PLOS Computational Biology: New Articles

A prototype-augmented graph representation learning framework for identifying brain disorder-associated genes and facilitating drug repurposing

2026-05-29T14:00:00Z

by Jiafang Li, Yifei Li, Siying Lin, Jiahua Rao, Huiying Zhao

Many genetic loci were identified as associated with neuropsychiatric disorders and neurodegenerative disorders by Genome-wide association studies (GWAS). How these loci impact these diseases is unclear. Advances in deep-learning approaches and multi-omics data have the potential to link GWAS findings with disease mechanisms. Here, we proposed the Multi-omics Graph Transformer Network (MOGT), a semi-supervised graph neural network that leverages graph representation learning to model biological networks derived from multi-omics data to predict disease-associated genes. MOGT outperforms the current approaches in disease gene prediction for two psychiatric disorders and three neurodegenerative/neurological diseases. High-risk genes (HRGs) for Parkinson’s disease (PD) predicted by MOGT were used to drug discovery by integrating with the CMAP database. Finally, 10 drugs were identified as potential candidates. Among them, the effect of drug UK-356618 was experimentally verified in a primary neuron model, showing that UK-356618 reversed the abnormal expression of PD-associated genes and improved the cell-level phenotypes of PD. Together, these results indicate that MOGT can be used to identify HRGs for brain disorders, and these predicted HRGs provide high-level insights into the mechanisms and treatments of brain disorders.

Structural and dynamic basis of NOD2 tandem CARD association and NOD1/2–RIP2 signaling complexes

2026-05-29T14:00:00Z

by Jitendra Maharana, Aritra Bej, Debasish Biswal, Debashis Panda, Arjun Sharma

NOD1 and NOD2, founding members of the NOD-like receptor (NLR) family, play a crucial role in host defense against bacterial infections. Recognition of peptidoglycan-derived ligands triggers ATP-dependent oligomerization of the NACHT domain, exposing the CARD domains that recruit the adaptor protein RIP2 via CARD-CARD interactions to activate the NF-κB signaling cascade. Although NOD1/2-RIP2 interactions and RIP2^CARD filament assembly are established, the precise interfaces that stabilize hetero-CARD filaments remain poorly defined. Here, we integrate in silico structural modeling with molecular dynamics (MD) simulations to elucidate structurally compatible arrangements of NOD1–RIP2 and NOD2–RIP2 hetero-CARD filaments. Our results reveal that NOD1^CARD subunits form a structurally compatible homomeric scaffold via canonical (type-I–III) interfaces, accommodating multiple tiers of RIP2^CARD rings at both filament termini. Meanwhile, the NOD2 tandem CARDs adopt multiple discrete conformations, reflecting a more intricate structural mechanism. In stable filament conformations, tandem CARDs converge at the type-II interface, with RIP2^CARD rings stacking onto CARDa (top-down) and CARDb (bottom-up) interfaces, highlighting the structural role of NOD2^CARDb in RIP2-mediated CARD-CARD interaction. In silico mutagenesis, involving charge-reversal and alanine scanning at key interfacial residues, disrupts NOD1–RIP2 and NOD2–RIP2 interactions at both top-down and bottom-up interfaces, leading to rapid interface destabilization within 0.1–0.4 μs of simulation. Together, these results reveal conserved and receptor-specific structural mechanisms governing NOD1/2–RIP2 CARD–CARD interactions and provide deeper structural and dynamic insights into the complex structural mechanisms for NLR-mediated inflammatory signaling.

Fully synthetic replication of complex real biological cell clusters using a novel cluster-based ‘Rosetta-Routine’ computational modelling process

2026-05-29T14:00:00Z

by Bradley Mason, Laura Justham, Liam Whitby, Alison Whitby, Stuart Scott, Samuel Nti, Jon Petzing

Flow cytometry (FC) is essential for the precise quantification and characterisation of individual cell populations in a larger heterogenous cell suspension. FC analysis provides a foundation for advanced clinical diagnostics and is a key component in many life-saving therapeutic strategies across a broad range of medical conditions. However, clinical, industrial and research laboratories alike face significant challenges in validating the metrological and biological accuracy of FC data analysis. Due to the inherent relative nature of FC data and the lack of definitive ‘ground truth’ associated with processed biological samples. This study specifically focuses on generating realistic fully synthetic flow cytometry cell clusters and demonstrating their suitability as substitutes for traditional FC data. The inherent model-based heritage of synthetic data enables the robust ability to generate distributionally-equivalent replicate datasets with explicit knowledge of cluster membership for each individual datapoint. Thereby, reducing the uncertainty issues associated with real cluster data and its analysis. This research uses meticulously optimised synthetic cluster-generating benchmarking software to simulate real monocyte clusters. A central component of the protocol is the ‘Rosetta-Routine’, a novel codebase which deciphers the statistical properties of real data and translates them into the computational coefficients required to generate accurate cluster-based synthetic replicates. This innovative approach ensures that the synthetic datasets faithfully represent the statistical characteristics of real-world data while retaining the benefits of computational traceability. This approach addresses a critical gap in current practices by enabling the ability to provide a controlled and reproducible validation framework for assessing clustering methods applied to analyse FC data. These features allow the ability to score and subsequently enhance the analysis confidence in many FC applications such as in diagnostics or in ‘mock-up’ training scenarios. Future synthetic-data-driven enhancements in FC analysis confidence will translate into more accurate clinical decision-making and subsequent overall improvements in patient care.

TIPP-SD: A new method for species detection in microbiomes

2026-05-28T14:00:00Z

by Chengze Shen, Eleanor Wedell, Mihai Pop, Tandy Warnow

In this study, we present TIPP-SD (i.e., TIPP for Species Detection), a new technique for species detection in a microbiome sample. TIPP-SD uses a substantially modified version of TIPP3, which is a recently developed abundance profiling tool based on maximum likelihood phylogenetic placement into marker gene taxonomies. TIPP-SD depends on a parameter (i.e., “threshold”) for the required support for species detection, thus allowing us to compute a precision-recall curve as we vary this parameter. In comparing the precision-recall curves for TIPP-SD, TIPP3, Kraken2, Bracken, Metabuli, and Metapresence, we find that TIPP-SD improves on the other methods with respect to accuracy under conditions where there is a highly variable distribution of species abundance or where there is sequencing error. Under other conditions, TIPP-SD is close to the best of these methods. Finally, although TIPP-SD is slower than the other methods, it is still fast enough to be used on large datasets. TIPP-SD is available in github as part of the TIPP3 software package.

Correction: Bayesian network models to assess antimicrobial resistance patterns of Streptococcus suis isolated from swine production systems in the United States between 2014–2021

2026-05-28T14:00:00Z

by Ruwini Rupasinghe, Brittany L. Morgan Bustamante, Rebecca C. Robbins, Maria J. Clavijo, Beatriz Martínez-López

Accurate computation of ionic concentrations in the synaptic cleft requires the full Poisson–Nernst–Planck (PNP) equations

2026-05-28T14:00:00Z

by Karoline Horgmo Jæger, Aslak Tveito

The synaptic cleft between neighboring neurons is the site of neurotransmitter-mediated communication that underlies normal brain function, including learning and memory. When an action potential reaches the presynaptic terminal, released neurotransmitters cross the cleft under the combined influence of diffusion and electrical forces to activate postsynaptic receptors. Despite this, synaptic-cleft transport is commonly modeled using a pure diffusion model, neglecting electrical drift. Here, we quantify the relative contributions of diffusion and electrical terms in the Poisson–Nernst–Planck (PNP) framework and assess whether the pure diffusion approximation is adequate. We solve the full PNP system in a three-dimensional computational model of the synaptic cleft at nanometer-scale resolution, tracking five ionic species (Na⁺, K⁺, Ca²⁺, Cl⁻, Glu⁻) with full spatial and temporal detail. Solutions are compared directly with those of the pure diffusion (D) model. The D and PNP models produce markedly different ionic concentration fields. Analysis of ionic fluxes confirms that diffusive and electrical contributions are of comparable magnitude across all species. These discrepancies are robust across parameter variations, including the number of AMPA receptors, the amount of released glutamate, the cleft height, and the cleft diffusion coefficient, and are amplified as the number of AMPA receptors or released glutamate ions increases, when the cleft becomes narrower or when diffusion becomes more restricted. However, because of competing effects, the resulting difference in the associated AMPA current is modest. The quantitative and qualitative differences between the pure D model and the full PNP model demonstrate that neglecting electrical forces in the synaptic cleft has consequences. These discrepancies are large enough to alter the predicted dynamics and biological interpretation of synaptic transmission, establishing that accurate computation of ionic concentrations in the synaptic cleft requires the full PNP equations.

UNified FramewOrk for reguLatory Dynamics (UNFOLD): Dissecting robustness, plasticity, evolvability and canalisation of biological function

2026-05-28T14:00:00Z

by Debomita Chakraborty, Raghunathan Rengaswamy, Karthik Raman

A unique balance of seemingly contradictory properties like robustness and plasticity, or evolvability and functional canalisation, characterises biological systems. To understand the basis of these properties, we investigate gene regulation, which is at the core of biological function. We simulate dynamical models of over 190 million genetic circuits covering all possible three-gene circuit structures. Our computational pipeline classifies these circuits into functional clusters by matching their temporal response shapes. Thus, we generate a dataset linking circuit structure, parameters and a corresponding functional label. Our key finding is a finite list of 20 functions that three-node genetic circuits can perform under step input and within the explored parameter space. Moreover, the structure-parameter space for these circuits tends to be primed for responses that stabilise over time following a perturbation. Every structure exhibits potential for multifunctionality with a range of 2–17 functions contingent upon parameters. We quantify network degeneracy, showing that many structural changes can be made to circuits without altering function. We define three quantities: structural, parametric, and functional diversities. Using these diversities, we construct a UNified FramewOrk for reguLatory Dynamics (UNFOLD) to analyse four key biological properties—robustness, plasticity, evolvability, and functional canalisation. Using UNFOLD, and within the explored parameter space, we identify that only 6.5% of network structures are non-plastic, while parameter sets enabling parametric robustness exist for every three-node network. We identify functionally canalised circuits from structure pairs that can be interchanged for a large number of parameter sets without a change in function. Overall, our framework offers insights into the fundamental organisation of biological networks by thorough analysis of three-node networks.

BINSEQ: A family of high-performance binary formats for nucleotide sequences

2026-05-28T14:00:00Z

by Noam Teyssier, Alexander Dobin

Modern genomics produces billions of sequencing records per run, which are typically stored as gzip-compressed FASTQ files. While this format is widely used, it is not optimal for high-throughput processing due to its reliance on single-threaded decompression and sequential parsing of irregularly sized records. This limitation is particularly problematic for applications that would benefit from parallel processing, such as read mapping, variant calling, and de novo assembly. Here, we present BINSEQ, a family of simple binary formats that enable high-throughput parallel processing of sequencing data. The BINSEQ family consists of two complementary implementations: BQ, optimized for fixed-length reads using a two-bit or four-bit encoding scheme with true random record access capability, and VBQ, designed for variable-length sequences with optional quality scores and block-based compression. We demonstrate that BINSEQ files are up to 90x faster than compressed FASTQ for parallel processing and can reduce analysis time from hours to minutes for large-scale genome and transcriptome analyses, particularly for resource-intensive applications like alignment, mapping, and de novo assembly. To facilitate adoption we provide high-performance libraries for reading and writing BINSEQ formats, native parallelization strategies with convenient APIs, and a command-line tool for conversion to and from traditional formats.

Hierarchical recurrent temporal prediction as a model of the mammalian dorsal visual pathway

2026-05-28T14:00:00Z

by Sebastian Klavinskis-Whiting, Andrew J. King, Nicol S. Harper

A major goal of neuroscience is to identify general principles that can explain the diverse structures and functions of the brain. The principle of temporal prediction provides one approach, arguing that the sensory brain is optimized to represent stimulus features that efficiently predict the immediate future input. Previous work has demonstrated that feedforward hierarchical temporal prediction models can capture the tuning properties of neurons along the visual pathway, and that recurrent temporal prediction models can explain local functional connectivity within primary visual cortex. However, the visual system is also characterized by extensive inter-areal feedback recurrency, which existing models lack. We aimed to better account for the dynamic features of neurons in the visual cortex by incorporating both local recurrency and inter-areal feedback connectivity into a hierarchical temporal prediction model. The resulting model captured tuning properties along the dorsal visual pathway, including pattern motion selectivity and surround suppression, and the contribution of inter-areal connectivity to these properties. Moreover, compared with several alternative normative models, the hierarchical recurrent temporal prediction model provided the closest fit to these tuning properties and was best able to explain neuronal responses to natural stimuli. Accordingly, temporal prediction accounts well for information processing along the visual pathway.

The interplay between ecological networks drives host-plasmid community dynamics

2026-05-26T14:00:00Z

by Ying-Jie Wang, Kaitlin A. Schaal, Johannes Nauta, Armun Liaghat, Manlio De Domenico, James P. J. Hall, Shai Pilosof

Plasmids drive evolution by transferring traits across microbial hosts. Transmission depends on both host–plasmid (infection) and plasmid–plasmid (compatibility) interactions, yet how the structure of these networks shapes transmission remains poorly understood. We hypothesized that these two ecological networks interact in non-additive ways to influence community outcomes. To test this, we developed a stochastic agent-based model that embeds both network structures and simulates coupled host–plasmid dynamics. We systematically varied the structure of each network, both individually and in combination, to isolate the effect of structure on host-plasmid dynamics. A modular (interactions organized into clusters) and hub (interactions concentrated on the highly connected) plasmid-plasmid compatibility network promoted transient host coexistence, while a modular host-plasmid infection network promoted plasmid diversity and stable host coexistence. Importantly, structured networks interacted non-additively, and their impact was most apparent when plasmid carriage imposed a moderate fitness cost on hosts. For example, combining a modular infection network with a hub compatibility network reversed the expected plasmid prevalence patterns, demonstrating that the structure of one network can counteract the effects of the other. We further re-parameterized our model to recapitulate empirical host-plasmid community dynamics, showing that infection network structure can strongly shape plasmid prevalence even in the presence of substantial biological heterogeneity. Our results highlight the necessity of jointly considering host–plasmid infection and plasmid–plasmid compatibility networks to understand host–plasmid community dynamics and their eco-evolutionary potential. More broadly, this work provides an initial mechanistic framework for generating testable hypotheses and underscores that systems involving multiple hosts and infectious agents require explicit consideration of how different ecological networks interact to shape community dynamics.

Utilizing virus genomic surveillance to predict vaccine effectiveness

2026-05-26T14:00:00Z

by Jiye Kwon, Ke Li, Joshua L. Warren, Sameer Pandya, Anne M. Hahn, Yale SARS-CoV-2 Genomic Surveillance Initiative , Virginia E. Pitzer, Daniel M. Weinberger, Nathan D. Grubaugh

Background

Since the development of the first vaccines targeting the original SARS-CoV-2 virus sequence in 2020, mRNA-based vaccines have been updated three times: targeting Omicron BA.4/BA.5 in 2022, the XBB lineage in 2023, and the KP.2 variant in 2024. While genomic surveillance has advanced our understanding of pathogen diversity, gaps remain in incorporating genomic information to evaluate vaccine effectiveness (VE) against emerging variants. This study aims to characterize the relationship between VE and sequence-based genetic distance, to establish a framework for predicting near real-time changes in the level of vaccine protection from virus surveillance data.

Methods

We analyzed 10,156 whole genome sequences of SARS-CoV-2 cases from Connecticut, USA, between April 2021 to July 2024. We first assessed how genetic distance, specifically the number of amino acid substitutions in the spike gene between COVID-19 case sequences and the mRNA vaccine formulation sequence(s), correlates with vaccine protection levels. Incorporating data from over 1 million test-negative controls, we developed a Bayesian time-varying model with autoregressive terms to assess VE at a weekly level. The analysis was adjusted for ZIP-code-level income, age, sex, and prior vaccine doses received. We then employed a random effects meta-regression to explore the relationship between VE and amino acid distance over time. Finally, we used the meta-regression model to estimate potential vaccine protection against emerging variants.

Findings

We found that spike gene amino acid distance showed a negative correlation with VE over time. Stepwise increases in amino acid distance aligned with sharp VE declines during variant emergence, while accumulation of within-variant changes was also associated with gradual VE decline. Each 10 amino acid increase in distance in the spike gene corresponds to a predicted 15.4% (95% credible intervals (CrI): –2.0%, 34.6%) reduction in VE. For the 2023/24 updated vaccine, spike distance rose from 12.25 to 30.23, predicting a 43.4% (95% CrI: –5.7%, 90.1%) drop in VE using sequence information alone.

Conclusion

Our framework quantifies how the emergence of new variants is expected to affect VE for SARS-CoV-2. By quantifying the relationship between amino acid substitutions and time-varying VE, we leverage intrinsic pathogen features, such as spike amino acid distance, to inform future vaccine updates using genomic sequences. As genomic surveillance data becomes more widely available across pathogens, this framework can serve as a near-real time surveillance tool to infer population-level protection and offers valuable insights for vaccine update decisions.

The role of pragmatic mechanisms in referential communication and categorization: An emergent communication model

2026-05-26T14:00:00Z

by Kristina Kobrock, Xenia Ohmer, Elia Bruni, Nicole Gotzner

We model pragmatic mechanisms of referential communication and categorization in a multi-agent framework of emergent communication. Pragmatic theories and experimental work predict that speakers consider the context in their choice of referring expressions. In addition to this context-based reasoning, utility-based pragmatic reasoning about the listener’s likely interpretation of an utterance influences the speaker’s production choices. We aim to investigate these two factors and their role in referring expression generation and categorization in a computational model of language emergence and language use. We model communication in interaction and consider an efficiency tradeoff between speaker and listener utilities. Our results show that an emerging language becomes more effective, ambiguous, and efficient when speakers and listeners communicate in a shared context. This is achieved by an efficient tradeoff between production and comprehension where languages can afford to be simpler in production when they are sufficiently informative in context, placing more burden on the listener’s side. We further demonstrate that incorporating utility-based pragmatics, as modeled with the Rational Speech Acts framework, improves the linguistic efficiency of language use only in languages that emerged with a shared context between interlocutors, but not in languages that emerged without such contextual information during training. We conclude that context-based pragmatics plays a role in referential communication and categorization by shaping an emerging language. Efficient reference in a communicative situation can benefit especially from utility-based pragmatics if the language that is being used has emerged in context. This might suggest that utility-based pragmatics hinges on mechanisms that naturally emerge when context is available during the evolution of a language. In summary, we show that human-like language and category systems emerge as an optimized tradeoff between speaker and listener needs in interaction, i.e., under efficiency considerations of simplicity and informativeness.

Multilabel prediction of virus target proteins via multimodal graph representation learning

2026-05-26T14:00:00Z

by Kuang Ma, Kaiyu Liu, Yuhui Xin, Rong Liu

Identification of virus target proteins (VTPs) is crucial for understanding viral pathogenesis. Existing computational studies have addressed this issue by predicting host-virus protein interactions, typically framed as a single-label problem. However, targets can be identified using only intrinsic information of host proteins. Moreover, a host protein may participate in the infection processes of multiple viruses, a scenario that can be treated as a multilabel prediction problem. Herein, we present MultiVTP, a multilabel framework for VTP prediction that employs graph learning with multimodal information. This algorithm samples subgraphs centered on query proteins to capture topological properties, while multimodal features are extracted to represent proteins from complementary perspectives. A graph transformer integrates and upgrades these attributes, followed by a progressive layered extraction module that captures both shared and virus-specific binding patterns to predict VTPs. Ablation experiments reveal that graph-based attributes and modules are the key contributors to performance, with additional components leading to further improvements in accuracy. Comprehensive evaluations demonstrate that MultiVTP not only surpasses various baseline models but also remains robust under limited training data. Applying our approach to the human proteome enables the systematic identification of novel VTPs for both individual and multiple viruses.

Cooperative molecular interaction networks govern PARP1 inhibitor selectivity and binding affinity

2026-05-26T14:00:00Z

by Alejandro Feito, Natàlia DeMoya‐Valenzuela, Cristian Privat, Andrés R. Tejedor, Lucía Paniagua-Herranz, Adiran Garaizar, Alberto Ocana, Jorge R. Espinosa

Selective inhibition of PARP1 represents a promising strategy to improve the therapeutic index of PARP inhibitors, a class of anticancer agents that exploit defects in DNA repair pathways. While PARP inhibitors have shown remarkable clinical benefit, particularly in BRCA-mutated tumors, the lack of discrimination between PARP1 and its close homolog PARP2, often leads to hematological toxicity and limits treatment efficacy. Thus, achieving molecular selectivity for PARP1 remains a central challenge in the rational design of safer and more potent inhibitors. To explore the molecular determinants of ligand selectivity, we focus on four clinically relevant PARP inhibitors—two PARP1-selective (saruparib and NMS-P118) and two non-selective (veliparib and olaparib) inhibitors—and perform atomistic potential-of-mean-force calculations of the PARP1 catalytic binding domain in the presence of these molecules. Our simulations near-quantitatively capture the experimental relative binding preferences, demonstrating that our approach reliably reflects selectivity patterns. Based on these findings, we analyze protein–ligand contact frequencies to identify the stabilizing interaction network and contact connectivity inducing protein selectivity. The most frequent protein–inhibitor contacts are primarily mediated by tyrosine triads and electrostatic interactions, showing a cooperative complex network of intermolecular contacts which strongly relies on protein multivalency. To dissect the decisive role of individual residues across the binding site, we also perform targeted mutagenesis of the PARP1 catalytic pocket in complex with saruparib, replacing several active-site amino acids by glycines. Progressively increasing the number of mutations markedly reduces binding stability, with distinct residue combinations exerting two primary effects: destabilization of the final bound state and the emergence of energetic barriers along the ligand association pathway. Together, our results provide a coherent mechanistic framework for understanding PARP1 selectivity and informs the rational design of next-generation inhibitors with improved efficacy and safety.

MIAAIM: Multi-omics image integration with dimensional reduction for tissue state mapping

2026-05-26T14:00:00Z

by Joshua M. Hess, Richard K. Dzeng, Iulian Ilieş, Denis Schapiro, John J. Iskra, Divya Mirgh, John Nam, Erin H. Seeley, David E. Verrill, Walid M. Abdelmoula, Michael S. Regan, Georgios Theocharidis, Chin Lee Wu, Aristidis Veves, Nathalie Y. R. Agar, Ann E. Sluder, Mark C. Poznansky, Ruxandra F. Sîrbulescu, Patrick M. Reeves

High-parameter tissue imaging enables detailed molecular analysis of single cells within their spatial environment. A current challenge to more complete tissue and single-cell spatial profiling is in situ data alignment across imaging platforms that quantify multiple types of biomolecules at differing resolutions. Here, we describe MIAAIM (Multi-omics Image Alignment and Analysis by Information Manifolds), a modular framework to align and process data from separate imaging technologies with distinct imaging resolutions and data complexity. MIAAIM is designed to be applied to align and analyze images of clinical biopsies from histological staining, imaging mass cytometry, and mass spectrometry imaging. A key advantage of the MIAAIM approach is its capacity to identify unbiased molecular phenotypes that correlate with cell identities and states determined using high-resolution targeted immunodetection. In a large diabetic foot ulcer (DFU) biopsy, this strategy allowed the identification of unique molecular characteristics of infiltrating immune cells as a function of local tissue health. In multi-core tissue microarrays (TMAs) of prostate cancer, MIAAIM allowed the classification of adjacent tumor grades with high accuracy, with over 90% of classification signal sourced from spatial features, generated from segmented cells across multiple imaging modalities while revealing novel cell/ immune signatures of the disease state. MIAAIM provides a disease and cell type agnostic general framework to construct multimodal tissue imaging datasets, yielding novel insights into the association of molecular analytes with cell subsets and their activation states for the analysis of complex tissue states.

Agent-based modeling demonstrates how target-independent processes supplement killing by antibody-drug conjugates in cancer therapy

2026-05-26T14:00:00Z

by Melissa C. Calopiz, Jennifer J. Linderman, Greg M. Thurber

Antibody-drug conjugates (ADCs) have had remarkable clinical success in recent years with multiple new approvals. However, for some ADCs, the response rates don’t closely correlate with clinical target expression. One particular ADC targeting HER2, trastuzumab deruxtecan or T-DXd, is notable due to its success at expression levels ranging from high to low and ultralow. This raises the question of the relative contributions of target-independent mechanisms on ADC efficacy in the clinic, and several such mechanisms have been proposed. However, in vitro and preclinical data have different doses and exposures, making it challenging to quantitatively extrapolate preclinical data to the clinic. In this work, we use our computational hybrid agent-based model, SimADC, to simulate target-dependent and -independent mechanisms, scaling from mice to humans. We first demonstrate that CD8 + T cells can significantly contribute to tumor regression, especially when the ADC further activates the immune cells. Next, we test target-independent payload-driven mechanisms including: 1) Fc-mediated internalization of ADC by intratumoral macrophages and payload release to neighboring cancer cells, 2) free payload circulating in the blood and re-entering the tumor, and 3) extracellular linker cleavage and payload release due to an abundance of proteases in the tumor. We find that free payload in the blood and extracellular linker cleavage had low and moderate impacts, respectively, while macrophage uptake and payload release resulted in high levels of efficacy. This is due to the macrophages’ ability to sustain free payload in the tumor. Moderate and high HER2 expression were more efficacious than target-independent mechanisms. Overall, our simulations demonstrate that moderate to high HER2 expression, immune activation, or macrophage uptake and payload release are sufficient for T-DXd tumor regression. Additionally, SimADC provides a robust framework for modeling both target-dependent and target-independent mechanisms for any ADC, providing the opportunity to engineer more effective therapeutic agents.

DREAMER-S: Deep leaRning-Enabled Attention-based Multiple-instance approaches with Explainable Representations for Spatial biology

2026-05-26T14:00:00Z

by M. Rifqi Rafsanjani, Alison Dooney, Rahul Suresh, Alice C. O’Farrell, Monika A. Jarzabek, Liam Shiels, Annette T. Byrne, Jochen H. M. Prehn, Aidan D. Meade

Identifying image features that associate strongly with diagnostic or prognostic classes in large-scale, multi-channel spatial imaging is challenging without pixel-level annotations. We present DREAMER-S, an attention-based multiple-instance learning (MIL) framework that, using only image- or slide-level labels, learns spatial features within 3D imaging hypercubes that are most informative for downstream classification. We demonstrate DREAMER-S on Quantum Cascade Laser infrared (QCL-IR) tissue imaging, where attention weights are rendered spatially to highlight class-relevant spectral instances without manual annotation. Because the MIL attention layer assigns interpretable importances to spatial instances, the method is broadly transferable to spatial-biology applications that require instance-level filtering to focus towards salient regions of interest in high-content datasets. We further evaluate DREAMER-S on a chemotherapy-response task in a colorectal cancer patient-derived xenograft (PDX) model. After tuning, DREAMER-S separated spectral instances from a chemo-sensitive PDX (CRC0344) and a less responsive PDX (CRC0076) with an F1 score of ~0.95. To validate explainability, we linked model saliency to cellular physiology, observing that, (i) unsupervised UMAP embeddings of high-attention spectra stratified samples by treatment (chemotherapy, apoptosis sensitizer, combination, vehicle), and (ii) selected spectral markers correlated with pro-apoptotic proteins measured independently in the same PDX system. Together, these results support a mechanistic link between spectral signals and apoptosis pathways and position DREAMER-S as an efficient, interpretable approach for analysing high-content spatial-biology imaging datasets.

Evaluating place cell detection methods in Rats and Humans: Implications for cross-species spatial coding

2026-05-26T14:00:00Z

by Weijia Zhang, Thomas Donoghue, Salman E. Qasim, Joshua Jacobs

Place cells, first identified in the rat hippocampus as neurons that fire selectively at specific locations, are central to investigations of the neural underpinnings of spatial navigation. Recent spatial studies in human patients with drug-resistant epilepsy have made identifying and characterizing place cells across species increasingly important for understanding the extent to which decades of rodent research generalize to humans and for uncovering fundamental principles of spatial cognition. One challenge, however, is that detection methods differ: rodent studies often rely on spatial information (SI) in conjunction with place field stability measures, whereas human studies employ analysis of variance (ANOVA) based approaches. These methodological differences may affect the identified place cell populations, which complicates how their properties are interpreted and cross-species comparisons. To address this, we systematically applied multiple detection pipelines to human and rat datasets, supported by simulations that vary place-field properties. Our analyses and simulations demonstrate that spatial information and ANOVA-based approaches are responsive to distinct place field properties: spatial information primarily reflects the contrast between peak and average firing rates, while ANOVA emphasizes consistency across trials. Across species, rodent place cells revealed a broad spectrum of spatial tuning, including strongly tuned neurons with high spatial information and high ANOVA values. In contrast, human place cells lacked this strongly tuned population and exhibited a narrower distribution of tuning scores, concentrated at the lower end of both spatial tuning metrics. Despite these differences, both species had an overlapping population of neurons with weaker yet consistent spatial tuning, which may support important functional roles such as generalization and mixed selectivity. Addressing these analytical differences allows for more direct comparisons between species, though differences in spatial tuning may still relate to variations in experimental paradigms that warrant further investigation. Together, our study provides a roadmap showing how spatial tuning metrics shape place cell detection and interpretation.

When one race is not enough: A relay model explains multisensory response times

2026-05-26T14:00:00Z

by Kalvin Roberts, Thomas U. Otto

Humans typically respond faster to multisensory signals than to their unisensory components, a phenomenon known as the redundant signal effect (RSE). One of the earliest and most influential accounts, the race model, attributes the RSE to statistical facilitation, which arises from parallel, independent processing across sensory modalities. While this model captures some key features of the RSE, it frequently underestimates the observed speed-up leading to violations of the race model inequality (RMI), a benchmark used to test the model’s validity. To reconcile this discrepancy, we introduce the relay model, a minimal extension of the race architecture that incorporates cross-modal initiation. In this model, responses result from two sequential race processes, allowing a signal in one modality to initiate the onset of perceptual decision processing in another. This structure retains statistical facilitation as a core principle while introducing a single free model parameter that partitions unisensory processing into gating and decision stages. Through simulations and fits to foundational empirical datasets, we show that the relay model captures both the magnitude and distributional shape of the RSE, including RMI violations. It also accounts for changes in the RSE under asynchronous stimulus onsets and manipulations of signal intensity, which are critical tests in multisensory research. By extending the classical race model with minimal added complexity, the relay model offers a mechanistically explicit and biologically plausible framework for explaining the dynamics of multisensory decision-making.

Integrated computational and experimental analysis explores FOLH1 expression patterns across cancers and nominates melatonin as a potential modulator in prostate cancer models

2026-05-22T14:00:00Z

by Rui Zhang, Junyu Zhou, Sihan Dong, Guoquan Liu, Xunbin Wei

Background

Growing evidence indicates that Folate Hydrolase 1 (FOLH1, also known as prostate-specific membrane antigen, PSMA) is aberrantly expressed across multiple malignancies, particularly showing significant upregulation in prostate cancer. However, systematic investigations into its pan-cancer expression patterns, immunomodulatory roles, and immune cell infiltration remain limited. The potential role of FOLH1 in prostate cancer is also not fully elucidated.

Methods

We analyzed FOLH1 mRNA expression, prognostic relevance, and immune infiltration across multiple malignancies, with a particular focus on prostate cancer. A machine learning (ML) workflow incorporating a deep learning model was developed to screen the therapeutic potential of drugs targeting FOLH1. The therapeutic potential of these candidates was validated through in vitro cellular assays and nude mouse xenograft models.

Results

FOLH1 expression was significantly altered in 27 cancer types and showed cancer-specific immune correlations. Our AI platform identified melatonin as a computationally predicted FOLH1-interacting candidate. In vitro and in vivo experiments demonstrated that melatonin suppresses FOLH1 expression in a concentration-dependent manner, inhibits invasive and migratory capacities, and restricts tumor growth under physiological circadian melatonin levels.

Conclusion

This study highlights FOLH1’s pan-cancer expression patterns and nominates melatonin as an exploratory therapeutic candidate for prostate cancer requiring further mechanistic validation. Our integrated computational-experimental framework highlights the promise of AI-driven drug discovery in oncology, while emphasizing the need for further mechanistic validation.

Scaffold-Lab: Critical evaluation and ranking of protein backbone generation methods in a unified framework

2026-05-22T14:00:00Z

by Zhuoqi Zheng, Bo Zhang, Bozitao Zhong, Jinyu Yu, Kexin Liu, Zhengxin Li, Junjie Zhu, Ting Wei, Hai-Feng Chen

Recent advances in de novo protein design have significantly progressed, particularly in protein backbone generation, which remains a challenging yet valuable task. However, there has been a lack in standardized evaluation framework to accurately assess diverse methods, nor is there a comprehensive analysis of their practical applications. In this study, we proposed Scaffold-Lab, a unified framework for systematic evaluation of protein backbone generation methods, encompassing a wide range of metrics including designability, novelty, diversity, efficiency and structural properties. We thoroughly evaluated seven representative methods, providing a detailed analysis of their performance and utilities. Our findings highlight that generating long proteins for unconditional generation and accurately reconstructing motifs are key bottlenecks for most methods. These results underscore both the strengths and limitations of existing protein backbone generation methods, offering new insights for further development and improvement in this field.

Spatial richness of neural magnetic fields

2026-05-22T14:00:00Z

by Ziad Ali, Ada S. Y. Poon

Brain implants that measure neural magnetic fields, rather than electrical potentials, are expected to confer significant clinical advantages related to implant longevity and signal fidelity due to the elimination of the electrode-tissue interface. However, the informational differences between neural electrical potentials and magnetic fields remain poorly understood. Using a mathematical formalism based on neuronal current sources, we directly establish the complementary informational content of extracellular magnetic fields and electrical potentials. This formalism also reveals that extracellular magnetic fields generated by spiking neurons inherently exhibit one order lower spatial polarity than electric fields, resulting in more favorable distance-scaling characteristics. We then use computational modeling to illustrate how dense networks of neurons are easier to distinguish and spike sort on the basis of their magnetic, rather than electrical, spike templates. Lastly, we show how the solenoidal nature of neural magnetic fields facilitates approximate morphological reconstruction, even with sparse sensor arrays. Our findings highlight the unique experimental advantages of neural magnetic field sensing, motivating the development of compact, low-noise devices capable of meeting the stringent sensitivity requirements for cortical recordings.

A simple model captures key characteristics of biological non-deterministic genotype-phenotype maps

2026-05-22T14:00:00Z

by Nora S. Martin

By connecting genotypic mutations to the higher-level phenotypes relevant for selection, genotype-phenotype (GP) maps play a key role in evolution. GP maps are typically investigated using computational models of biophysical phenotypes (for example, RNA secondary structures and simplified models of protein tertiary and quaternary structures), but GP map concepts are relevant beyond these specific models. While there has been significant progress in quantifying GP map properties and their evolutionary implications, this is largely limited to the simplest case, where each genotype corresponds to a single, categorical phenotype. Here, I turn to a more realistic, but also more complex, non-deterministic (ND) treatment, meaning that each genotype generates an ensemble of phenotypes. To provide a tool for tackling the additional complexity of ND GP maps, this paper identifies a tuneable synthetic model that produces an ND GP map reproducing central features of biophysical ND GP maps: phenotypic bias, genetic correlations, a tradeoff between genotypic robustness and evolvability and a non-negative trend between phenotypic robustness and evolvability. These features are reproduced for several alternative models combining additive genotype dependencies with non-linearities, suggesting that few ingredients are needed for these shared features to appear. Moreover, the synthetic ND GP map may be useful as a conceptually and computationally simpler model for addressing open questions about ND GP maps: for simulations linking GP map properties to evolutionary implications, for the development of sampling methods for ND GP maps and for extrapolations.

MIRAGE: Robust multi-modal architectures translate fMRI-to-image models from vision to mental imagery

2026-05-22T14:00:00Z

by Reese Kneeland, Cesar Kadir Torrico Villanueva, Tong Chen, Jordyn Ojeda, Shubh Khanna, Jonathan Xu, Paul S. Scotti, Thomas Naselaris

To be useful for downstream applications, vision decoding models that are trained to reconstruct seen images from human brain activity must be able to generalize to internally generated visual representations, i.e., mental images. In an analysis of the recently released NSD-Imagery dataset, we demonstrated that while some modern vision decoders can perform quite well on mental image reconstruction, some fail, and that state-of-the-art (SOTA) performance on seen image reconstruction is no guarantee of SOTA performance on mental image reconstruction. Motivated by these findings, we developed MIRAGE, a method explicitly designed to train on vision datasets and cross-decode mental images from brain activity. MIRAGE employs a linear backbone and multi-modal text and image features as input to a diffusion model. Feature metrics and human raters establish MIRAGE as SOTA for mental image reconstruction on the NSD-Imagery benchmark. With ablation analysis we show that mental image reconstruction works best when decoders use image features with relatively few dimensions and include guidance from text-based and both high- and low-level image-based features. Our work indicates that–given the right architecture–existing large-scale datasets using external stimuli are viable training data for decoding mental images, and warrant optimism about the future success and utility of mental image reconstruction.

Fast reconstruction of degenerate populations of conductance-based neuron models from spike times

2026-05-21T14:00:00Z

by Julien Brandoit, Damien Ernst, Guillaume Drion, Arthur Fyon

Inferring the biophysical parameters of conductance-based models (CBMs) from experimentally accessible recordings remains a central challenge in computational neuroscience. Spike times are the most widely available data, yet they reveal little about which combinations of ion channel conductances generate the observed activity. This inverse problem is further complicated by neuronal degeneracy, where multiple distinct conductance sets yield similar spiking patterns. We introduce a method that addresses this challenge by combining deep learning with Dynamic Input Conductances (DICs), a theoretical framework that reduces complex CBMs to three interpretable feedback components governing excitability and firing patterns. Our approach first maps spike times directly to DIC densities at threshold using a lightweight neural network that learns a low-dimensional representation of neuronal activity. The predicted DIC values are then used to generate degenerate CBM populations via an iterative compensation algorithm, ensuring compatibility with the intermediate target DICs, and thereby reproducing the corresponding firing patterns, even in high-dimensional models. Applied to two neuronal models, this algorithmic pipeline reconstructs spiking, bursting, and irregular regimes with high accuracy and robustness to variability, including spike trains generated under noisy current injection mimicking physiological stochasticity. It produces diverse degenerate populations within milliseconds on standard hardware, enabling scalable and efficient inference from spike recordings alone. Beyond methodological advances, we provide an open-source software package with a graphical interface that allows experimentalists to generate and explore CBM populations directly from spike trains without requiring programming expertise. Together, this work positions DICs as a practical and interpretable link between experimentally observed activity and mechanistic models. By enabling fast and scalable reconstruction of degenerate populations directly from spike times, our approach provides a powerful way to investigate how neurons exploit conductance variability to achieve reliable computation and provides the foundation for experimental applications that span from neuromodulation studies to real-time model-guided interventions.

Energetic constraints shape the diversity of feasible ecological networks

2026-05-20T14:00:00Z

by Chengyi Long, Marco Tulio Angulo, C. Brandon Ogbunugafor, Ricard Solé, Serguei Saavedra

The relationship between energy supply and biodiversity is a longstanding question in ecology. Although a monotonic increase in diversity with energy availability is often assumed, unimodal species–energy relationships have been widely documented across ecosystems, and their origin from first principles remains unclear. Here, we develop a geometric framework that recasts ecological feasibility in explicitly energetic terms. By treating total energy supply as a system-level constraint on an energy-based network model, we define nested feasibility domains in the space of energy capture rates and quantify feasibility probabilities as their volume ratios. We show that the probability of initializing a feasible network increases monotonically and saturates with energy supply, whereas the probability of sustaining steady-state biomass follows a unimodal relationship—revealing a bounded energetic window within which network maturation is most likely. Extending this analysis to all candidate subcommunities via feasibility partitions, we find that different community sizes are most feasible at different energy levels, and that average diversity itself peaks at intermediate supply. Together, these results suggest that energetic constraints determine the diversity of ecological networks not through energy scarcity alone, but through the geometric interplay between external energy supply and internal energy exchange.

DENcode: A model for haplotype-informed transmission probability of dengue virus

2026-05-20T14:00:00Z

by Sachith Maduranga, Braulio Mark Valencia, Chathurani Sigera, Praveen Weeratunga, Deepika Fernando, Senaka Rajapakse, Andrew R. Lloyd, Rowena A. Bull, Haley Stone, Chaturaka Rodrigo

Dengue virus transmission networks are often only partially resolved, due to gaps in sampling, unobserved mosquito-mediated transmission, and using methods (phylogenetics) that describe evolutionary relatedness but not explicit, probabilistic transmission links between individual infections. We developed DENcode, a framework to estimate the relative likelihood of vector-mediated transmission between pairs of dengue cases by combining a temperature- and time-modulated epidemiological kernel, which captures the extrinsic incubation period and human infectiousness, with a phylogenetically informed genetic similarity kernel derived from patristic distances between viral haplotypes or consensus sequences. Validation with a real-life dataset of 90 dengue infections sampled from Colombo, Sri Lanka between 2017 – 2020 and sequenced to resolve within-host haplotypes, DENcode estimates were stable across 100 Monte Carlo iterations, yielding narrow credible intervals (median width <0.001) and consistent top-ranked transmission pairs. Sensitivity analyses using ablation experiments showed that removing either the genetic or epidemiological component substantially altered the distribution of linkage probabilities, indicating that both contribute meaningfully to the inferred transmission structure. Serotype-specific transmission networks constructed from pairwise linkage probabilities from DENcode were analysed using degree- and path-based centrality measures at probability thresholds of 0.1 and 0.5, revealing relative importance of cases to disease transmission within the community. Haplotype-derived networks were more informative than consensus-based networks (x 3.6 and x 1.6 times more edges for DENV2 and 3 respectively). DENcode is a robust framework to explore dengue transmission within a community that provides an output of network of transmission probabilities informed by pathogen genetic similarity and clinical epidemiological parameters.

Dynamics of trachoma infection in West Africa revealed by a hidden state model

2026-05-20T14:00:00Z

by Jake Carson, Thomas Crellen, Anna Borlase, Joaquin M. Prada, Robin Bailey, T. Déirdre Hollingsworth, Simon E. F. Spencer

Trachoma is estimated to be the leading infectious cause of blindness globally, predominantly affecting low-income populations with poor sanitation and hygiene. Over a decade of mass drug administration with antibiotics has led to substantial progress in control and elimination, but hotspots remain where infection persists or rebounds following mass drug administration for reasons that remain unclear. Transmission modelling is a key component of understanding these dynamics, but the complex dynamics of infection and reinfection with Chlamydia trachomatis are challenging to infer from cross–sectional surveys. Here, we analyze longitudinal data collected over six months in 1991 using multiple diagnostics from two villages in The Gambia by developing and fitting a Bayesian epidemiological model that classifies individuals into disease states at each time point using a forward-filtering backward-sampling algorithm. We find that infection risk is clustered within households and the weekly probability of transmission within a shared room is 40–fold higher than in a shared village. Infected children are estimated to contribute disproportionately to transmission, accounting for 70–90% of the force of infection within the observed period. We estimate the basic reproduction number, R₀, to be 2.2 by simulation and find that the distribution of secondary cases per individual is less aggregated than for other directly-transmitted pathogens. We further quantify heterogeneity in predisposition to becoming infected and estimate the sensitivity and specificity for PCR, antigen detection tests, and clinical examinations. Our study uncovers the natural history of trachoma infection, with implications for simulating pathogen dynamics and designing interventions to halt transmission and prevent avoidable cases of blindness.

Testing the validity and adequacy of linguistic phylogenetic analyses

2026-05-20T14:00:00Z

by Benedict King

Bayesian phylogenetics has become a standard tool in historical linguistics, and for the most part implements models borrowed from evolutionary biology. Not enough work has been done to validate the analysis set-up that has become standardised in phylolinguistics, which consists of binary data with ascertainment bias, data partitions with correlated cognate count and rate, the binary covarion substitution model, and the uncorrelated lognormal branch rate model. Here I perform a set of simulation-based calibration studies to test a typical phylolinguistic analysis in the software BEAST2. Although the analysis can correctly recover the parameters of the substitution model, complications arise due to the combination of ascertainment bias and partitions of unequal length and rate. Reweighting the partition-specific rates by the number of cognates, as is the default behaviour, leads to poorly calibrated posteriors. An alternative approach, where each meaning is assumed to come from a set of cognates of equal size, behaves correctly in simulations and is found to fit better to empirical data. I also assess the adequacy of the covarion substitution model through posterior predictive simulations. The covarion is found to fall short of approximating the true process of lexical evolution, likely due to the prevalence of semantic shift and the non-independence of cognate substitutions in real data. This work highlights the importance of thorough testing of models and their implementation in phylolinguistics, as well as the need for further research on improving models of lexical evolution.

Trial-level sequence modeling reveals hidden dynamics of dual-task interference

2026-05-20T14:00:00Z

by Rick den Otter, Anna Dame, Sjoerd Stuit, Leendert van Maanen

Theories of dual-task interference assume that the same cognitive operations underlie multitasking regardless of stimulus timing, yet this core assumption has remained untested due to methodological limitations of behavioral averaging. Here, we combine hidden multivariate pattern (HMP) analysis with deep spatiotemporal sequence modeling of single-trial EEG to uncover the neural dynamics of multitasking in the psychological refractory period (PRP) paradigm. Using a deep spatiotemporal sequence model trained on Long stimulus-onset asynchrony (SOA) trials, we identify Encoding, Central, and Response operations and show that these same operations occur in the Short SOA condition, demonstrating shared cognitive processes across interference conditions. Additionally, trial-level decoding reveals multiple distinct sequences of cognitive operations across both tasks during interference, varying both within and across individuals. These sequences predict behavioral differences in reaction time and accuracy, revealing how interference timing within the cognitive operation sequence influences performance. In other words, we found trial-by-trial variability related to individual strategies directly affecting accuracy and reaction time (RT). Our findings challenge static bottleneck accounts and establish trial-level sequence modeling as a powerful tool to investigate the hidden dynamics of multitasking.