Papers

At the water-land interface, amphibious fishes encounter wet flowable substrates made of granular solid-water mixtures, which can stay solid or flow like a fluid. As these substrates become wetter or drier, their yield strength (at which solid-fluid transition occurs) and cohesion (how sticky they are) both change, challenging locomotion. Despite substantial understanding of tetrapod locomotion on flowable substrates (mostly dry sand), we know little about how amphibious fishes cope with wet flowable substrates of various wetness. Here, we studied mudskippers on clay mud of controlled, variable wetness over the range where solid-fluid transition occurs. As mud became wetter, its strength decreased by 100-fold, leading the animal to sink deeper, with larger areas of body and fins contacting mud. By contrast, mud stuck most easily at intermediate wetness. The increased sinkage and contact and stickiness change caused more mud to stick to and pull against the animal on wetter mud. We also tested dry mud, which stuck to animal fins as its mucus dried. Despite these challenges, the mudskipper predominately used a conserved crutching gait on all except the wettest mud tested, with a modest performance reduction. When normal crutching became less effective, the animal assisted it with tail thrusting, by bending and straightening it to push downward and backward to generate additional thrust and lift, or even thrusting the tail to jump. These observations suggest that mudskipper's crutching motor program is well adapted to its native muddy substrates but inflexible, with most novelty in tail use.

Genomic and metagenomic analyses play critical roles in many fields, such as precision medicine, urgent clinical settings, discovering early warnings of communicable diseases, ensuring food safety through pathogen monitoring, agriculture, and scientific discovery. Due to the challenges of analyzing and storing massive volumes of genomic and metagenomic sequence data, significant efforts have been made to accelerate (meta)genomic analyses and store sequence data compressed. Despite the benefits of these techniques, we identify two major outstanding problems in accessing stored sequence data and supplying it to the analysis units: (i) the data movement bottleneck due to moving large amounts of low-reuse data from storage and the unnecessary burden on the rest of the system, and (ii) the data preparation bottleneck, where compressed sequence data needs to be first decompressed and formatted before analysis. In this dissertation, we present customized storage-centric systems, which efficiently (i) analyze (meta)genomic data inside the storage system, and (ii) enable highly-compressed storage and high-performance access of large-scale sequence data, thereby alleviating the overheads of data movement, computation, and data preparation. We demonstrate that the proposed systems significantly improve system performance, energy efficiency, and system cost-efficiency of (meta)genomic analysis. We hope that the storage-centric systems proposed in this dissertation facilitate the broader adoption of (meta)genomic analyses and inspire future research to fundamentally improve the performance, energy efficiency, and cost-effectiveness of other data-intensive application domains related to health and life sciences.

Peptide-protein affinity models are often evaluated with a single data split, obscuring whether they interpolate among measurements for observed targets or generalize across peptide or target shifts. We integrated three sources of quantitative peptide-protein binding data to obtain 11,349 deduplicated pairs and benchmarked ten peptide representations, ESM-2 protein embeddings, and six regressors under peptide-similarity, within-target, and leave-target-out partitions. Across 60 matched representation-regressor configurations, mean test Spearman correlations were 0.462, 0.669, and 0.530, respectively. The top configuration shifted from ECFP-16 count fingerprints with random forest in the first two settings to HELM-BERT with Extra Trees when exact target sequences were excluded. Representation-rank correlations ranged from -0.042 to 0.624 across partitions, whereas regressor-rank correlations ranged from 0.771 to 0.943. Learning curves showed that representation differences were largest with limited supervision and narrowed as training data increased. PeptideCLM-2 adaptation and simple element-wise interaction features provided no consistent gain over a frozen encoder and direct concatenation under the tested protocols. These conclusions are specific to a dataset that pools transformed Kd, Ki, and IC50 measurements and to target exclusion at the exact-sequence level. Peptide-protein affinity benchmarks should therefore align data partitions with the intended use and jointly assess the effects of data scale, molecular representation, and downstream learner.

Scientific progress depends not only on finding solutions, but on learning the rules that explain why they work and using that understanding to design better experiments. We introduce science sandboxes, a framework for studying this capability in AI agents through repeated cycles of experimentation, feedback, and hypothesis revision. Science sandboxes invite an agent to query the natural world in different ways, ranging from "wet" physical experiments, to "damp" predictive models trained on empirical data, to "dry" invented rules. By establishing a common experimental loop and a protocol for evaluating agents within it, science sandboxes allow assessment of both quantitative performance on specific metrics and qualitative scientific reasoning, across a spectrum of empirical verifiability. Here, we instantiate this framework in two biological settings, models of regulatory genomics and protein fitness prediction, and examine the capabilities of frontier agents. Across these settings, we could see when agents successfully optimized a quantitative metric without understanding the rules underlying the system. In particular, their scientific reasoning deteriorated when they encountered systems whose rules fell outside familiar biological priors. By highlighting such failure modes, science sandboxes make the frontier of scientific capability measurable and provide a controlled setting in which to study and ultimately expand it.

Insect pollination is critical for global food production, yet monitoring pollinators at commercial farm scale remains a challenge. Recent advances in computer vision and deep learning have enabled detailed analysis of pollinator behaviour, but monitoring must trade-off detail against spatial coverage and human or technological resources. This paper presents the Automated Guided Robot for Insect and Crop Activity Monitoring (AGRICAM), a purpose-built robotic system designed to meet the requirements of large-scale pollination monitoring in protected cropping systems. AGRICAM operates autonomously on low-cost, easily installed track for movement along crop rows, without disrupting farm operations or insect behaviour. The platform integrates two RGB cameras, microclimate sensors, GPS and RFID modules, motion sensors, and 4G cellular network connectivity for data transmission. A web interface enables remote device configuration and scheduling. The system autonomously captures video and image data of insects' locations and local environmental conditions. These are transferred to the cloud and analysed using computer vision models to quantify pollinator visitation and spatio-temporal activity variation. We deployed the system on a commercial blueberry farm to demonstrate and test its capability. It successfully mapped insect pollination patterns across 80 m long industrial polytunnels over 30 hours. This data enabled spatial analyses of insect activity we used to confirm a uniform pollinator distribution within polytunnels, as desired by the farm management team. The data also highlighted variation of insect activity associated with time of day and microclimate. AGRICAM therefore has been shown to be a scalable, automated crop pollination monitor that can support data-driven decisions to enhance pollination management, thereby improving crop productivity and food security.

Spatial transcriptomics (ST) provides spatially resolved gene expression profiling but remains expensive, motivating the prediction of ST from histology images. Generative models have emerged as a mainstream paradigm for ST prediction due to their ability to model the conditional distribution of gene expression and capture its inherent stochasticity. However, these methods typically treat genes as independent prediction targets and overlook the intrinsic gene-gene interactions in biological systems, which limits their ability to preserve biologically meaningful co-expression patterns. We argue that gene-gene interactions, which reflect shared pathways and regulatory mechanisms, are essential for generating numerically accurate and biologically coherent ST profiles. In this paper, we propose CorrFlow, a correlation-guided flow matching framework for histology-to-ST prediction that explicitly models gene-gene dependencies through two complementary mechanisms. First, we introduce an annealed masked flow matching strategy, where subsets of genes are progressively masked following a timestep-dependent annealing schedule, encouraging the model to infer masked genes conditioned on the remaining genes and promoting joint conditional modeling beyond per-gene marginal estimation. Second, we devise a gene graph-regularized optimization scheme that integrates prior knowledge from the STRING database and data-driven co-expression estimated by WGCNA to construct a gene affinity graph, which enforces both local consistency and global smoothness in the predicted expression. Extensive experiments across 12 datasets show that CorrFlow achieves the best average PCC and HPCC among evaluated methods, leading to more biologically coherent ST predictions.

Cryo-electron microscopy (cryo-EM) is crucial for analyzing 3D biological structures, in which automated particle picking is essential for the workflow. However, fully supervised methods require extensive manual annotations. While few-shot learning offers a potential solution, existing approaches struggle to handle the diverse contaminations inherent in real micrographs owing to insufficient negative supervision, resulting in false positives that degrade the quality of the 3D reconstruction. Although synthetic data provides abundant and perfect labels, their use has primarily been restricted to validating identical proteins or augmenting full-shot training, leaving the potential for few-shot adaptation to novel proteins unexplored. In this study, we investigate the effective utilization of synthetic data for few-shot particle picking. We identify that direct transfer fails due to a “clean-vs-contaminated” Sim2Real gap. To overcome this, we propose CryoAnomaly, which is a framework that turns this gap into an advantage. By employing an anomaly detector trained on clean synthetic data, we identify real-world contaminants as anomalies and suppress them via a novel anomaly-guided hard negative suppression loss. On the CryoPPP benchmark, CryoAnomaly achieves the best picking accuracy and reconstruction resolution among state-of-the-art methods in the few-shot setting. The proposed anomaly-guided loss is confirmed to be effective on datasets with diverse contamination, where reliable pseudo-anomaly masks can be generated. Our code, dataset, and project page are available at: https://github.com/riku359/CryoAnomaly, https://huggingface.co/datasets/rikrikrik/CryoAnomaly, https://riku359.github.io/CryoAnomaly-page/.

AlphaFold 3 (AF3) enables structure prediction of biomolecular complexes through co-folding multiple interacting molecules, making it increasingly useful for de novo protein design and for large-scale studies of protein-protein, protein-peptide, and other biomolecular interactions. However, systematic co-folding experiments can produce large volumes of output data, particularly when multiple random seeds and samples are generated for each input complex. We introduce FoldKit, a Python package for efficient storage and analysis of large-scale AF3 co-folding results. FoldKit converts raw AF3 outputs into a compact, structured representation while preserving the metadata needed for downstream analysis. The FoldKit Python library provides convenient programmatic access to global, single chain, and interface confidence metrics such as pLDDT, pTM, ipTM, ipAE, and ipSAE, as well as an ensemble-level interface for accessing and aggregating these metrics for a single input across multiple seeds and samples. We benchmark FoldKit on three types of AF3 co-folding datasets: (i) a protein design campaign with 2 chains per input, (ii) a TCR-pMHC dataset with 4 chains per input, and (iii) a pooled-AF3 protein-protein interaction dataset with up to 22 chains per input. We find that FoldKit reduces storage requirements by approximately 5-15-fold compared to native AF3 outputs, depending on dataset composition, while maintaining direct programmatic access to individual predictions, ensembles, and confidence metrics. By reducing storage requirements and facilitating programmatic access to relevant outputs, FoldKit facilitates large-scale computational studies of biomolecular interactions. FoldKit is available from PyPI and can be installed using pip.

Neuromorphic computing is closely associated with spiking neuronal networks. However, an alternative class of so-called "rate-based" models arising from computational neuroscience and machine learning forgoes spiking interactions and instead relies on continuous coupling between neurons. Existing neuromorphic implementations designed around spike-based interactions are not well-suited for emulating such models. Here view the distributed simulation of these models as message-passing algorithms on parallel hardware. Leveraging prior art in numerical algorithms and distributed simulation, we outline steps that enable the design of efficient digital neuromorphic accelerators for non-spiking neuronal models. In particular, we show that multi-bit packets, rather than spikes, are the most efficient communication strategy in packet-switched networks and that compared to basic numerical integration methods, higher-order differential equation solvers decrease both computation and communication costs while achieving lower numerical error, but that these benefits are ultimately limited by arithmetic precision. Using our proposed design principles, we convert an existing neuromorphic architecture into a distributed numerical solver - a spikeless neuromorphic system - for continuously-coupled neuronal models. We thereby demonstrate that our theoretical considerations indeed translate into practical advantages, namely reduced energy consumption and delay.

Banana crop diseases threaten food security across the world, yet field diagnosis remains difficult because of limited expert access and visual similarity among disease classes. We present BananaVLM, a domain-adapted vision-language model for banana disease diagnosis built on LLaVA-v1.5-7B and fine-tuned with Low-Rank Adaptation (LoRA). To address the lack of multimodal agricultural instruction data, we introduce BananaInstruct, an automated pipeline that converts raw disease images into $\approx$80,000 question–answer pairs without manual annotation, using LLaVA-1.5-13B for symptom descriptions and Mistral-7B for agricultural Q&A generation and label grounding. Evaluated against 14 open-source and 5 closed-source VLMs on a nine-class benchmark, BananaVLM achieves 92.21% in-domain and 83.28% out-of-domain (OOD) classification accuracy, outperforming the best closed-source baseline, Gemini~2.5~Pro, by 49.8 and 63.3 percentage points, respectively. Binary healthy/diseased identification reaches 98.38% OOD accuracy with 1.00 recall. A controlled LoRA–DoRA comparison shows that DoRA performance degrades across epochs, while LoRA remains more effective for fine-grained classification. Qualitative evaluation using nine LLM judges yields G-Eval win rates of 0.81–0.96, while five domain experts prefer BananaVLM in 98.48% of 990 blind pairwise comparisons. These results show that lightweight domain adaptation with automated instruction tuning is an effective and scalable approach for specialized agricultural AI. Code, datasets, and model weights are available at https://github.com/samy101/banana-vlm

Sparse autoencoders (SAEs), and other interpretability methods could turn AI models in Biology and other fields into engines of scientific discovery by explaining the superhuman capabilities of those models. However, a latent is only useful if we know three things: whether it is coherent, whether it can be described, and whether that description has predictive power. These questions are routinely conflated. We assemble them into a single pipeline and report the practical innovations each stage required. First, cross-seed dictionary stability prioritises which latents are worth spending resources to investigate. Second, an intruder-detection task asks whether a latents activating examples share a recognizable pattern. Third, a separate pass proposes a candidate biological description which we convert into falsifiable predictions which can be tested in silico. Deployed on the Boltz-1 Pairformer trunk, stability prioritisation finds interpretable latents using about 4.4 times fewer latent evaluations each, and at 5.2 times lower measured cost, while recovering over half of them, and the external check shows the surfaced motifs are significantly enriched for their claimed annotations. The results also suggest a possible tension: the cross- seed stability might be selecting for some types of features, like structure-related ones, much more than others, such as function-related features.

Motivation: Accurate prediction of mutation-induced protein–protein binding free-energy changes is important for antibody affinity maturation, yet scarce labels and complex interface geometry limit generalization. Heterogeneous predictors may process three-dimensional complexes without preserving the cross-chain signals most relevant to a mutation in their final scalar output. Results: We introduce CIR-DDG, a lightweight residual adapter that combines a fixed base prediction with 22 interpretable descriptors of cross-chain distance, contact density and site–partner context. In complex-level five-fold evaluation on SKEMPI 2.0 measurements from 343 complexes, CIR-DDG improved all six tested backbones on antibody–antigen interface mutations: Spearman correlation increased by 0.0346–0.1296, while RMSE decreased by 0.0074–0.0408\kcalmol. Cross-validated probing, equal-capacity controls and feature ablations support the complementarity of explicit geometry. On an independent SARS-CoV-2 RBD–ACE2 deep-mutational-scan benchmark of 3669 substitutions, the fold-specific adapters transferred without any retraining: the absolute interface Spearman correlation increased by 0.026–0.081 for all four evaluable backbones, showing that the learned geometric correction generalizes beyond SKEMPI thermodynamic measurements. Availability and implementation: CIR-DDG is available at https://github.com/ecnuabmlab/CIR-ddG.

Many volume-based neuroimaging workflows require NIfTI label volumes, whereas a growing number of cortical parcellations are distributed in CIFTI surface space. We present Surf_2_Volume, a Python package for converting categorical CIFTI cortical parcellations into NIfTI volumes by resampling surface labels and filling a target cortical ribbon. The package provides configurable control over cortical-ribbon support and allows users either to specify the mapping threshold manually or to select suitable thresholds automatically according to the similarity between parcel-size distributions in the converted volume and the source CIFTI, without requiring a volumetric reference for threshold selection. We evaluated Surf_2_Volume using the Schaefer2018 100-parcel 7-network and 17-network atlases and compared its automatically selected outputs with multiple Connectome Workbench surface-to-volume mappings. Surf_2_Volume produced more continuous cortical coverage while avoiding the substantial underfilling observed with conservative Workbench mappings and the wider spatial expansion observed with more permissive nearest-vertex mappings. Across both atlases, the automatically selected Surf_2_Volume outputs showed consistently closer agreement with the paired reference volumes than the tested Workbench mappings. In HCP resting-state data, parcel time series extracted from Surf_2_Volume volumes also showed consistently high agreement with those obtained from the paired reference volumes. Together, these results support Surf_2_Volume as a configurable Python-based solution for converting CIFTI cortical parcellations to NIfTI volumes, with automatic threshold selection and stable performance across the two tested network parcellations.

Objective: Choosing between noninvasive respiratory support (NIRS) and invasive mechanical ventilation (IMV) for acute respiratory failure is a time-sensitive decision. Although guidelines provide population-level guidance, it remains unclear who benefits more from NIRS than IMV. The standard outcome, ventilator-free days at 28 days (VFD-28), scores death and prolonged ventilation equally, and current estimators do not distinguish between them. We developed and validated a censoring-aware Digital Twin framework for Individualized Treatment Effects of Non-Invasive Respiratory Support (DINIRS) to estimate individualized treatment effects (ITEs) that capture both mortality and ventilation duration. Materials and Methods: We emulated a target trial in 5,336 MIMIC-IV patients and trained DINIRS on 23 baseline clinical variables measured during the first 24 ICU hours. A transformer encoder with a survival attention gate decomposed VFD-28 into survival probability and conditional ventilation duration. A cross-fitted, doubly robust learner estimated ITEs. We externally validated DINIRS in 2,540 patients from the multi-site eICU-CRD dataset without retraining. Results: The DINIRS policy achieved a mean benefit of 2.07 ventilator-free days per patient (207 per 100 patients) compared with observed practice. Predicted NIRS benefit was higher among patients with less organ dysfunction (88.4% versus 49.0%) and persisted across hypoxemia severity. External validation reproduced this pattern. Discussion: The NIRS benefit stemmed from shorter ventilation among survivors rather than from reduced mortality, indicating that avoiding intubation-associated complications was the primary mechanism. Conclusion: Prospective validation is needed before these estimates inform treatment decisions. The decomposition framework can be extended beyond respiratory support to any zero-inflated composite outcome.

Protein language models organize sequence and structure at scale, but a global representation of how proteins respond to mutation remains lacking. We present RegimeFormer, a large protein perturbation model coupled to RegimeAtlas, constructed by harmonizing and indexing 202,556,313 non-redundant protein sequences across the tree of life. A diversity-preserving one-million-protein subset provides the high-resolution training and inference layer, with 995,995 proteins yielding residue-level summaries across 407,048,356 residues and substitution-specific predictions available on demand. Across experimental deep mutational scanning, molecular benchmarks, structural confidence and evolutionary constraint, RegimeFormer identifies reproducible protein-level perturbation regimes that organize residue fragility, adaptability and predictive uncertainty. Regime conditioning improves substitution-specific prediction, with the largest relative gains under unseen-protein, unseen-family and low-homology evaluation. RegimeFormer-derived molecular priors further improve downstream transcriptomic and drug-response modelling. Together, RegimeFormer and RegimeAtlas provide a scalable framework for mapping, predicting and querying protein perturbation landscapes across global sequence space.

Masked discrete diffusion models perform strongly on text, code, and biological sequences, but their training objective rewards only naturalness, and retraining the generator for every new reward is expensive. Inference-time steering of a frozen model either guides the sampler by the reward gradient or searches over several trajectories, and recent samplers combine the two. Such combinations are assembled as pipelines that leave three choices at their defaults: a guidance estimate resting on one Gumbel draw per sample, a reward tilting placed without reference to the distribution the combination then targets, and a selection temperature held fixed although the spread of per-step rewards drifts. We identify that distribution and settle the three choices against it. We therefore propose Variance-reduced Guidance and Adaptive Selection (VGAS), a simple yet effective inference-time framework that reduces the variance of the guidance estimate for both reward types, applies the reward tilting in the clean-token logits, where the pretrained schedule is preserved, and sets the selection temperature per step. Across regulatory DNA, protein and small-molecule benchmarks, VGAS attains the best training-free reward and matches or surpasses a reward-fine-tuned generator.

Identifying patients with similar clinical characteristics is important for cohort discovery, clinical decision support, and personalized medicine. However, it is unclear whether learned patient representations preserve the similarities present in the original clinical data. We propose a correspondence score to directly assess how well these relationships are preserved, independently of labeled outcomes. The score combines complementary relationship-preservation metrics, with dataset-specific weights estimated through Bayesian optimization. We evaluated the approach on four complete blood count datasets using representations generated by five tabular foundation models and six classical transformation methods. Correspondence differed substantially across methods and representation dimensionalities. Patient relationships were generally best preserved at intermediate dimensionalities, while stronger reduction or expansion introduced greater deviations from the original data structure. Foundation models preserved correspondence more consistently than classical methods, with improvements ranging from approximately 5% to over 125%. The proposed score provides a way to evaluate representation quality before downstream use and supports informed selection of representation methods and dimensionalities.

Protein language models (pLMs) encode information about protein sequences which enable downstream tasks such as structure prediction, but their internal representations are not well understood. Sparse autoencoders (SAEs) provide a promising tool to disentangle latent pLM representations into interpretable features, but existing annotation pipelines largely rely on protein-level annotations derived from database labels and LLM annotations of top activating sequences. Such annotations can overlook the localized residue-level and geometric patterns encoded by sparse features. We introduce an automated and scalable method for interpreting SAE features in ESM-2 by using geometrically inspired features of the protein $\text{C}_\alpha$ backbone. Across ESM-2 8M layers, an FDR-controlled discovery analysis shows that local geometry is significantly associated with many SAE features, with varying levels of predictive strength, expanding coverage beyond database and sequence-based methods. In particular, geometry can distinguish SAE features sharing the same database annotation, revealing substructure within known biological labels. A significant portion of SAE features activate on unannotated metagenomic protein sequences enabling us to use our SAE annotations to better understand these sequences. In addition, ablation experiments at the level of contact prediction show that removing found geometric features shifts ESM-2's predicted contact maps in the direction of the descriptor. This provides a robust method of annotating proteins activated within SAE neurons at a residue level, providing a bridge between mechanistic interpretability and structural biology.

With activity against multidrug-resistant pathogens and mechanisms distinct from conventional antibiotics, antimicrobial peptides (AMPs) offer a promising approach to combating antibiotic-resistant infections. However, their potency varies substantially across pathogen species, making accurate prediction of the minimum inhibitory concentration (MIC) for specific peptide-pathogen pairs essential for prioritizing candidates before costly experimental validation. Existing predictors are trained mainly on a few well-represented pathogens and rarely exploit biological relationships across species, limiting their generalization to low-resource and unseen pathogens. We introduce PathoMIC, the largest and most pathogen-diverse unified dataset for quantitative antimicrobial peptide activity prediction, containing 74,751 experimentally reported MIC measurements across 424 pathogen species. PathoMIC integrates peptide sequences, standardized MIC values, pathogen descriptions, and taxonomic relationships to facilitate knowledge transfer across related species. We establish few-shot and zero-shot cross-species evaluation protocols and develop a knowledge-enhanced framework that leverages pathogen descriptions and taxonomy. The framework yields substantial improvements for low-resource species with limited supervision, while gains for entirely unseen species remain modest, highlighting the difficulty of zero-shot cross-species MIC prediction. PathoMIC provides a standardized foundation for cross-species activity modeling and pathogen-specific virtual screening. Code is available at https://anonymous.4open.science/r/PathoMIC-546D/.

Dementia comprises biologically heterogeneous disorders, yet current risk assessment provides limited insight into how subtype-specific risk emerges and diverges before clinical diagnosis. We developed NetMoint, a multimodal framework integrating partially observed plasma proteomic, structural magnetic resonance imaging and cerebral haemodynamic phenotypes to predict individualized risks of Alzheimer's disease (AD), vascular dementia (VD) and frontotemporal dementia (FTD) across 1-, 5-, 10- and 20-year horizons. Among 104,120 UK Biobank participants free of dementia at baseline, NetMoint achieved mean area under the receiver operating characteristic curve (AUC) values of 0.937, 0.930 and 0.932 for AD, VD and FTD, respectively. The biological determinants of prediction shifted with time, from structural brain vulnerability at shorter horizons towards circulating molecular signatures at longer horizons, with distinct subtype-specific biological profiles. Multi-horizon risk profiling identified distinct temporal trajectories of dementia susceptibility. Among participants who subsequently developed AD, 0.7% followed a persistently very-high-risk trajectory, with predicted risk reaching 53.50% at 20 years, whereas 8.3% of those who developed FTD followed an increasing very-high-risk trajectory, reaching 67.17%. These high-risk trajectories were marked by distinct molecular signatures, with lower TGFB1 characterizing the AD group and higher NDRG1 the FTD group. In an independent ADNI-to-UK Biobank analysis, AD risk prediction remained informative after harmonization to 138 shared features, with an AUC of 0.741 at 20 years. Together, these findings establish a multimodal framework for trajectory-resolved dementia risk stratification, identifying small but high-risk populations within dementia subtypes and linking their divergent risk trajectories to distinct molecular signatures.

Wearable electrocardiogram (ECG) and photoplethysmogram (PPG) sensors are complementary but individually fragile: motion artifact, poor contact, and sensor dropout can degrade one or both signals. Fusion strategies that assume both modalities are equally trustworthy can become less reliable than a single clean modality under degradation. We present CardioFusion-AI, a framework whose signal-processing front end, including R-peak and systolic-peak detection, an Orphanidou-type signal-quality index, and beat-by-beat pulse transit time estimation, is validated on 53 real intensive-care recordings (848 windows; heart-rate mean absolute error 1.61 bpm for ECG and 2.78 bpm for PPG) and a real annotated fetal ECG database (R-peak F1 0.89-0.98). We then conduct a controlled synthetic degradation study comparing eight ECG-PPG fusion strategies across six degradation regimes spanning graded corruption and complete modality loss, using five independent training seeds. Attention fusion achieved the lowest descriptive overall error (1.66+/-0.43 bpm). Both adaptive gates reallocated weight toward the healthy modality under complete modality loss, but showed near-zero correlation between gate weight and signal quality under graded degradation (r = 0.10-0.24). Signal-quality conditioning produced a specific improvement under missing-PPG conditions (1.56+/-0.59 bpm), approaching the 1.48 bpm unimodal ceiling. With only five training seeds, no pairwise comparison survives Holm-corrected significance testing; effect sizes and confidence intervals are therefore reported. These results indicate that modality availability and modality quality are functionally distinct problems for adaptive fusion.

Continuous-time Markov chains (CTMCs) provide the backbone for modeling discrete stochastic dynamics across applied, physical, and biological sciences. Their integration with modern gradient-based machine learning, however, is limited by the hard categorical event selection intrinsic to Gillespie-type simulation algorithms. We exploit the affine state update to obtain the exact one-step conditional-mean sensitivity by differentiating normalized reaction propensities. We pair this backward rule with exact forward trajectories to define the propensity straight-through (PST) estimator. At the trajectory level, we show that one-step sensitivities composed across events can depart from the exact multistep sensitivity. We derive the resulting per-step discrepancy in closed form and prove that it vanishes identically for affine downstream dependence. PST matches the accuracy of Gumbel-Softmax straight-through across all benchmarks: reversible dimerization (0.06% error), a genetic oscillator (1.7% error), a 50-task repressilator suite (0.17% median error), and patch-clamp ion-channel recordings ($R^2$ = 0.988). Under matched settings, PST converges 3.0-fold faster on the oscillator and 2.1-fold faster on the ion channel. At deep-learning scale, PST trains a 203,796-parameter stochastic reaction network with hard sampling, reaching 98.22% MNIST digit classification accuracy. By differentiating an exact conditional mean rather than a relaxed sample, PST offers a temperature- and Gumbel-free path to scalable gradient-based learning through exact stochastic trajectories.

Recently, there has been a growing adoption of protein language models (PLMs) in biomedical science. Their embeddings provide a rich numerical representation of protein sequences which achieve state-of-the-art performance on several downstream tasks including protein fitness prediction. However, PLM embeddings are not directly interpretable and, thereby, it remains unclear what features they encode. To gain insight into which biochemical properties of the protein are driving the prediction, we leverage an orthogonal projection technique that removes linear effects of known tabular features from embeddings and extend it to high-order and interaction effects. In this way, we remove the effects of interpretable biochemical features from PLM embeddings. In an ablation study, we show that this leads to a decrease in performance for a downstream classifier trained only on the embeddings to predict protein fitness. In an additional evaluation, we find that these biochemical features explain a substantial part of the variance in the predictions of this classifier. Hence, we can show that PLM embeddings encode patterns correlated with biochemical properties and quantify their contribution to predicting protein fitness. This computationally efficient approach is not limited to the features or embeddings considered here and is readily transferable to problem settings beyond protein fitness prediction.

Artificial intelligence (AI) promises to accelerate biological research by automating computational analyses. Yet the ability of AI agents to execute on computational biology at the scale of complete research studies has not been systematically evaluated. Here we introduce BixBench3, a benchmark that measures the capacity of AI agents to process raw biological data through to scientific results. We designed BixBench3 tasks to mirror the delegation of work from a scientist to an agent: the scientist chooses the research question and high-level methods, then delegates implementation of all analyses to the agent. In each task, an agent receives a research objective, methodological guidance, and raw data derived from a published scientific study, and must execute a sequence of analyses to achieve the research objective. The data artifacts resulting from these analyses, such as peak call matrices or differential expression tables, are programmatically graded against the corresponding artifacts generated and reported in the original study. Across 20 BixBench3 tasks encompassing the generation of 138 unique artifacts, we find that 13 frontier models achieve scores ranging from 0.00 for Gemini 3.1 Flash Lite to 0.48 for GPT 5.6 Sol. Agents perform worse on tasks with larger raw datasets (0.36 on tasks with <100 GB versus 0.10 on tasks with >100 GB) and on analyses requiring more sequential steps (0.36 at 1-2 steps vs 0.24 at 3+). On average, agents use 6.8 hours, 102 million tokens, and \$43 to complete each task, with the longest attempts consuming 24 hours, 1.07 billion tokens, and \$525. Notably, the highest-scoring agents used fewer tokens and were cheaper than less performant options. These results reveal that LLMs vary substantially in their ability to (1) execute multiple sequential analysis steps coherently, (2) manage large quantities of raw data, and (3) work across scientific domains.

Histology and transcriptomics provide complementary views of tissue biology, capturing spatial morphology and molecular activity, respectively. Pathology foundation models (PFMs) learn rich morphological representations from H&E images, yet strong downstream performance alone does not establish whether these representations encode biologically meaningful and robust molecular information. We present a **training-free framework for auditing biological alignment in frozen PFMs** using spatially paired histology and transcriptomics from HEST-1k, evaluated on **240 samples spanning three organs**. Multiple frozen PFMs are used to extract H&E representations, while gene expression is aggregated into biologically interpretable pathway-level programs. We use TabPFN as a pretrained probe to quantify the extent to which these molecular programs can be decoded from frozen image representations without task-specific gradient updates. Beyond predictive performance, our audit examines whether pathway decodability generalizes across tissue sections, patient groups, and tissue types; whether representations exhibit section-level or other shortcut dependencies; and whether predictions remain stable under small image perturbations and context resampling. This multi-tissue evaluation distinguishes molecular programs that are consistently encoded from those that are tissue-specific, unstable, or shortcut-sensitive. Our framework therefore provides a systematic approach for assessing not only **what biological information pathology foundation models encode, but also how reliably that information persists under clinically relevant sources of variation**.

Natural language driven autonomous co-scientist workflows involve a fundamental trade-off between flexibility and reasoning at the expense of determinism, reproducibility, and observability. Such agents increasingly must communicate across institutional boundaries, where federation topology can shape latency and cost. We systematically evaluated these tradeoffs using a controlled ablation on a production agentic platform for science. We use a verifiable task: given a protein sequence, we ask an agent to confidently characterize its function by routing across common tools. We compare federation topology, classic RL vs LLM-driven harnesses, language model, and prompt expertise. We also stratify results by protein novelty. We find that the choice of LLM dominated prediction quality far more than topology or prompting (Opus ~92%-94% vs o4-mini ~40%-50%). The PPO policy was nearly as accurate as the best LLM (88%) at zero token cost, fastest latency, and perfect consistency, but yields no reasoning trace. Expert prompted LLMs reached the highest accuracy but were high-cost and less consistent; prompt dependence was largest when the task was hardest. Federation imposed a negligible penalty on performance. These results offer actionable guidance for deploying agents for scientific workflows: for routine, verifiable tasks, a cheap deterministic policy delivers near-frontier accuracy with complete reproducibility, while flexible LLM reasoning is best reserved for open-ended discovery.

RNA inverse folding asks for an RNA sequence whose prescribed secondary structure is the unique maximum-base-pair compatible fold. In the four-letter Watson-Crick model (A-U and C-G pairs only, no pseudoknots, and zero minimum base-pair span), Hales et al. introduced a separated-coloring certificate and an even-odd device, while Boury et al. generalized this to modulo-$m$ separability, gave an $O(n 2^m)$ decision algorithm, and guaranteed designability when every helix has length at least 3. We prove that the guarantee still holds when a motif-free target has at most two maximal helices of length 2, no maximal helix of length 1, and all remaining helices of length at least 3. The proof builds on Boury et al.'s local helix-coloring transfers and adds a global counting argument showing that the demands created by at most two short helices can always be coordinated. This is a structural success guarantee for the existing modulo-2 algorithm, not a new general decision capability. The resulting coloring yields an explicit sequence whose every distinct compatible noncrossing fold has fewer pairs. No claim is made for nearest-neighbor thermodynamic energy models. The theorem and supporting lemmas are formalized in Lean 4 against pinned Mathlib and reproduced from a frozen public artifact; the kernel-reported axiom set is $\{\mathrm{propext},\mathrm{Classical.choice},\mathrm{Quot.sound}\}$. The work was developed with foundational generative-AI assistance under the author's direction and has not yet received independent human expert review.

Frozen hematology foundation-model (FM) embeddings reach near-saturated in-domain white-blood-cell (WBC) accuracy, but clinical deployment demands reliability across scanners, sites, stains and preparation pipelines. We audit 15 frozen encoders (hematology, pathology, and general vision) across four public single-cell acquisition domains along two axes: accuracy robustness and calibration. In-domain linear-probe macro-F1 is saturated (0.98-0.997), yet cross-dataset macro-F1 drops 34-72% and rankings re-order: DinoBloom-L, the in-domain best, falls to 10th of 15 on the most-shifted target (MLL23) at the benchmark's shared 224-px input, behind RedDino and several general and pathology encoders. Rank transfer is probe-dependent: 1-NN retrieval is more stable on average than a source-fitted linear head (median $\rho$ 0.65 vs 0.45), but neither probe universally predicts target robustness. Calibration also collapses: source-trained probes are nearly calibrated in-domain (expected calibration error, ECE, 0.004) but confidently wrong off-domain (ECE 0.35), and source-fitted temperature scaling transfers poorly. We further audit pretraining exposure and identify MLL23 as DinoBloom's internal cohort; because DinoBloom's only held-out dataset is also our source domain, this benchmark cannot isolate exposure from scanner-associated shift. Label-free adaptation and marginal-entropy-based model selection appear safe under balanced evaluation but fail under realistic WBC class-prior shift. Class-Balanced Re-standardization (CBR), a training-free pseudo-label-balanced feature normalization, improves all evaluated target-prior scenario means and partially improves calibration, although encoder-level exceptions and residual miscalibration remain. Hematology FM benchmarks must therefore jointly audit accuracy, calibration, exposure, and class-prior robustness.

Prenatal maternal distress has been linked to differences in offspring brain and behavioural development, but is rarely framed computationally. Predictive-processing (PP) accounts hold that perception, cognition, and action arise from the interplay of incoming input and internally generated predictions, governed in part by precision. We reappraised eleven publications from two non-clinical prenatal cohorts for a coherent PP-compatible developmental pattern. Eight Leuven publications spanned adolescence to adulthood: externally cued response inhibition was largely preserved, whereas differences emerged when task goals, stimulus relevance, expected outcomes, and response policies had to be generated or maintained endogenously. At age 28, higher prenatal anxiety was associated with lower radial and mean diffusivity in the left superior posterior corona radiata, lower Vocabulary scores, weaker medial prefrontal to left inferior frontal connectivity, and greater unpleasantness of induced dyspnoea. Three Tilburg ERP publications covered infancy and early childhood: higher prenatal anxiety was associated with increased processing of repeated standard tones, of fearful over happy vocalizations, and of neutral over pleasant or unpleasant pictures. Maternal mindfulness, the only positive prenatal characteristic examined, showed the opposite association for the repeated standard, suggesting that positive and distress-related states may act differently on shared mechanisms. Across cohorts, findings were compatible with differences in how relevance and processing resources are allocated to predictable, threat-related, ambiguous, contextual, feedback-related, and bodily information. Because these publications did not formally estimate precision, prediction errors, or priors, we treat them as hypothesis-supporting rather than a formal test of the account.

Spatially resolved biology requires representations that preserve biological neighborhood structure rather than only exact cross-modal correspondences. Existing histology–transcriptomics objectives can emphasize instance-level matching even when non-paired spots share molecular or spatial context. We introduce BioKERN, a multimodal spatial representation-learning framework that incorporates biological structure as an explicit, learnable inductive bias. BioKERN constructs a training-time biological kernel by combining transcriptomic similarity and spatial proximity, then uses it to provide graded neighborhood supervision and regularize embedding geometry. Evaluation uses a fixed, model-independent biological neighborhood definition shared by all methods. Across Mouse Brain Visium and Human Liver GSE240429, BioKERN consistently improves biological-neighborhood retrieval over BLEEP in both single- and multi-scale settings. Controlled shared-architecture experiments show that most of the improvement arises from biological-kernel regularization rather than increased model capacity. These results support explicit biological geometry as an interpretable inductive bias for multimodal learning in spatial biology.