFold3, a structure prediction model, under a strict computational budget. In a large-scale evaluation, we apply RosettaSearch to 400 suboptimal sequences generated by LigandMPNN (a state-of-the-art model trained for protein sequence design), recovering high-fidelity designs that LigandMPNN's single-pass decoding fails to produce. RosettaSearch's designs show improvements in structural fidelity metrics ranging between 18% to 68%, translating to a 2.5x improvement in design success rate. We observe that these gains in success rate are robust when RosettaSearch-designed sequences are evaluated with an independent structure prediction oracle (Chai-1) and generalize across two distinct LLM families (o4-mini and Gemini-3), with performance scaling consistently with reasoning capability. We further demonstrate that RosettaSearch improves the sequence fidelity of ProteinMPNN designs for de novo backbones from the Dayhoff atlas, showing that the approach generalizes beyond native protein structures to computationally generated backbones. We also demonstrate a multi-modal extension of RosettaSearch with vision-language models, where images of predicted protein structures are used as feedback to incorporate structural context to guide protein sequence generation. To our knowledge, this is the first large-scale demonstration that LLMs can serve as effective generative optimizers for backbone-conditioned protein sequence design, yielding systematic gains without any model retraining.

In the last century, most sensorimotor studies of cortical neurons relied on average firing rates. Rate coding is efficient for fast sensorimotor processing that occurs within a few seconds. Much less is known about the neural mechanisms underlying long-term working memory with a time scale of hours. Cognitive states may not have sensory or motor correlates. For example, you can sit in a quiet room making plans without moving or sensory processing. You can also make plans while out walking. In this perspective, I make the case for a possible second tier of neural activity that coexists with the well-established sensorimotor tier. The prominent physiological feature of the second tier is coordinated spike timing activity. The interplay of data supporting this hypothesis involves three puzzling yet highly intriguing experimental observations, without any obvious indication that they might actually represent different aspects of a single functional organization. First, consider the precision of spiking in individual neurons. The discovery of millisecond-precision spike initiation in cortical neurons was unexpected (Mainen and Sejnowski, 1995). Even more striking was the precision of spiking in vivo, in response to rapidly fluctuating sensory inputs. Second, high temporal resolution can also mediate spike timing-dependent plasticity (STDP) by controlling the relative timing of presynaptic and postsynaptic spikes at the millisecond scale. Third, we observe waves across many frequency bands traveling across the cortex. Strikingly, their timing is highly precise. Gamma waves, for example, which are triggered by attention, can plausibly trigger STDP that lasts for hours in cortical neurons. This temporary cortical network, ostensibly a second tier of functionality, rides astride the long-term sensorimotor network and could support cognitive processing and long-term working memory.

It is known that listeners lose the ability to discriminate the direction of motion of a revolving sound (clockwise vs. counterclockwise) beyond a critical velocity ("the upper limit"), likely due to degraded front-back discrimination. Little is known about how this ability is affected by simultaneously present distractor sounds, despite the real-life importance of tracking moving sounds in the presence of distractors. We hypothesized that the presence of a static distractor sound would impair the perception of moving target sounds and reduce the upper limit, and show that this is indeed the case. A distractor on the right was as effective as a distractor at the front in reducing the upper limit despite the likely importance of resolving front-back confusions. By manipulating the spectral content of both the target and distractor, we found that the upper limit was reduced if and only if the distractor spectrally overlaps with the target in the frequency range relevant for front/back discrimination. Our findings form the first steps toward a better understanding of the tracking of multiple sounds in the presence of distractors.

Adaptive control in biological systems, such as intestinal immunity, remains poorly understood despite detailed knowledge of underlying regulatory networks. We propose an alternative framework based on stochastic martingale turnover, in which cells proliferate through mutual competition and decay without cell-type-specific regulation. Through stochastic simulations and mathematical analysis, we show that this process autonomously generates balanced population compositions associated with low decay probabilities. The compositional dynamics can be described as a random walk whose step lengths decrease in low-decay regions. Reduced decay leads to larger total population sizes and an increase in the number of compatible microscopic states, which in turn shapes the distribution of compositions under fluctuating conditions. More generally, the dynamics follow a modified Langevin equation, in which constant mass is replaced by a fitness-dependent effective mass proportional to the total population size. Thus, biological systems regulate resistance to change, not merely direction, in shaping their macroscopic behavior.

Microbial populations exhibit a broad spectrum of nutrient utilization strategies, ranging from strategies utilizing diverse nutrients, called "generalists," to those being highly adapted to specific nutrients, called "specialists." The mathematical conditions for the diversification of nutrient utilization strategies are central questions in theoretical ecology. Previous studies have shown that trade-offs among different resource utilization functions that cells cannot utilize broad types of substrates at near-maximum speed are crucial for the emergence of diverse strategies. However, in natural settings, nutrient availability often fluctuates over time, imposing additional trade-offs on cells. Cells that grow rapidly under nutrient-rich conditions will suffer a higher death rate under nutrient-poor conditions, creating a growth-death trade-off that intersects with the classical resource-use trade-off. Here, we introduce a unified mathematical model that simultaneously incorporates the resource-use trade-off and the growth-death trade-off. The nutrient supply was modeled as discrete stochastic events, capturing realistic temporal fluctuations. We show that the relative balance between growth and death rates critically influences the dominance of either generalist or specialist strategies. Specifically, under conditions of high average growth rates among different environments and a weak trade-off between growth and death rates, generalists prevail. In contrast, when the growth-death trade-off is intense, specialists emerge as the dominant strategy. Our findings reveal that accounting for the growth-death trade-off is crucial for understanding how microbial communities adapt and evolve in temporally varying environments.

Biological evolution depends on the passing down to subsequent generations of genetic information encoding beneficial traits, and on the removal of unfit individuals by a selection mechanism. However, selection acts on phenotypes, and is affected by random contingencies. Thus, a combination of fitness and luck determines which individuals will successfully reproduce and give rise to the next generation. To understand how randomness in the selection mechanism affects the long-term patterns of evolution, we studied an idealized evolution model. We show through simulations and mathematical analysis, that the speed of adaptation increases with increasing selection pressure only up to a threshold. Beyond the threshold, any increase of the selection pressure results in more weight given to random effects rather than on genetic fitness in determining which individuals will successfully reproduce. This severely reduces the speed of adaptation and the diversity in the gene pool. Our findings may be considered as a biological instance of Goodhart's law: "When a measure becomes a target, it ceases to be a good measure". Finally, we show that this intricate response of evolution to natural selection can be mathematically explained by a novel phase transition for pulled traveling waves.

Early diagnosis and discovery of therapeutic drug targets are crucial objectives for effective management of Alzheimer's Disease (AD). Current approaches for AD diagnosis and treatment planning are based on radiological imaging and largely inaccessible for population-level screening due to prohibitive costs and limited availability. Recently, blood tests have shown promise in diagnosing AD and highlighting possible biomarkers that can be used as drug targets for AD management. Blood tests are significantly more accessible to disadvantaged populations, cost-effective, and minimally invasive. However, biomarker discovery in the context of AD diagnosis is complex as there exist important associations between various biomarkers. Here, we introduce BRAIN (Biomarker Representation, Analysis, and Interpretation Network), a novel machine learning (ML) framework to jointly optimize diagnostic accuracy and biomarker discovery processes to identify all relevant biomarkers that contribute to AD diagnosis. Using a holistic graph-based representation for biomarkers, we highlight their interdependencies and explain why different ML models identify different discriminative biomarkers. We apply BRAIN to a publicly available blood biomarker dataset, revealing three novel biomarker subnetworks whose interactions vary between the control and AD groups, offering a new paradigm for drug discovery and biomarker analysis for AD.