Machine-learning interatomic potentials (MLIPs) have advanced rapidly, with many top models relying on strong physics-based inductive biases. However, as models scale to larger systems like biomolecules and electrolytes, they struggle to accurately capture long-range (LR) interactions, leading current approaches to rely on explicit physics-based terms or components. In this work, we propose AllScAIP, a straightforward, attention-based, and energy-conserving MLIP model that scales to O(100 million) training samples. It addresses the long-range challenge using an all-to-all node attention component that is data-driven. Extensive ablations reveal that in low-data/small-model regimes, inductive biases improve sample efficiency. However, as data and model size scale, these benefits diminish or even reverse, while all-to-all attention remains critical for capturing LR interactions. Our model achieves state-of-the-art energy/force accuracy on molecular systems, as well as a number of physics-based evaluations (OMol25), while being competitive on materials (OMat24) and catalysts (OC20). Furthermore, it enables stable, long-timescale MD simulations that accurately recover experimental observables, including density and heat of vaporization predictions.

Designing messenger RNA (mRNA) sequences for a fixed target protein requires searching an exponentially large synonymous space while optimizing properties that affect stability and downstream performance. This is challenging because practical mRNA design involves multiple coupled objectives beyond classical folding criteria, and different applications prefer different trade-offs. We propose a general sampling-based continuous optimization framework, inspired by SamplingDesign, that iteratively samples candidate synonymous sequences, evaluates them with black-box metrics, and updates a parameterized sampling distribution. Across a diverse UniProt protein set and the SARS-CoV-2 spike protein, our method consistently improves the chosen objective, with particularly strong gains on average unpaired probability and accessible uridine percentage compared to LinearDesign and EnsembleDesign. Moreover, our multi-objective COMBO formulation enables weight-controlled exploration of the design space and naturally extends to incorporate additional computable metrics.

Understanding how neural networks transform inputs into outputs is crucial for interpreting and manipulating their behavior. Most existing approaches analyze internal representations by identifying hidden-layer activation patterns correlated with human-interpretable concepts. Here we take a direct approach to examine how hidden neurons act to drive network outputs. We introduce CODEC (Contribution Decomposition), a method that uses sparse autoencoders to decompose network behavior into sparse motifs of hidden-neuron contributions, revealing causal processes that cannot be determined by analyzing activations alone. Applying CODEC to benchmark image-classification networks, we find that contributions grow in sparsity and dimensionality across layers and, unexpectedly, that they progressively decorrelate positive and negative effects on network outputs. We further show that decomposing contributions into sparse modes enables greater control and interpretation of intermediate layers, supporting both causal manipulations of network output and human-interpretable visualizations of distinct image components that combine to drive that output. Finally, by analyzing state-of-the-art models of neural activity in the vertebrate retina, we demonstrate that CODEC uncovers combinatorial actions of model interneurons and identifies the sources of dynamic receptive fields. Overall, CODEC provides a rich and interpretable framework for understanding how nonlinear computations evolve across hierarchical layers, establishing contribution modes as an informative unit of analysis for mechanistic insights into artificial neural networks.

The spatial Muller's ratchet is a model introduced by Foutel-Rodier and Etheridge to study the impact of cooperation and competition on the fitness of an expanding asexual population. The model is an interacting particle system consisting of particles performing symmetric random walks that reproduce and die with rates that depend on the local number of particles. For each particle, we keep track of the number of deleterious mutations that it carries, and after each birth event, with some positive probability, the offspring particle can acquire an additional mutation that gives it a lower reproduction rate than its parent. We show that under an appropriate scaling, the process converges weakly to the solution of an infinite system of partial differential equations (PDEs), confirming non-rigorous computations of Foutel-Rodier and Etheridge. In the PDE limit, when the reaction term of the system of PDEs is monostable, we establish bounds on the ratio between the density of particles with a given number of mutations and the density of particles without mutations. If the reaction term satisfies a Fisher-KPP condition, we can also rigorously determine the spreading speed of the population into an empty habitat. Finally, by considering the PDE limit of a form of tracer dynamics, we answer the question of whether deleterious mutations can surf population waves in this setting.

Saliency maps are increasingly used as design guidance in siRNA efficacy prediction, yet attribution methods are rarely validated before motivating sequence edits. We introduce a pre-synthesis gate: a protocol for counterfactual sensitivity faithfulness that tests whether mutating high-saliency positions changes model output more than composition-matched controls. Cross-dataset transfer reveals two failure modes that would otherwise go undetected: faithful-but-wrong (saliency valid, predictions fail) and inverted saliency (top-saliency edits less impactful than random). Strikingly, models trained on mRNA-level assays collapse on a luciferase reporter dataset, demonstrating that protocol shifts can silently invalidate deployment. Across four benchmarks, 19/20 fold instances pass; the single failure shows inverted saliency. A biology-informed regularizer (BioPrior) strengthens saliency faithfulness with modest, dataset-dependent predictive trade-offs. Our results establish saliency validation as essential pre-deployment practice for explanation-guided therapeutic design. Code is available at https://github.com/shadi97kh/BioPrior.

Electroporation is increasingly used as a percutaneous ablation technique for tumours located near vital structures. Although effective, tumour regrowth may still occur. At the same time, in vitro studies on cell monolayers have shown that electroporation can trigger immunogenic cell death (ICD) through the release of damage-associated molecular patterns (DAMPs). These molecules can stimulate the immune system and could counteract tumour regrowth. To fully exploit electroporation, two key questions must be addressed: (1) what dynamics drive tumour regrowth, and (2) how ICD unfolds in space and time within three-dimensional cellular structures, which better mimic in vivo conditions than 2D cultures. Here, we combine in vitro experiments with a hybrid individual-based/continuous computational model to explore tumour spheroid regrowth and ICD potential under different pulse intensities. Experiments quantify spheroid viability, growth rate, and the release of ATP and HMGB1. In parallel, the hybrid model predicts the dynamics of proliferative, quiescent, and necrotic cells. Both approaches show that cell death and DAMP release scale with pulse intensity. The model, validated against experimental data, further highlights the dual role of quiescent cells: some die and free space and resources, while others survive and resume proliferation. Together, these findings demonstrate how spheroid fate depends on pulse strength and emphasize the importance of accounting for quiescent cells when designing electroporation-based therapies.

With significant population fractions in many societies who refuse vaccines, it is important to reconsider how vaccination is incorporated into compartmental epidemiology models. It is still most common to apply the vaccination rate to the entire class of susceptibles, rather than to use the more realistic assumption that the vaccination rate function should depend only on the population of susceptibles who are willing and able to receive a vaccination. This study uses a simple generic disease model to address two questions: (1) How much error is introduced in key model outcomes by neglecting vaccine unwillingness?, and (2) Can the error be reduced by incorporating vaccine unwillingness into the vaccination rate constant rather than the rate diagram? The answers depend greatly on the time scale of interest. For the endemic time scale, where longterm behavior is studied with equilibrium point analysis, the error in neglecting unwillingess is large and cannot be improved upon by decreasing the vaccination rate constant. For the epidemic time scale, where the first big epidemic wave is studied with numerical simulations, the error can still be significant, particularly for diseases that are relatively less infectious and vaccination programs that are relatively slow.

Brain-wide recordings of large-scale networks of neurons now provide an unprecedented view into how the brain drives behavior. However, brain activity contains both information directly related to behavior as well as the potential for many internal computations. Moreover, observable behavior is executed not only by the brain, but also by the spinal cord and peripheral nervous system. Behavior is a coarse-grained product of neural activity, and we thus take the view that it can be best represented by lower-dimensional latent neural dynamics. Capturing this indirect relationship while disambiguating behavior-generating networks from internal computations running in parallel requires new modeling approaches that can embody the parallel and distributed nature of large-scale neural populations. We thus present behavior-decomposed linear dynamical systems (b-dLDS) to disentangle simultaneously recorded subsystems and identify how the latent neural subsystems relate to behavior. We demonstrate the ability of b-dLDS to decouple behavioral vs. internal computations on controlled, simulated data, showing improvements over a state-of-the-art model that uses behavior to supervise all dynamics based on behavior. We then show that b-dLDS can further scale up to tens of thousands of neurons by applying our model to large-scale recording of a zebrafish hindbrain during the complex positional homeostasis behavior, wherein b-dLDS highlights behavior-related dynamic connectivity networks.

Multiple Sclerosis (MS) is a chronic autoimmune disease of the central nervous system whose molecular mechanisms remain incompletely understood. In this study, we developed an end-to-end machine learning pipeline to analyze transcriptomic data from peripheral blood mononuclear cells and cerebrospinal fluid, integrating both bulk microarray and single-cell RNA sequencing datasets (concentrating on CD4+ and B-cells). After rigorous preprocessing, batch correction, and gene declustering, XGBoost classifiers were trained to distinguish MS patients from healthy controls. Explainable AI tools, namely SHapley Additive exPlanations (SHAP), were employed to identify key genes driving classification, and results were compared with Differential Expression Analysis (DEA). SHAP-prioritized genes were further investigated through interaction networks and pathway enrichment analyses. The models achieved strong performance, particularly in CSF B-cells (AUC=0.94) and microarray (AUC=0.86). SHAP gene selection proved to be complementary to classical DEA. Gene clusters identified across multiple datasets highlighted immune activation, non-canonical immune checkpoints (ITK, CLEC2D, KLRG1, CEACAM1), ribosomal and translational programs, ubiquitin-proteasome regulation, lipid trafficking, and Epstein-Barr virus-related pathways. Our integrative and explainable framework reveals complementary insights beyond conventional analysis and provides novel mechanistic hypotheses and potential biomarkers for MS pathogenesis.

Collaborative training across multiple institutions is becoming essential for building reliable medical image segmentation models. However, privacy regulations, data silos, and uneven data availability prevent hospitals from sharing raw scans or annotations, limiting the ability to train generalizable models. Latent-space collaboration frameworks such as privacy-segmentation framework (SF) offer a promising alternative, but such methods still face challenges in segmentation accuracy and vulnerability to latent inversion and membership-inference attacks. This work introduces a privacy-preserving collaborative medical image segmentation framework (PPCMI-SF) designed for heterogeneous medical datasets. The approach combines skip-connected autoencoders for images and masks with a keyed latent transform that applies client-specific orthogonal mixing and permutation to protect latent features before they are shared. A unified mapping network on the server-side performs multi-scale latent-to-latent translation, enabling segmentation inference without exposing raw data. Experiments on four datasets: PSFH ultrasound, ultrasound nerve segmentation, FUMPE CTA, and cardiac MRI show that the proposed PPCMI-SF consistently achieves high Dice scores and improved boundary accuracy, as reflected by lower 95th percentile Hausdorff distance (HD95) and average symmetric surface distance (ASD) compared to the current state-of-the-art and performs competitively with privacy-agnostic baselines. Privacy tests confirm strong resistance to inversion and membership attacks, and the overall system achieves real-time inference with low communication overhead. These results demonstrate that accurate and efficient medical image segmentation can be achieved without compromising data privacy in multi-institution settings.

Confounding pathology with normal anatomical variation remains a significant challenge in unsupervised medical-image anomaly detection, resulting in numerous false positives. To enhance integration of healthy variation, we augment the latent representation of a CNN autoencoder with contextual similarities within a normal cohort through batch-wise hypergraph estimation and a shared-weights graph convolution layer, producing a population-aware embedding. On a heterogeneous brain-tumor dataset of 2D MRI scans, the method improves separability between healthy and pathological samples, achieving an AUC-ROC of 0.90 (95% CI 0.84-0.95, 5.7% absolute gain), and a 16% absolute improvement in average precision (0.78 AP, 95% CI 0.66-0.89), thereby lowering false-positive rates. Moreover, both anomaly detection and downstream tumor versus no-tumor classification performance improve with the size of the mini-batch context captured in the augmented representation, suggesting a tunable lever for integrating healthy variation.

When biological foundation models such as scGPT and Geneformer process single-cell gene expression, what geometric and topological structure forms in their internal representations? Is that structure biologically meaningful or a training artifact, and how confident should we be in such claims? We address these questions through autonomous large-scale hypothesis screening: an AI-driven executor-brainstormer loop that proposed, tested, and refined 141 geometric and topological hypotheses across 52 iterations, covering persistent homology, manifold distances, cross-model alignment, community structure, and directed topology, all with explicit null controls and disjoint gene-pool splits. Three principal findings emerge. First, the models learn genuine geometric structure. Gene embedding neighborhoods exhibit non-trivial topology, with persistent homology significant in 11 of 12 transformer layers at p < 0.05 in the weakest domain and 12 of 12 in the other two. A multi-level distance hierarchy shows that manifold-aware metrics outperform Euclidean distance for identifying regulatory gene pairs, and graph community partitions track known transcription factor target relationships. Second, this structure is shared across independently trained models. CCA alignment between scGPT and Geneformer yields canonical correlation of 0.80 and gene retrieval accuracy of 72 percent, yet none of 19 tested methods reliably recover gene-level correspondences. The models agree on the global shape of gene space but not on precise gene placement. Third, the structure is more localized than it first appears. Under stringent null controls applied across all null families, robust signal concentrates in immune tissue, while lung and external lung signals weaken substantially.

Rare disease diagnosis requires matching variant-bearing genes to complex patient phenotypes across large and heterogeneous evidence sources. This process remains time-intensive in current clinical interpretation pipelines. To overcome these limitations, We present LA-MARRVEL, a knowledge-grounded, language-aware LLM framework and designed for clinical robustness and practical deployment. LA-MARRVEL delivers a 12-15 percentage-point absolute improvement in Recall@1 over established gene prioritization approaches, showing that architectural design can drive substantial accuracy gains. We found that the central contributor is structured, phenotype-rich prompt construction that explicitly encodes patient and disease phenotypes, preserving clinically meaningful context more effectively than disease labels alone. Across three real-world cohorts, LA-MARRVEL consistently improves gene-ranking performance, including in challenging cases where the causal gene was initially ranked lower by first-stage prioritization. For each candidate gene, the system delivers clinically relevant, ACMG-aligned reasoning that integrates phenotype concordance, inheritance patterns, and variant-level evidence into auditable explanations, enabling streamlined clinical review. These findings suggest that knowledge-grounded LLM layer can enhance existing rare-disease gene prioritization workflows without altering established diagnostic pipelines.

T cell receptor (TCR) recognition of peptide-MHC (pMHC) complexes is a central component of adaptive immunity, with implications for vaccine design, cancer immunotherapy, and autoimmune disease. While recent advances in machine learning have improved prediction of TCR-pMHC binding, the most effective approaches are black-box transformer models that cannot provide a rationale for predictions. Post-hoc explanation methods can provide insight with respect to the input but do not explicitly model biochemical mechanisms (e.g. known binding regions), as in TCR-pMHC binding. ``Explain-by-design'' models (i.e., with architectural components that can be examined directly after training) have been explored in other domains, but have not been used for TCR-pMHC binding. We propose explainable model layers (TCR-EML) that can be incorporated into protein-language model backbones for TCR-pMHC modeling. Our approach uses prototype layers for amino acid residue contacts drawn from known TCR-pMHC binding mechanisms, enabling high-quality explanations for predicted TCR-pMHC binding. Experiments of our proposed method on large-scale datasets demonstrate competitive predictive accuracy and generalization, and evaluation on the TCR-XAI benchmark demonstrates improved explainability compared with existing approaches.

The modeling of genomic sequences presents unique challenges due to their length and structural complexity. Traditional sequence models struggle to capture long-range dependencies and biological features inherent in DNA. In this work, we propose TrinityDNA, a novel DNA foundational model designed to address these challenges. The model integrates biologically informed components, including Groove Fusion for capturing DNA's structural features and Gated Reverse Complement (GRC) to handle the inherent symmetry of DNA sequences. Additionally, we introduce a multi-scale attention mechanism that allows the model to attend to varying levels of sequence dependencies, and an evolutionary training strategy that progressively adapts the model to both prokaryotic and eukaryotic genomes. TrinityDNA provides a more accurate and efficient approach to genomic sequence modeling, offering significant improvements in gene function prediction, regulatory mechanism discovery, and other genomics applications. Our model bridges the gap between machine learning techniques and biological insights, paving the way for more effective analysis of genomic data. Additionally, we introduced a new DNA long-sequence CDS annotation benchmark to make evaluations more comprehensive and oriented toward practical applications.

CD8+ "killer" T cells and CD4+ "helper" T cells play a central role in the adaptive immune system by recognizing antigens presented by Major Histocompatibility Complex (pMHC) molecules via T Cell Receptors (TCRs). Modeling binding between T cells and the pMHC complex is fundamental to understanding basic mechanisms of human immune response as well as in developing therapies. While transformer-based models such as TULIP have achieved impressive performance in this domain, their black-box nature precludes interpretability and thus limits a deeper mechanistic understanding of T cell response. Most existing post-hoc explainable AI (XAI) methods are confined to encoder-only, co-attention, or model-specific architectures and cannot handle encoder-decoder transformers used in TCR-pMHC modeling. To address this gap, we propose Quantifying Cross-Attention Interaction (QCAI), a new post-hoc method designed to interpret the cross-attention mechanisms in transformer decoders. Quantitative evaluation is a challenge for XAI methods; we have compiled TCR-XAI, a benchmark consisting of 274 experimentally determined TCR-pMHC structures to serve as ground truth for binding. Using these structures we compute physical distances between relevant amino acid residues in the TCR-pMHC interaction region and evaluate how well our method and others estimate the importance of residues in this region across the dataset. We show that QCAI achieves state-of-the-art performance on both interpretability and prediction accuracy under the TCR-XAI benchmark.

Recent remarkable advancements in geometric deep generative models, coupled with accumulated structural data, enable structure-based drug design (SBDD) using only target protein information. However, existing models often struggle to balance multiple objectives, excelling only in specific tasks. BInD, a diffusion model with knowledge-based guidance, is introduced to address this limitation by co-generating molecules and their interactions with a target protein. This approach ensures balanced consideration of key objectives, including target-specific interactions, molecular properties, and local geometry. Comprehensive evaluations demonstrate that BInD achieves robust performance across all objectives, matching or surpassing state-of-the-art methods. Additionally, an NCI-driven molecule design and optimization method is proposed, enabling the enhancement of target binding and specificity by elaborating the adequate interaction patterns.

In recent years it has become evident the need of understanding how failure of coordination imposes constraints on the size of stable groups that highly social mammals can live in. We examine here the forces that keep animals together as a herd and others that drive them apart. Different phenotypes (e.g. genders) have different rates of gut fill, causing them to spend different amounts of time performing activities. By modeling a group as a set of semi-coupled oscillators on a disc, we show that the members of the group may become less and less coupled until the group dissolves and breaks apart. We show that when social bonding creates a stickiness, or gravitational pull, between pairs of individuals, fragmentation is reduced.

A dramatic increase in the number of outbreaks of Dengue has recently been reported, and climate change is likely to extend the geographical spread of the disease. In this context, this paper shows how a neural network approach can incorporate Dengue and COVID-19 data as well as external factors (such as social behaviour or climate variables), to develop predictive models that could improve our knowledge and provide useful tools for health policy makers. Through the use of neural networks with different social and natural parameters, in this paper we define a Correlation Model through which we show that the number of cases of COVID-19 and Dengue have very similar trends. We then illustrate the relevance of our model by extending it to a Long short-term memory model (LSTM) that incorporates both diseases, and using this to estimate Dengue infections via COVID-19 data in countries that lack sufficient Dengue data.