Papers

A foundational principle of connectionism is that perception, action, and cognition emerge from parallel computations among simple, interconnected units that generate and rely on neural representations. Accordingly, researchers employ multivariate pattern analysis to decode and compare the neural codes of artificial and biological networks, aiming to uncover their functions. However, there is limited analytical understanding of how a network's representation and function relate, despite this being essential to any quantitative notion of underlying function or functional similarity. We address this question using analysable two-layer linear networks and numerical simulations in non-linear networks. We find that function and representation are dissociated, allowing representational similarity without functional similarity and vice versa. Further, we show that neither robustness to input noise nor the level of generalization error constrain representations to the task. In contrast, networks robust to parameter noise have limited representational flexibility and must employ task-specific representations. Our findings suggest that representational alignment reflects computational advantages beyond functional alignment alone, with significant implications for interpreting and comparing the representations of connectionist systems.

Single-cell transcriptomics enables profiling of cellular states at unprecedented resolution, but its high dimensionality, sparsity, and technical batch effects pose significant challenges for representation learning. Existing single-cell foundation models typically encode each cell independently or only model cells from the same batch for denoising, thereby underutilizing the rich relational information across batches and cell types to model gene expression patterns. We argue that single-cell models can benefit from more informative cell-context modeling. By comparing consistency and variation across cells, models can capture fine-grained gene-gene dependencies associated with cell states, which are essential for learning high-quality representations. Inspired by the use of multiple sequence alignment (MSA) context in protein modeling, we propose CellMSA, a single-cell representation learning framework that introduces an MSA-inspired inductive bias into transcriptomic modeling. For each target cell, CellMSA retrieves relevant cells from different batches and biologically related cell types as context, and summarizes cross-cell patterns into a context-dependent gene-pair representation. This representation is then injected into a pair-aware target-cell encoder for fine-grained representation learning. We pretrain CellMSA on a large-scale human single-cell corpus of approximately 109 million cell observations, including 65.6 million primary observations. Experiments show that our framework consistently outperforms existing methods across multiple benchmarks. Code is available at the following repository: https://github.com/PharMolix/CellMSA.

Studying the alignment between the internal representations of vision models and the responses of the visual cortex to the same observed visual stimuli has enabled us to better understand human visual processing. However, studies so far have largely overlooked the fact that the human brain not only processes observed visual stimuli, but also predicts upcoming stimuli based on what has been observed. Accordingly, we hypothesize that internal representations for generating future video frames are better aligned with the predictive nature of human visual processing than representations of the observed video itself. To this end, we compare the alignment between human video-watching fMRI responses in the visual cortex and the internal representations from two types of video diffusion models, an autoregressive (AR) model and its non-AR base model. We first conduct a within-model analysis of the AR video diffusion model and show that the representations for future video generation align better with the visual cortex than the representations of the observed video. We then compare the internal representations of the AR model with those of its non-AR base model and again show that the representations for future video generation align better with the visual cortex than the representations for observed video reconstruction by the base model. Specifically, the alignment of observed video reconstruction is concentrated in lower-order visual cortex, whereas that of future video generation is concentrated in higher-order visual cortex. Finally, we show in a human behavioral experiment that humans prefer videos generated by amplifying the contributions of individual layers that align better with the visual cortex.

Tau and amyloid-$\beta$ (A$\beta$) deposits co-occur in Alzheimer's disease, and tau fibrils have recently been shown in vitro to catalyze A$\beta$ primary nucleation in a fold-specific manner. A two-compartment kinetic model is developed here, describing tau aggregation within the neuronal soma and A$\beta$ aggregation in the surrounding interstitium. The adjustable coupling parameter is the fraction f$_{acc}$ of somatic tau fibrillar material accessible to interstitial A$\beta$. Homotypic A$\beta$ secondary nucleation is restricted to the accessible surface of the consolidating plaque rather than scaling with total fibril mass, which preserves the in vitro calibration of the rate constant while preventing the surface term from growing without bound over decades. Three model predictions emerge. First, the model predicts that tau-catalyzed nucleation can increase the soluble A$\beta$ oligomer concentration and the number of A$\beta$ fibrillar species, by factors of 4.6 and 2.0, respectively, at complete accessibility, while leaving total A$\beta$ fibril mass and plaque volume unchanged, since in the absence of clearance these are fixed by the monomer supply; oligomer burden and visible plaque burden are therefore decoupled. Second, the predicted effect requires substantial accessibility: a 50 % increase in accumulated oligomer exposure needs f$_{acc}$ $\approx$ 0.035-0.15, which is difficult to reconcile with tangles enclosed by an intact neuronal membrane and points instead to neuronal lysis and the formation of extracellular ghost tangles. Third, physiologically relevant oligomer dissociation and proteolytic turnover each suppress accumulated exposure, by approximately three orders of magnitude and, at zero tau accessibility, about seventyfold, respectively, and remove the effect on a biological-age measure entirely.

Drug development is an expensive and lengthy process requiring developers to select drugs for human trials and diseases to target. Computational tools that can estimate before testing which drug-disease candidates are most likely to reach FDA approval could reduce development time and costs through early program advancement and disease prioritization. Yet existing approaches predict outcomes at a single level or stage or do not distinguish between disease outcomes for the same drug. We ask (1) whether FDA approval can be forecast for a drug-disease pair using only pre-clinical trial information, and (2) whether these forecasts can help prioritize diseases for the same drug. We built a cohort of 14,134 drug-disease pairs by reconstructing programs from ClinicalTrials.gov, extracting disease-level approvals from FDA label documentation, and enriching each pair with molecular, disease, ADMET, target, and pathway features. We divided the cohort into pre- and post-2019 programs and trained a classifier on pre-2019 data to predict FDA approval. On the post-2019 cohort (2,637 programs), the model would have prioritized a program that ultimately gained approval over one that did not approximately three out of four times. For drugs tested across multiple diseases, selecting the model's highest-ranked disease would have identified an ultimately approved disease 61.3% of the time, compared with 40.7% under random selection. These results show that public pre-trial information contains signal for forecasting FDA approval at the drug-disease level and can help prioritize diseases for the same drug. We provide a public web tool for SMILES- and ICD-10-based drug-disease approval forecasting.

Tetrahymena calcium-binding protein (Tcb2) forms Ca$^{2+}$-responsive networks that exhibit contractile behavior, yet how Ca$^{2+}$ concentration controls their local mechanical response remains poorly understood. Here, we use optical tweezers to perform active microrheology on reconstituted Tcb2 networks inside a microfluidic device that enables precise control of Ca$^{2+}$ concentration, allowing systematic tuning of network structure and mechanics. We find that increasing Ca$^{2+}$ from 1 to 100~mM enhances the effective stiffness by nearly two orders of magnitude, from $\sim$1$\times$10$^{-3}$ to $\sim$7$\times$10$^{-2}$~pN/nm, corresponding to a transition from a viscosity-dominated to a more elastic and mechanically robust network. Recoil assays further reveal rapid release of stored elastic energy following deformation. After two orthogonal pulls, the bead recoils along the diagonal rather than retracing the loading path, indicating that stresses from different directions combine to produce a resultant restoring response. These results establish Tcb2 networks as a minimal, tunable system for probing chemomechanical coupling and viscoelasticity in Ca$^{2+}$-regulated protein networks.

Connectome models of the fly nerve cord generate walking-like motor rhythms, but oscillation alone does not show that the specific wiring matters. Here we provide, to our knowledge, the first test of which features of motor output depend on the specific wiring. We simulated the leg motor systems of two independent Drosophila connectomes, with synapse counts as fixed weights and glutamatergic synapses treated as inhibitory, and compared each with six families of rewired networks that preserve progressively more of its structure, using pre-registered criteria. We find that rhythm is generic but antagonist coordination is not: many rewired networks were more rhythmic than the real ones, yet the real wiring coordinated antagonistic motor pools more strongly than every rewired network, most of all at the thorax–coxa joint. We trace this specificity to how premotor input is allocated between antagonistic pools. Both connectomes carry Sherrington's reciprocal innervation—neurons that excite one pool inhibit its antagonist—and no rewired network does. Reassigning premotor inputs between the pools abolished coordination even when motor neurons' typical input strength changed little (all pre-registered criteria met in one connectome; same direction in the other). Coordination, not rhythm, therefore reveals whether a connectome model's wiring matters.

Recent advances in musculoskeletal modeling and reinforcement learning have enabled muscle-actuated agents to reproduce increasingly complex human motions. Yet these capabilities remain largely confined to flat ground, in part because motion datasets rarely include aligned terrain geometry and because retargeting terrain interactions to complex musculoskeletal bodies is challenging. We present TERRA, an end-to-end pipeline for terrain-aware retargeting and control of musculoskeletal locomotion. From kinematic trajectories alone, TERRA combines terrain priors, estimated contacts, and negative free-space evidence to recover task-relevant support geometry. TERRA further considers anatomical, tendon-continuity, and contact constraints during retargeting. Using the resulting motion-terrain pairs from five datasets, we successfully train a single muscle-actuated control policy on 9.4 hours of diverse locomotion. Across reconstruction, retargeting, and held-out tracking benchmarks, TERRA improves terrain accuracy, sharply reduces anatomical and interaction violations, and achieves the highest observed completion rate over supported terrain families. Overall, TERRA provides a practical route from scene-less motion data to muscle-actuated locomotion over diverse non-flat terrain. Project website: https://cnai.epfl.ch/terra/

Accurate classification of viral families is essential to understanding genomic diversity, but rapid evolution and heterogeneous genome structure complicate reliable assignment. Newly discovered viruses, particularly those from underrepresented ecosystems, expand the known diversity of viral genomes. This growing diversity poses challenges for classification methods that rely on similarity to existing references. We present Fragment Path Complex Networks (FPCN), a framework that represents viral genomes as sequences of ordered fragments at various scales. FPCN captures contextual sequence information within each fragment and uses path complexes to model complex relationships between neighboring fragments. We evaluated FPCN on well-established NCBI benchmarks, and FPCN achieved higher classification accuracy than competing methods. Further evaluations on newly released datasets revealed FPCN's ability to generalize to genomes that were not presented in the training. Together, these results position FPCN as an effective tool for classifying viral genomes within known families across benchmark datasets and newly released genomes.

Tree-like branching structures are common in nature, from botanical trees to neurons, blood vessels and respiratory trees. Their branching shape often reflects function, making structural modelling central to understanding how these systems work. Because acquiring real-world 3D data is often expensive or infeasible, realistic generative models are valuable for simulation and data augmentation. Existing morphology-specific models either constrain how topology is generated or rely on hand-tuned, mechanistic procedures. Generic 3D graph generators, by contrast, do not exploit or enforce the structure of trees. We propose Autoregressive Frontier Expansion, a generative framework that constructs trees through an iterative expansion process, simulating the biological growth of real trees. At each step, a flow-matching model parameterised by an SO(2)-equivariant GNN expands the frontier by predicting whether each active branch bifurcates or terminates. We evaluate our method on cortical neurons and botanical trees in unconditional, class-conditioned, and morphology-guided generation. Across both domains, the generated morphologies agree closely with the reference distributions and, in conditional experiments, with the specified targets.

Global Neuronal Workspace Theory (GNWT) is one of the most extensively developed empirical research programmes on access consciousness. By contrast, the Motivated Emotional Mind (MEM) model proposes an embodied, semi-hierarchical associative memory in which representational selection, action, recurrent reconstruction of modality-specific fields, interoception, and valence form a single functional cycle. This article assesses whether MEM mechanisms can reproduce the functions explained by GNWT and whether they yield additional hypotheses concerning phenomenal quality. We first reconstruct the origins, mechanistic core, strengths, and limitations of GNWT, and then present the origins of MEM, the architecture of semblions, and the FFS-RP cycle. In the comparative analysis, we distinguish functional similarity from mechanistic identity and empirical equivalence. We examine 25 publications: twelve canonical studies of GNWT or critiques thereof, and thirteen studies of perception, interoception, and affect. For each study, we juxtapose the GNWT interpretation with a possible MEM reinterpretation. The findings indicate that most classical effects - including threshold dynamics, maintenance, recurrence, long-range connectivity, and flexible use of information - are compatible with both theories. This compatibility weakens the claim that the GNWT interpretation is exclusive, but does not yet provide theory-specific support for MEM. MEM advances a broader hypothesis integrating access, modality-specific reconstruction, bodily state, affect, and action; its potential explanatory advantage nevertheless remains a hypothesis requiring preregistered quantitative tests. We propose a series of interventions designed to dissociate sensory re-entry, global availability, report, and interoceptive valuation.

Retrosynthesis enables the discovery of viable synthetic routes to target molecules. It plays a central role in modern drug discovery and materials design. Retrosynthesis involves molecular graph transformations that can change both connectivity and graph size. These transformations may introduce reactant components absent from the target while revising the product-derived structure. To model these transformations, we propose RetroGEF, a flow-based generative model for single-step retrosynthesis. Starting from the target molecule, it constructs possible reactants by adding atoms and changing bonds in the molecular graph. RetroGEF models molecular transformations and changes in graph size within the same generative process, rather than relying on a fixed-size graph canvas. It learns this process directly from product–reactant pairs without requiring a prescribed edit order. Experiments on representative retrosynthesis benchmarks demonstrate that RetroGEF achieves state-of-the-art performance.

Teleoperation systems are tuned for channel fidelity, while whether the operator experiences the device as part of the body, the Sense of Embodiment (SoE), is measured only afterwards, by questionnaire. Predictive-processing accounts suggest controlling devices to reduce the mismatch between the operator's predictions and the returned feedback, but those predictions are unobservable, and an objective that only penalizes mismatch is minimized by removing feedback. We formulate an embodiment-aware controller, the Universal Embodiment Engine (UEE), that infers the operator's embodiment and visuo-proprioceptive cue weighting from implicit gaze and pupil signals and task outcome, and chooses bounded device settings under explicit preferences, cast as a discrete Active Inference agent. In simulations with 300 heterogeneous synthetic operators, the UEE found the suitable setting within half a minute for most operators, before identifying their exact type, and came close to an oracle in the second half of the session (embodiment 1.68 against 0.73 for the best fixed setting, on a 0-2 scale). Adapting without reading the operator did no better than fixed control, and model-free bandits did worse, whereas an expected-utility controller with the same inference did exactly as well: the benefit comes from Bayesian inference over the operator with explicit preferences, not from the information-seeking term of Active Inference. A naive prediction-error minimizer withheld feedback, as its objective implies, and lost task success (0.74 vs 0.90). The benefit shrank but persisted for operators outside the controller's model family, grew with the variety of the population, and vanished when the controller trusted an uninformative signal or when cue weighting changed mid-session without being modeled. These failures show what studies with people must establish first: calibrated signals and a model of change.

We study transient solute dispersion in pressure-driven micropolar flow through a concentric annulus representing a catheterized artery. The catheter is impermeable to solute, whereas the arterial wall combines irreversible removal with reversible surface retention. The exact steady micropolar velocity field is coupled to the Gill-Sankarasubramanian generalized-dispersion framework; the resulting bulk-surface hierarchy is advanced with Rannacher-damped Crank-Nicolson time stepping to determine $K_0(t)$, $K_1(t)$ and $K_2(t)$. The exchange coefficient is independent of the micropolar parameters and satisfies $-K_0(0^+)=2(\beta+\theta Da)/(1-\lambda^2)$, whereas $K_2-Pe^{-2}\sim\sigma_v^2t$ initially, so the leading short-time dispersion is hydrodynamic and independent of wall kinetics. In the weak-reaction regime, micropolarity modifies convection approximately linearly and shear dispersion quadratically through the velocity-amplitude factor. A narrow-gap analysis with $\varepsilon=1-\lambda$ gives $\bar v\sim(2-N_c)\varepsilon^2/6$ and $K_2-Pe^{-2}\sim(2-N_c)^2\varepsilon^6/7560$ for a fixed pressure gradient, revealing a sixth-power suppression of shear dispersion as the catheter approaches the arterial wall. Increasing $\lambda$ from $0.01$ to $0.30$ reduces effective convection by about a factor of $2.45$, while chemically passive shear-induced dispersion falls by a factor of $24$. The coupled formulation also yields the exact partition $\Phi_m+\Phi_s+\Phi_a=1$, separating mobile, reversibly retained and irreversibly absorbed solute. The reconstructed field quantifies reaction-modulated transverse non-uniformity, while the results distinguish hydrodynamic effects of confinement and microrotation from kinetic wall effects.

On a single task, deep networks can learn many solutions, depending on their optimizer, training data, architecture, and hyperparameters. Many of these solutions are mode-connected: rather than isolated points in weight space, they are connected by low-loss regions. Yet how their internal computation varies within these regions is unknown. A parallel line of work has identified the degeneracy of neural representations: many networks reach similar training loss with distinct internal structures. However, it is unclear how these solutions are related in weight space. We unify these subfields and show for the first time that many different internal mechanisms exist within a local mode-connected region in weight space. To do so, we introduce Hessian Null Space Continuation (HNC), a scalable method that uses local curvature to traverse regions of weight space that preserve network function, and can be steered toward solutions with specified properties. In RNNs trained on a memory task, HNC reaches drastically different representations and dynamics with maintained behavior. In ImageNet-trained Vision Transformers, HNC finds representations that differ more from the original network than any independently trained model with a different architecture or objective. In reinforcement-learning agents, HNC uncovers a distinct navigation strategy at comparable return and exposes reward hacking in an AI Safety Gridworld. Finally, HNC measures the local geometry of the solution set, showing how model size and task complexity shape its dimension and functional sensitivity. Our results show that a surprisingly large amount of representational diversity exists near a single trained solution, unseen by standard gradient-based optimization. HNC identifies and quantifies this diversity, opening new possibilities for mechanistic understanding of solution spaces and for model merging, editing, and fine-tuning.

Brain-AI alignment is often interpreted as a sign that model and brain perform similar computations. Whether the aligned units are causally involved in model computation is rarely checked. On an abstract pattern-completion task (AAABAAA $\rightarrow$ B), we compare LLM attention-head representations with human EEG and test how ablating those heads affects task performance. Alignment and causation dissociate: brain-aligned heads contribute to performance, but their removal is substantially less disruptive than removal of heads selected via attribution patching. We compare two head sets that prior interpretability work defines without reference to the brain: concept vectors (CVs), which represent abstract patterns across formats, and function vectors (FVs), selected for their contribution to correct-answer prediction. Brain alignment shows little association with FV scores, while its association with CV scores varies across models. Among brain-aligned heads, we find recurring attention profiles: one emphasizes distinctive elements (novelty heads), the other repeating elements (repetition heads). The novelty family tracks salience and attends to the same elements that humans look at, yet its removal is less damaging than random ablation on average. Repetition heads contribute modestly to performance and are associated with abstract-pattern representation (CVs). Across 17 models spanning 3B-72B parameters, FV-ranked removal is substantially more disruptive than brain-ranked removal. Brain alignment thus captures how the model reads the stimulus, and only faintly captures how it represents the pattern and solves the task.

Isolating vocalizations from noise in bioacoustics recordings is a prerequisite to many ecological analyses, including species identification, animal communication understanding, and individual or population-level variability studies. However, when recordings are acquired in open environments, vocalizations, noise, or signal-to-noise ratio can vary widely across individuals, species, environment, and recording conditions, making it hard to develop robust and generalizable methods for ecological analyses. To account for these challenges, noise reduction techniques are used to remove noise before downstream analyses. Popular methods such as Noisereduce rely on spectral gating, which estimates a noise threshold for each frequency channel. We propose a further improvement to Noisereduce, leveraging the fact that most animal sounds have structure across multiple frequencies. We apply our method on bird and marine mammals recordings and show that our extension of Noisereduce leads to improved denoising in both above and under water acoustic recordings, without impacting speed of preprocessing.

Flocking and schooling are thought to have evolved partly as defences against predation, but how prey should balance social and escape tendencies may depend on the predator's hunting strategy. We extend the predator-prey boids model of Ojo et al. (2023), itself based on Reynolds' boids, by combining six prey movement tendencies (alignment, cohesion, separation, dodge, repel and wiggle) into a single weighted acceleration update, and by reformulating wiggle as a sinusoidal manoeuvre. We then use an evolutionary strategy to optimise the six behaviour coefficients for collective prey survival against four predator hunting strategies: attack-centroid, attack-nearest, attack-random and attack-peripheral. Across five independent trials per strategy, coefficients converged within trials and mean fitness remained stable or increased, although trials often settled in different local optima. Prey survival was highest under attack-centroid and lowest under attack-nearest, in line with our hypotheses. Against attack-centroid, prey evolved individualistic predator avoidance with high escape coefficients, whereas against the other three strategies they largely kept their flock formation. Across all strategies, evolution favoured a low repel coefficient and relatively high dodge and wiggle coefficients. Our results suggest that optimal anti-predator behaviour depends on the interplay between escape tendencies and the predator's hunting strategy.

Nitrogen limitation is a key mechanism regulating the mutualism between photosymbiotic cnidarians and their algal symbionts. Algal-derived photosynthate facilitates host assimilation of inorganic nitrogen, specifically ammonium. The resulting competition between symbiotic partners for inorganic nitrogen leads to a negative feedback loop wherein symbiont density is effectively controlled through facilitation of their competitor. However, it is unclear whether this feedback loop is sufficient to stabilize the interaction absent other forms of regulation. Using a generalized modeling approach, we identify the conditions under which cnidarian-algal competition for nitrogen yields dynamically stable symbioses. We show that nitrogen limitation is sufficient to stabilize the interaction through the interplay of intraspecific (symbiont-symbiont) and interspecific (host-symbiont) competition. As host assimilation of inorganic nitrogen increases, the system shifts from being stabilized by intraspecific competition between symbionts to interspecific competition between host and symbionts, necessitating a positive coupling of symbiont density to host nitrogen assimilation via photosynthate translocation for dynamic stability.

Artificial Intelligence Virtual Cells (AIVCs) are envisioned as scientific agents that simulate cellular responses, explain underlying mechanisms, and support hypothesis-driven discovery. Existing AIVC benchmarks, however, operate primarily at the simulation layer, motivating complementary evaluation of how models interpret experimental evidence and formulate biological hypotheses. We introduce OmniVCBench, a figure-centric, source-traceable benchmark for the interpretation component of an AIVC. It contains 6,077 curated single- and multi-subfigure question–answer pairs derived from figures and experimental contexts in the scientific literature. Guided by Bloom's taxonomy, we instantiate interpretation-layer counterparts of the AIVC Predict–Explain–Discover agenda through three scientific reasoning tasks. We further introduce AIVC-Judge, a task-conditioned MLLM-as-a-judge framework with category-specific, reference-aware rubrics for evaluating open-ended responses. A complementary Model-Derived Hard-Negative Mining (MDHNM) strategy converts plausible errors observed during model inference into MCQ distractors for lower-cost evaluation. Within the evaluated heterogeneous model pool, MCQ accuracy correlates positively with AIVC-Judge scores, providing a complementary view of performance alongside open-response evaluation. Code and data demo are available at https://anonymous.4open.science/r/OmniVCBench.

Single-cell cancer datasets are predominantly cross-sectional and rarely provide paired or longitudinal observations linking individual healthy-like cells to tumor-associated states. We introduce CancerZigZag, a seed-initialized diffusion-based framework for exploratory generation of tumor-associated single-cell candidate clouds from unpaired epithelial cell populations. For each cancer context, a diffusion model is trained exclusively on tumor-derived epithelial cells and applied through repeated partial latent-space perturbation and reverse diffusion to held-out healthy-like seeds, generating stochastic candidate clouds without paired measurements or classifier guidance. We applied CancerZigZag to colorectal, breast, lung, and renal cell carcinoma contexts and explored parameter landscapes defined by perturbation depth and the number of ZigZag cycles. Across the reported operating configurations, candidate clouds contained outputs classified toward held-out tumor-derived reference populations for each evaluated seed. Residual seed-dependent organization varied across contexts, with the clearest structure in colorectal cancer, more modest organization in lung cancer, and limited cloud-level structure in breast and renal cell carcinoma. Representative candidates also showed directional concordance with transcriptional shifts observed between held-out healthy-like and tumor-derived reference populations. CancerZigZag is not interpreted as a model of deterministic healthy-to-tumor transformation or cellular progression. Instead, it provides a reference-informed framework for exploring tumor-associated candidate distributions from unpaired healthy-like seeds and quantifying the context-dependent relationship between tumor-associated displacement and residual seed dependence.

Phenotypic plasticity, the ability of cells to switch between states, is central to development, differentiation, and therapy resistance. Although it is modeled at several scales, from compartmental ODEs to phenotype-structured PDEs and single-cell stochastic equations, a framework connecting these descriptions is missing. We present such a framework. Starting from a general $n$-compartment ODE model encompassing nonlinear growth and linear transitions between phenotypes, we show, with a new, elementary and generalizable proof of a recent theorem, that under uniform competition, the long-term population distribution is solely governed by transition rates. All phenotypes become selectively neutral at saturation, and the imprint of fitness differences during growth fades at an explicit rate. Restricting transitions to neighboring states transforms the model into a discretization of a phenotype-structured reaction-diffusion-advection PDE. In this continuum model, diffusion and advection are identified from switching rates, and fitness remains asymptotically irrelevant. Interpreting the advection velocity as the negative gradient of an effective epigenetic potential transforms the PDE into a Fokker–Planck equation and converts single-cell trajectories into stochastic gradient descent (SGD) in the Langevin sense on the phenotypic landscape. Non-local transitions, such as mutations, are incorporated via an integro-differential term, yielding a unified model with reaction, gradient flow, diffusion, and jumps. State-dependent noise reshapes the effective landscape without altering the underlying potential. This gives cancer a route to elevated plasticity that static, single-cell snapshots cannot distinguish from a changed landscape. This framework provides a physical interpretation of Waddington's landscape, where cells perform SGD, and cancer corresponds to a corrupted landscape.

Protein Language Models (PLMs) have made remarkable progress following scaling laws established in natural language processing across sequence- and structure-based tasks, yet the potential of tokenization remains underexploited. Unlike human language, proteins preserve structure despite extensive sequence variation a property standard tokenization strategies fundamentally fail to capture. We introduce ZEST (Zoned Encoding of Sequence Traits), an evolution-informed vocabulary derived from conserved regions of multiple sequence alignments. ZEST allows embedding domain-level biological priors directly at the tokenization stage rather than learning them implicitly through scale. ZEST natively compresses sequences to an average token length of 4 residues, enabling our model to process 4,000 residues within a standard 1024-token context window. Building on this, we present LEMON (Layered Extraction of Molecular Ordering from Nature), a compact 200M-parameter sequence-based model for detection of remote homology between protein sequences trained on a single H100 GPU for one week. Despite its modest size, LEMON outperforms state-of-the-art models ranging from 600M to 3B parameters. Our results demonstrate that evolution-informed tokenization can substitute for massive parameter scaling, opening a new direction for efficient, biologically-grounded protein representation learning. All code, model weights, and results are publicly available under the MIT license.

Self-supervised pretraining reshaped prediction in language and vision, and brain foundation models (BFMs) inherited its promise. Representations learned from large unlabelled corpora should capture individual functional dynamics and generalise across cohorts. However, kernel ridge regression (KRR) fitted on functional connectivity (FC) matrices still predicts individual phenotypes more accurately than any BFM we tested. In this paper, we show that KRR is weighted by the eigenvalues of the FC which are miscalibrated for phenotype prediction. We apply an efficient spectral filter to recalibrate the eigenvalues of each subject's FC matrix, enabling the model to exploit more inter-individual variance. Across the 5 datasets, 11 parcellations and 6 prediction targets we tested, we match or exceed the KRR baseline. Based on this finding, we then pretrain a small encoder model on about 4,000 hours of fMRI from 162 open datasets, whereby we align the pairwise similarities between the embeddings of recording snippets with those between the recalibrated connectomes. Our model performs on par with the best of the 6 published BFMs we tested while having an order of magnitude fewer parameters. Our encoder performs better than FC on short scans and in smaller cohorts, especially in fingerprinting. We release the pretrained model weights, the code and the pretraining data, preprocessed and parcellated.

Drug discovery is a costly and high-risk process, where toxicity-related failures remain a major cause of attrition in both preclinical and clinical stages. As a result, accurate early prediction of chemical toxicity is essential to reduce downstream costs and improve compound prioritization. In this context, graph deep learning (GDL) has emerged as a powerful paradigm for toxicity prediction, leveraging molecular graph representations to learn directly from chemical structure with improved expressivity over traditional approaches. Despite the growing number of proposed models, current literature-based comparisons are often difficult to interpret due to inconsistencies in datasets, preprocessing pipelines, and evaluation protocols. To address this limitation, we introduce a unified and standardized benchmarking framework for GDL-based toxicity prediction. We systematically evaluate more than 20 representative approaches under consistent experimental conditions and across multiple datasets and partitioning strategies, enabling a fair and reproducible comparison of model performance. In addition, we complement this empirical study with a structured literature analysis to contextualize existing methodological trends and performance claims. Our results provide a clearer and more reliable assessment of the current state of the field, highlighting both the strengths and limitations of existing graph-based approaches. To support transparency and reproducibility, we release our benchmarking framework as open-source software https://gitlab.citius.gal/noel.suarez/benchtox, allowing the community to evaluate and compare models under consistent conditions.

Large-scale models have permeated the molecular sciences, yet what makes a model 'foundational' in this domain remains poorly defined. This paper proposes three testable criteria for assessing the foundational nature of molecular models: (i) generality across molecular entities, properties, and tasks; (ii) transferability to new applications with no or minimal task-specific retraining; and (iii) generalization beyond the training distribution. Applying these criteria to the state of the art reveals promising progress, particularly visible in biomolecular structure prediction and machine-learned interatomic potentials, although none of the approaches examined fully satisfies all three. Success is concentrated in domains where target properties are consistently defined and training data are abundant, with low label noise relative to physically meaningful variation. More broadly, progress in the molecular sciences appears to depend less on model scale alone than on the quality and structure of available data, as well as the incorporation of prior knowledge into models, prediction tasks, or downstream applications. This work shifts the notion of a molecular foundation model from a descriptive label to a testable hypothesis, offering a framework for assessing current models and guiding future developments.

Recent advances in biomolecular structure prediction have enabled accurate modelling of increasingly complex molecular systems. However, nucleic acid structure prediction remains challenging because of conformational flexibility and the limited availability of high-quality 3D structural data. Secondary structure (SS) provides a more readily available layer of structural information that captures base-pairing relationships and folding topology. Here we present OFoldNA, an all-atom diffusion model that incorporates SS information into nucleic acid folding and protein–nucleic acid co-folding. Without external SS information, OFoldNA achieved leading performance on FoldBench for both nucleic acid monomer folding and protein–nucleic acid co-folding, with particularly strong performance on DNA monomers and protein–DNA interfaces involving longer nucleic acid chains. When accurate base-pairing information was provided, OFoldNA-SS2TS further improved both folding and co-folding accuracy, while partial SS information also yielded consistent gains. The same auxiliary branch can also be used for RNA SS prediction as OFoldNA-SS, which achieved the best out-of-distribution performance on CHANRG. Together, these results show that intermediate structural information such as nucleic acid SS can be leveraged to improve all-atom 3D modelling, providing a general direction for incorporating complementary structural modalities into molecular structure prediction and design.

Cortical circuits must respond flexibly to new inputs while integrating information about the past, yet the way in which neural activity reconciles these competing demands remains unclear. Combining Neuropixels recordings from six mouse visual areas with mechanistic circuit modeling, we identify a dynamical regime in which heavy-tailed superdiffusive fluctuations coexist with long-range temporal dependence and oscillations. We formalize this regime as the adaptive fractional (AF) state, using an effective mean-field theory in which spatial and temporal fractional derivatives capture heavy-tailed excursions and long-range memory, respectively. We find that dynamical exponents characterizing the AF state vary systematically across the visual hierarchy: higher visual areas exhibit weaker superdiffusion and stronger temporal memory. In the circuit model, this hierarchical shift emerges from a progressive weakening of effective inhibition. These findings extend the classical hierarchy of timescales to a hierarchy of dynamical regimes and suggest that the AF state allows cortical areas to jointly express rapid responses and long-term temporal integration.

Laboratories often face a new molecular assay with 16-64 labels and a bank of predictors whose training data and parameters are unavailable. The practical question is which frozen outputs to include in a small local model. AssayRouter treats completed assays as pseudo-targets and labels each candidate by its post-fit utility: the reduction in held-out discovery loss when the candidate is added to the local target predictor. A shared regressor learns to predict this utility from candidate behavior on the support set, without source identity; on a new assay, one frozen ranking selects four sources and separate labels fit a convex combiner. We train only on completed ChEMBL-MT assays and evaluate 24 external regression assays across six frozen interface families. AssayRouter-C lowers strict four-call negative log-likelihood (NLL) by 0.0409 relative to Support-CV@4. Frozen candidate-label permutations confirm that candidate-utility correspondence carries the transferred information, and leave-one-interface-out training shows that the mapping generalizes to unseen predictor families. Completed assays therefore provide transferable supervision for scarce-label routing through frozen prediction interfaces.

Simulation-based inference is challenging when many heterogeneous observations must be composed, hierarchical latent structure must be preserved, and the simulator is misspecified relative to observed data. We develop sampling and fine-tuning methods for diffusion-based inference in design-conditional settings, where the same simulator is queried across different experimental conditions $\xi$. We extend compositional score-based inference with a continuous-time diffusion coefficient that accounts for the number of observations, avoiding Jacobian and auxiliary-covariance corrections. We introduce Hierarchical Blockwise Diffusion Sampling (HBDS), which infers shared parameters and group-specific latent states using a single pretrained model, with the hierarchy specified only at sampling time. Together, these methods support variable observation sets and groupings without retraining. To address misspecification, we introduce path-regularized fine-tuning that adapts the learned likelihood to observations and transfers corrections to posterior inference. Using Girsanov's theorem, we quantify path divergence between pretrained and fine-tuned models across experimental designs and interpret it alongside predictive errors to distinguish candidate misspecification correction from unnecessary adaptation. We evaluate compositional sampling on exact-score Gaussian and Simple Likelihood, Complex Posterior benchmarks, HBDS with analytic and learned scores on a controlled hierarchical model, and fine-tuning and localization on a separate analytic model with known design-dependent discrepancy. Finally, we apply the framework to 940 measurements across four cell lines in a mechanistic Bone Morphogenetic Protein signaling model, where fine-tuning improves posterior-predictive accuracy relative to the pretrained model and shifts posterior marginals toward the least-squares reference while retaining spread.