PLOS https://journals.plos.org/ploscompbiol/ customercare@plos.org https://journals.plos.org/ploscompbiol/feed/atom All PLOS articles are Open Access. https://journals.plos.org/ploscompbiol/resource/img/favicon.ico https://journals.plos.org/ploscompbiol/resource/img/favicon.ico 2026-07-10T08:01:47Z Adam Wu Jakub Lála Quentin Trolliet Abhinav Rajendran Stefano Angioletti-Uberti 10.1371/journal.pcbi.1014433 2026-07-09T14:00:00Z 2026-07-09T14:00:00Z <p>by Adam Wu, Jakub Lála, Quentin Trolliet, Abhinav Rajendran, Stefano Angioletti-Uberti</p> We present a hybrid approach combining a protein language model (pLM) with Monte Carlo (MC) sampling for generating enzyme mutants free of mutations deleterious for structural preservation. Given the amino acid sequence of the original enzyme and a set of residues for which the local environment should be conserved, i.e., the catalytic site, our approach generates mutants that differ vastly in the overall sequence while retaining the geometry of the conserved region, thereby representing promising candidates for further experimental screening. Unlike end-to-end deep-learning approaches, whose results are harder to interpret and control, the use of a well-established, classic technique such as MC sampling allows us to easily interpret the generative process as the sampling of an energy landscape determined by the pLM. In turn, such an interpretation enables us to steer this generative process and control its outcome by making use of robust statistical mechanics concepts, e.g., temperature, thereby explicitly guaranteeing certain properties of the generated mutants. We further show, through comparison to experimentally characterised chorismate mutase variants, that low embedding energy is a necessary condition for catalytic function, providing direct experimental grounding for the energy function at the core of our approach. Given the increasing relevance of generative algorithms in the design and search for novel, optimised enzymes, we believe that our results constitute an important step for the future development of this class of techniques. To facilitate experimental verification, we finally provide over 12,500 sequences in total for 13 different enzymes involved in catalytic processes ranging from biomass degradation to DNA replication. Paolo Agliati André Urbano Pablo Lanillos Nasir Ahmad Marcel van Gerven Sander Keemink 10.1371/journal.pcbi.1014432 2026-07-09T14:00:00Z 2026-07-09T14:00:00Z <p>by Paolo Agliati, André Urbano, Pablo Lanillos, Nasir Ahmad, Marcel van Gerven, Sander Keemink</p> Neurons communicate with downstream systems via sparse and incredibly brief electrical pulses, or spikes. Using these events, they control various targets such as neuromuscular units, neurosecretory systems, and other neurons in connected circuits. This gave rise to the idea of spiking neurons as controllers, in which spikes are the control signal. Using instantaneous events directly as the control inputs, also called ‘impulsive control’, is challenging as it does not scale well to larger networks and has low analytical tractability. Therefore, current spiking control usually relies on filtering the spike signal to approximate analog control. This ultimately means spiking neural networks (SNNs) have to output a continuous control signal, necessitating continuous energy input into downstream systems. Here, we circumvent the need for rate-based representations, providing a scalable method for task-specific spiking control with sparse neural activity. In doing so, we take inspiration from both control theory and neuroscience, and define a spiking rule where spikes are only emitted if they bring a dynamical system closer to a target. From this principle, we derive the required connectivity for an SNN, and show that it can successfully control linear systems. We show that for physically constrained systems, predictive control is required, and the control signal ends up exploiting the passive dynamics of the downstream system to reach a target. Finally, we show that the paradigm scales to both high-dimensional systems and bio-inspired motor control tasks. Importantly, in all cases, we maintain a closed-form mathematical derivation of the network connectivity, the network dynamics and the control objective. This work advances the understanding of SNNs as biologically-inspired controllers, providing insight into how real neurons could exert control, and enabling applications in neuromorphic hardware design. Arne Elofsson Nir Ben-Tal 10.1371/journal.pcbi.1014467 2026-07-08T14:00:00Z 2026-07-08T14:00:00Z <p>by Arne Elofsson, Nir Ben-Tal</p> We have seen more progress in computational biology for macromolecules in the last five years than we experienced in the five preceding decades. Thus, it is very challenging to forecast future progress. It is possible that we have reached a plateau, and we will be stuck with similar problems as we have today. Still, it is also possible that the field will continue its rapid progress and completely transform other fields, such as biochemistry, molecular and cell biology, and medicine. It is also possible that general AI will take over, and all scientific endeavours will be conducted without human input. To be honest, we do not know what will happen, but we will highlight a few of the challenges and the most critical research questions that we face today. Hopefully, these will be resolved within the following decades, or hopefully much earlier. Looking back over the last decade, we can see that machine learning and deep learning have become significantly more popular (<i>T</i>-test residual > 2) among the papers published within our section of PlosCB. We do believe that this trend will continue; therefore, we focus on the challenges that must be overcome for it to make significant and notable contributions. The future of computational biology for macromolecules in 20 years is likely to be characterised by transformative advances in accuracy, automation, integration, and explainability, with AI playing a role in one form or another. Arjun Singh Ali I. Ukperaj Gabriel F. Porto Gregory L. Dignon 10.1371/journal.pcbi.1014462 2026-07-08T14:00:00Z 2026-07-08T14:00:00Z <p>by Arjun Singh, Ali I. Ukperaj, Gabriel F. Porto, Gregory L. Dignon</p> Intrinsically disordered proteins (IDPs) exhibit phase separation behavior that is closely linked to their degree of single-chain compaction, which in turn is governed by both amino acid composition and sequence patterning. Existing metrics such as sequence charge decoration (SCD) and sequence hydropathy decoration (SHD) describe these effects but are largely limited to describing differences between sequences of similar length and overall composition. In this work, we present a shuffle-based normalization scheme for SCD and SHD, enabling comparison of sequence patterning between very different IDP sequences. Leveraging this normalization scheme toward design space, we develop a Monte Carlo based sequence design algorithm that generates novel IDPs with desired patterning features. Our design framework is further strengthened by incorporating additional metrics such as sequence aromatic decoration (SAD), compositional RMSD, and a previously developed sequence based ΔG predictor. We validate our approach through coarse-grained MD simulations, showing that the designed sequences exhibit tunable phase behavior. This strategy lays the groundwork for rational design of IDPs for biomedical and biotechnology applications, as well as basic biophysical research. Soha Ben Tahar Ester Comellas Timothy Duerr Dareen Bakr James Monaghan Jose J. Muñoz Sandra J. Shefelbine 10.1371/journal.pcbi.1014348 2026-07-07T14:00:00Z 2026-07-07T14:00:00Z <p>by Soha Ben Tahar, Ester Comellas, Timothy Duerr, Dareen Bakr, James Monaghan, Jose J. Muñoz, Sandra J. Shefelbine</p> The vertebrate limb provides an interesting system to study how tissue growth and molecular signaling interact to shape complex skeletal patterns. How these processes are coordinated across space and time is not fully understood. This study introduces a computational tool to examine how growth interacts with positional cues and self-organizing patterning mechanisms to shape skeletal structures in both mice and axolotl limbs. We developed the Growth-Reaction-Diffusion (GRD) framework, a reaction-diffusion system within a growing domain, where reaction represents the regulation of patterning cues and diffusion captures their spatial propagation. The relative contribution of growth, reaction and diffusion is modulated through two non-dimensional parameters, whose spatial variation is informed by positional cues derived from experimental morphogen maps. This formulation normalizes the reaction-diffusion equation relative to growth, enabling investigation of how different spatiotemporal regimes of growth interact with reaction and diffusion to produce whole limb patterning. The GRD framework captures the progressive formation of all limb segments: the humerus, radius/ulna, and the digits patterns. Our simulations indicate that in the proximal region (humerus, radius/ulna) the contributions of growth, reaction and diffusion are equally important to patterning, but in the distal elements (digits) the reaction and diffusion contributions are much greater than the contribution of growth to the formation of the digits. Through a single framework, we simulate the whole-limb skeletal patterns in both mice and axolotls, despite their morphological differences. These results highlight the model’s potential to explore conserved and divergent features of limb development from an evolutionary perspective through a unified mechanism across species. Matto Leeuwis Nikki van Aerts Ajay Seth Patrick A. Forbes 10.1371/journal.pcbi.1013522 2026-07-07T14:00:00Z 2026-07-07T14:00:00Z <p>by Matto Leeuwis, Nikki van Aerts, Ajay Seth, Patrick A. Forbes</p> Human movement control is shaped by competing objectives, among which minimizing energy expenditure plays a central role, particularly in determining preferred walking patterns. Whether energetic cost similarly influences standing balance remains unclear because it has not been systematically quantified across a range of natural postures. Importantly, standing is the resting state from which most walking begins, suggesting that the optimization of posture may also reflect the energetic demands of initiating gait. In this study, we use a combination of indirect calorimetry and musculoskeletal simulations to characterize the energetic cost of standing and gait initiation across natural standing postures and investigate whether humans optimize energy expenditure under these conditions. In Experiment 1 (<i>N</i> = 13), we measured metabolic cost at preferred and six different prescribed whole-body lean angles. Energy expenditure was lowest at a slight anterior lean (1.15°) and increased monotonically with whole-body lean angle in either direction, rising twice as fast posteriorly compared to anteriorly. This asymmetry challenges the common modeling simplification that effort is symmetric and linear or quadratic with lean angle. Furthermore, participants preferred body angles (1.50 ± 0.73°) with similar energy expenditure to the minimum-cost lean but with significantly more postural variability, suggesting that strict postural regulation was not necessary for minimizing energetic cost. In Experiment 2 (<i>N</i> = 20), participants initiated forward and backward walking from preferred or prescribed lean angles. Participants did not alter their standing posture before expected gait initiations in the forward or backward direction, consistent with musculoskeletal simulations showing that leaning further in the anticipated direction did not significantly improve gait initiation time or energetic costs. Together, these findings suggest that postural strategies optimize energy efficiency when permitted by the demands of movement readiness. Our study quantifies the energetic cost landscape that governs human postural control, challenges widely used symmetric estimations of this cost, and offers an empirical foundation for developing more accurate simulations of posture and energy expenditure. Balázs Erdős Christos Chatzis Jonathan Thorsen Jakob Stokholm Age K. Smilde Morten A. Rasmussen Evrim Acar 10.1371/journal.pcbi.1014486 2026-07-06T14:00:00Z 2026-07-06T14:00:00Z <p>by Balázs Erdős, Christos Chatzis, Jonathan Thorsen, Jakob Stokholm, Age K. Smilde, Morten A. Rasmussen, Evrim Acar</p> Longitudinal microbiome studies provide critical insights into microbial community dynamics and their relation to host health. Tensor decompositions offer a powerful framework for the unsupervised analysis of such data, yielding interpretable low-dimensional temporal patterns. However, existing approaches based on the CANDECOMP/PARAFAC (CP) model assume common temporal dynamics for all subjects and therefore cannot capture subject-specific trajectories. To address this limitation, we introduce a novel analytical framework based on PARAFAC2 to explicitly model subject-specific variations, such as shifts and delays in temporal patterns. Through systematic comparisons on simulated and real-world datasets—including studies of infant gut maturation and dietary interventions—we demonstrate that PARAFAC2 outperforms CP in capturing subject-specific temporal trajectories, and enables the discovery of biologically relevant patterns that are overlooked by CP. Furthermore, we introduce replicability as a robust criterion for selecting the number of model components, ensuring that the extracted patterns are replicable. The PLOS Computational Biology Staff 10.1371/journal.pcbi.1014482 2026-07-06T14:00:00Z 2026-07-06T14:00:00Z <p>by The PLOS Computational Biology Staff </p> Hugues Berry Lyle J. Graham Kim T. Blackwell 10.1371/journal.pcbi.1014468 2026-07-06T14:00:00Z 2026-07-06T14:00:00Z <p>by Hugues Berry, Lyle J. Graham, Kim T. Blackwell</p> Romesa Khan Hongsheng Zhong Shuvam Das Jack Cai Matthias Niemeier 10.1371/journal.pcbi.1014435 2026-07-06T14:00:00Z 2026-07-06T14:00:00Z <p>by Romesa Khan, Hongsheng Zhong, Shuvam Das, Jack Cai, Matthias Niemeier</p> Seminal frameworks of predictive coding propose a hierarchy of generative modules, each attempting to infer the neural representation of the module one level below; the predictions are carried by top-down feedback projections, while the predictive error is propagated by reciprocal forward pathways. Such symmetric feedback connections support visual processing of noisy stimuli in computational models. However, neurophysiological studies have yielded evidence of asymmetric cortical feedback connections. We investigated the contribution of neural feedback in visual processing for computing grasp parameters, by utilizing convolutional neural network models that had been augmented with predictive feedback and were trained to compute grasp positions for real-world objects. After establishing an ameliorative effect of symmetric feedback on grasp detection performance when evaluated on noisy stimuli, we characterized the performance effects of asymmetric feedback, similar to that observed in the cortex. Specifically, we tested model variants extended with <i>short</i>-, <i>medium</i>-, <i>long</i>- and <i>longer</i>-range feedback connections (i) originating at the same source layer or (ii) terminating at the same target layer. We found that the performance-enhancing effect of predictive coding under adverse conditions was optimal for <i>medium</i>-range asymmetric feedback. Moreover, this effect was most prominent when <i>medium</i>-range feedback originated at a level of representational abstraction that was proximal to the input layer, in contrast to more distal layers. To conclude, our simulations show that introducing biologically realistic asymmetric predictive feedback improves model robustness to noisy visual stimuli in a neural network model optimized for grasp detection. Joseph T. Benthal Justin A. Avila Jeffrey R. Smith E. Michelle Southard-Smith 10.1371/journal.pcbi.1014424 2026-07-06T14:00:00Z 2026-07-06T14:00:00Z <p>by Joseph T. Benthal, Justin A. Avila, Jeffrey R. Smith, E. Michelle Southard-Smith</p> Hirschsprung disease (HSCR) is characterized by absence of enteric ganglia (aganglionosis) along variable lengths of the distal intestine. This disorder results from deficient colonization of fetal intestine by enteric neural crest-derived cells (ENCDCs). HSCR exhibits complex, multifactorial inheritance with penetrance and severity varying widely even within families. <i>SOX10</i> is among causal genes that predispose to aganglionosis. Yet, how gene interactions influence severity of HSCR aganglionosis is not understood. Prior mapping of aganglionosis modifiers was achieved in a standard F₁-intercross utilizing the <i>Sox10</i>^Dom HSCR mouse model. Here we deploy a novel strategy of genotyping an extended pedigree pedigree of <i>Sox10</i>^Dom mice on a mixed genetic background. GWAS in this pedigree points to novel aganglionosis modifier intervals with replication and refinement of prior modifier regions. Complementary omics analysis of the developing Enteric Nervous System (ENS) enabled identification of multiple high-priority candidate genes within these modifier intervals based on gene expression, chromatin accessibility, and presence of conserved SOX10 binding motifs. We implemented a prioritization pipeline for ranking potential modifiers that generated candidate lists including several well-known for effects on ENS development as well as multiple novel genes. Among the novel genes, <i>Dach1</i> ranked as a top priority candidate gene for modifying migration of ENCDCs and thus influencing aganglionosis severity. The results identify genome intervals with intrinsic genes that are logical candidates for modifying <i>Sox10</i>^Dom aganglionosis severity. We also note that several human orthologs to aganglionosis modifier candidate genes are within linkage disequilibrium blocks containing genetic variants associated with human gut motility disorders, which offers opportunity for gaining biological insight into human HSCR severity. Eduardo H. Colombo Corina E. Tarnita Juan A. Bonachela 10.1371/journal.pcbi.1014411 2026-07-06T14:00:00Z 2026-07-06T14:00:00Z <p>by Eduardo H. Colombo, Corina E. Tarnita, Juan A. Bonachela</p> The problem of pattern and scale remains central in ecology, bridging fundamental and applied questions. Marine microbial communities are a case in point. For instance, to understand the role of zooplankton in oceanic biogeochemistry, their response to changes in environmental conditions, and the implications for ecosystem services (e.g., fisheries), it is critical to understand zooplankton trophic interactions and how they change in a rapidly changing climate. This understanding, however, remains elusive because, unlike for phytoplankton, for which remote sensing of macroscale patterns can provide insight into their microscale dynamics and community composition, obtaining this information for zooplankton largely rests on quantifying the difficult-to-monitor microscale interactions among millions of individuals with different behaviors, and between individuals and their environment. Here, we investigate whether it is possible to obtain indirect information on zooplankton from the macroscale spatial distribution of their prey. To tackle this “problem of scale,” we develop a rigorous coarse-graining methodology that connects individual-level properties with macroscale spatial patterns. We demonstrate that the shape of the prey spatial distribution can encode information about zooplankton feeding behavior and community dynamics. Specifically, we predict a change in dominant feeding behavior—from non-motile to motile feeding—as one moves from areas of high to areas of low prey density. These computational results are validated by our analysis of satellite images of oceanic blooms around the globe, which suggests novel opportunities for remote sensing approaches: the potential tracking of consumer behavioral signatures in the large-scale patterns of the resource. Importantly, the scaling-up methodology developed here to check for those signatures is general, and can be used to link scales rigorously and systematically in any system in which the complexity of individual dynamics makes connecting scales intractable. Naomi Auer Lars Chen Jakob Stubenrauch Benjamin Lindner Richard Kempter 10.1371/journal.pcbi.1013235 2026-07-06T14:00:00Z 2026-07-06T14:00:00Z <p>by Naomi Auer, Lars Chen, Jakob Stubenrauch, Benjamin Lindner, Richard Kempter</p> The brain can efficiently learn and form memories based on limited exposure to stimuli, often even in single trials. Two key factors are believed to support this ability: large synaptic plasticity to strongly encode new memories; and sparse coding, leading to low overlap between memory representations and to small interference. Therefore, increased sparseness can also improve memory capacity. However, it is not well understood how the strength of plasticity of synapses affects capacity. Here, we analyze the combined impact of population sparseness and strength of plasticity on memory capacity. Specifically, we explore how the strength of plasticity that maximizes capacity depends on the sparseness of the neural code. To this end, we study a feedforward network with Hebbian and homeostatic plasticity and a two-state synapse model. The network learns to associate sparse binary input-output pattern pairs. The strength of plasticity is modeled as the probability of synaptic changes. Our results are based on both network simulations and an analytical theory, predicting the expected memory capacity in dependence on strength of plasticity and population sparseness. For both perfect and noisy input patterns, we find that the optimal strength of plasticity increases with increasing pattern sparseness and that this effect is more pronounced for input than for output sparseness. Interestingly, the optimal strength of plasticity remains the same across different network sizes if the number of active units in an input pattern is constant. While the memory capacity obtained at the optimal strength of plasticity increases monotonically with output sparseness, its dependence on input sparseness is non-monotonic. Overall, we provide the first detailed investigation of the interactions between population sparseness, strength of plasticity, and memory capacity. Our findings suggest that differences in sparseness between brain regions may underlie observed differences in how strongly these regions adapt and how quickly they learn. Jean Fan 10.1371/journal.pcbi.1014471 2026-07-02T14:00:00Z 2026-07-02T14:00:00Z <p>by Jean Fan</p> Since the founding of PLOS Computational Biology 20 years ago, genomics research has advanced at a remarkable pace. In this 20th anniversary commentary, as an Editor for the Journal Section of Genomics, Epigenomics, & Proteomics, I take a data-driven dive into genomics research at PLoS Computational Biology by analyzing all submitted research papers in this section since 2017. This time window reflects the limits of the Editorial Manager records we were able to assemble, but it also coincides approximately with my own independent journey in this field. While this time window does not capture the journal’s full 20-year history, I hope this analysis will offer a data-driven reflection on how genomics research within the journal has evolved and provide a putative trajectory on how the field will surely continue to grow into the future. Giulia Pullano Shweta Bansal Stefania Rubrichi Vittoria Colizza 10.1371/journal.pcbi.1014427 2026-07-02T14:00:00Z 2026-07-02T14:00:00Z <p>by Giulia Pullano, Shweta Bansal, Stefania Rubrichi, Vittoria Colizza</p> Human mobility fundamentally shapes the spatial spread of infectious diseases, yet the level of detail required from mobility data to accurately inform epidemic models remains unclear. Mobile phone records offer unprecedented resolution on population movements, but little attention has been devoted however to determining (i) which aspects of mobility are epidemiologically relevant and (ii) what level of data resolution is necessary to capture spatial invasion dynamics. Using mobile phone records from 9.5 million users in Senegal (approximately 80% of the population), we systematically compare three approaches to aggregating mobility data for epidemic modeling. These approaches span a range of resolutions: high-resolution tracking of all individual displacements between consecutive visited locations (HR), medium-resolution accounting for time spent in all visited locations (MR), and low-resolution identification of the most-visited location (LR). We incorporate these mobility representations into a metapopulation epidemic model that explicitly accounts for transmission from residents, visitors, and returning travelers, and simulate diseases with varying transmissibility corresponding to controlled epidemic conditions, seasonal influenza–like transmission, and highly transmissible pathogens. We find that preserving all observed displacements in individual trajectories does not necessarily improve the epidemiological relevance of mobility in pathogens spatial transmission. Instead, displacement-based networks fragment long-range trips and underestimate key spatial connections relevant for disease spread. In contrast, approaches that capture where individuals spend most of their time (such as home, work, or school) more accurately reproduce spatial invasion patterns. Accounting for additional daily activities beyond these primary locations provides little additional epidemiological information. Our results suggest that lower-resolution mobility indicators capturing time spent at key locations are sufficient to inform predictive epidemic models. These findings have important implications for both epidemic modeling and data governance, indicating that mobile phone data can be aggregated to reduce privacy issues while still providing the essential information needed to model spatial disease transmission. Wenran Li Shijia Yu Yingyu Cheng Sijia Wang 10.1371/journal.pcbi.1014476 2026-07-01T14:00:00Z 2026-07-01T14:00:00Z <p>by Wenran Li, Shijia Yu, Yingyu Cheng, Sijia Wang</p> DNA methylation is a key epigenetic modification that regulates gene expression and plays a vital role in cell differentiation, development, and tumorigenesis. However, large-scale experimental profiling of genome-wide DNA methylation remains time-consuming and limited in coverage. We present DeepMethylation, a deep learning framework that integrates DNA sequence and tissue-specific epigenomic features to predict CpG methylation status across the genome. DeepMethylation achieves state-of-the-art performance (average AUROC 0.909) across tissues, accurately imputes methylation beyond array-covered sites, and enables robust extension from 450k to EPIC array coverage. Feature importance analysis revealed consistent patterns of epigenomic feature contributions across tissues. We also introduced Delta DeepMethylation (DDM), a variant evaluation model to estimate the epigenetic effects of SNPs on DNA methylation. DDM-predicted variant effects were consistent with methylation quantitative trait loci (mQTLs) and not confounded by linkage disequilibrium (LD). Our framework provides a powerful tool for genome-wide methylation prediction and regulatory variant interpretation across tissues. Michael Zietz Kathleen LaRow Brown Undina Gisladottir Nicholas P. Tatonetti 10.1371/journal.pcbi.1014431 2026-07-01T14:00:00Z 2026-07-01T14:00:00Z <p>by Michael Zietz, Kathleen LaRow Brown, Undina Gisladottir, Nicholas P. Tatonetti</p> Complex diseases are a major challenge, and genetics underlie a large fraction of the risk for these diseases. Observational data are helpful for this research due to large scale, cost-effectiveness, information on many different conditions, and future scalability, but they reflect factors such as healthcare processes, access to care, and broader societal effects like systemic biases. Here, we introduce MaxGCP, a phenotyping method designed to purify the genetic signal in observational data. MaxGCP optimizes a phenotype definition to maximize its coheritability—the genetic covariance between two traits normalized by their phenotypic standard deviations—with the complex trait of interest. Unlike previous phenotype-combination methods, MaxGCP is phenotype-specific, has linear computational complexity in the number of features, and does not require manual feature selection. In an analysis of stroke, we found that MaxGCP boosts study power by more than 13 percent compared to conventional, single-code phenotype definitions. MaxGCP is a powerful tool for genetic discovery in observational data, and we anticipate that it will be broadly useful for studying complex diseases using observational data. Boxuan Wang Eugenio Valdano 10.1371/journal.pcbi.1014425 2026-07-01T14:00:00Z 2026-07-01T14:00:00Z <p>by Boxuan Wang, Eugenio Valdano</p> Assessing epidemic risk following pathogen introduction is crucial in infectious disease epidemiology. Risk is commonly encoded through reproduction ratios, which underpin operational decision-making. In spatially structured populations, both local and cross-community transmission shape epidemic trends, a feature that standard reproduction ratios fail to capture simultaneously. Here, we use multitype branching processes to define the outbreak reproduction ratio <i>R</i>^ob, a reformulation applicable across pathogens, epidemics and transmission routes, enabling community-specific, but system-aware, risk assessment. We test <i>R</i>^ob on respiratory pathogens and estimate it prior to emergence using aggregated contact matrices, enabling spatially resolved risk assessment even with limited data and computational resources. Estimates across countries reveal heterogeneous spatial risk, not captured by standard metrics. <i>R</i>^ob can also be estimated from early-phase surveillance data, as we show using SARS-CoV-2 in Canada, where it correctly identifies community risk. <i>R</i>^ob represents a concise and practicable framework for interpreting epidemic risk in spatially structured populations. Elijah Gunther Pablo G. Camara 10.1371/journal.pcbi.1014487 2026-06-30T14:00:00Z 2026-06-30T14:00:00Z <p>by Elijah Gunther, Pablo G. Camara</p> The three-dimensional structure of proteins is intimately linked to their function, yet establishing comprehensive frameworks for systematically comparing and organizing protein structures across the proteome remains a significant challenge. Here, we introduce GWProt, a computational framework that leverages recent advances in metric geometry, such as Gromov-Wasserstein couplings, for protein structure alignment and analysis. GWProt enables the integration of biochemical information into structural comparisons and introduces the concept of local geometric distortion, a measure that captures local conformational differences. We demonstrate the utility of this framework by identifying conformational switches within individual proteins, detecting functional domains shared among evolutionarily distant viral proteins, revealing topological rearrangements in homologous folds, and uncovering recurrent short structural motifs underlying functional domains across the human proteome. Collectively, these results establish the use of metric geometry as a versatile and quantitative framework for the systematic comparative analysis of protein structures, complementing existing approaches for elucidating protein organization. Clemens Kohl Martin Vingron 10.1371/journal.pcbi.1014418 2026-06-30T14:00:00Z 2026-06-30T14:00:00Z <p>by Clemens Kohl, Martin Vingron</p> Clustering for single-cell RNA-seq aims at finding similar cells and grouping them into biologically meaningful clusters. Many available clustering algorithms however do not not provide the cluster defining marker genes or are unable to infer the number of clusters in an unsupervised manner as well as lack tools to easily determine the quality of the label assignments. Therefore, clustering quality is commonly evaluated by visually inspecting low-dimensional embeddings as produced by, e.g., UMAP or t-SNE. These embeddings can, however, distort the true cluster structure and are known to produce radically different embeddings depending on the chosen hyperparameters. In order to improve the interpretability of clustering results, we developed CAdir, a clustering algorithm that can infer the number of clusters in the data, determine cluster specific genes and provides easy to interpret diagnostic plots. CAdir exploits the geometry induced by correspondence analysis (CA) to cluster cells as well as cluster associated genes based on their direction in CA space. Using the angle between the cluster directions, it is able to automatically infer the number of clusters in the data by merging and splitting clusters. A comprehensive set of diagnostic and explanatory plots provides users with valuable feedback about the clustering decisions and the quality of the final as well as intermediary clusters. CAdir is scalable to even the largest data set and provides similar clustering performance to other state-of-the-art cell clustering algorithms in our benchmarking. CAdir can be downloaded from GitHub: https://github.com/VingronLab/CAdir. Niklas Müller Hongye Chen Sofie Wahlberg H. Steven Scholte Iris I. A. Groen 10.1371/journal.pcbi.1014371 2026-06-30T14:00:00Z 2026-06-30T14:00:00Z <p>by Niklas Müller, Hongye Chen, Sofie Wahlberg, H. Steven Scholte, Iris I. A. Groen</p> Retinotopic tuning of neural populations is a key organizing principle of human visual cortex. However, state-of-the-art models that predict neural recordings based on task-optimized Convolutional Neural Networks (CNNs) do not take this retinotopic organization into account. Furthermore, while retinotopic tuning in visual cortex has been studied extensively using functional magnetic resonance imaging, the temporal dynamics of processing information from distinct parts of the visual field are less well understood. Here, we reveal distinct temporal profiles for foveal and peripheral visual information processing by implementing multiple spatial sampling strategies on feature maps of CNNs into encoding models that predict human electroencephalography (EEG) responses. Using large, high-quality natural scene images, we show that processing of peripheral information precedes that of foveally sampled information. This temporal difference is best modeled when applying a differential spatial transform to CNN feature maps that is derived from empirical measurements of human retinal ganglion cells. We directly confirm this temporal difference experimentally by mutually exclusive stimulation of foveal and peripheral visual field regions. Last, we introduce a novel, data-driven method of recovering visual field information from neural data, highlighting and quantifying spatial, retinotopic information contained in temporally specific EEG recordings. Together, these results provide novel neural evidence for a temporal coarse-to-fine visual processing hierarchy in the processing of natural images that is directly linked to distinct spatial information sampling. Aligning the spatial sampling of humans and CNN encoding models not only improves predictions of neural responses but also demonstrates that EEG recordings contain a significant amount of temporally encoded retinotopic information. We make our large-scale EEG dataset including high-resolution natural scene images publicly available to enable future research into naturalistic visual processing. Irina Tolkova Holger Klinck Dana A. Cusano Anke Kügler Susan E. Parks 10.1371/journal.pcbi.1013321 2026-06-30T14:00:00Z 2026-06-30T14:00:00Z <p>by Irina Tolkova, Holger Klinck, Dana A. Cusano, Anke Kügler, Susan E. Parks</p> Anthropogenic noise has increased ambient sound levels across the globe, both underwater and on land. Among its many negative impacts, heightened noise can impair communication in vocal animals through acoustic masking. Conceptually, noise reduces the animal’s <i>communication space</i> – the area in which an individual animal can effectively convey information to a conspecific listener. Previous studies have estimated the communication space using sound propagation models and/or behavioral studies. However, studies frequently equate signal recognition with signal detection – a necessary but not sufficient precondition – thereby persistently overestimating spatial coverage and underestimating anthropogenic impacts. Measuring communication is inherently difficult, and varies with taxa, call type, and context, leading to significant data gaps in key parameters. We propose that deep learning creates an opportunity to estimate biologically-relevant communication, even for data-limited species. In particular, we present a case study with the critically endangered North Atlantic right whale (<i>Eubalaena glacialis</i>; hereafter NARW). Prior research has demonstrated that the upcall – a low-frequency contact call produced across ages and sexes – encodes individual identity. We therefore consider a dataset of NARW vocalizations recorded with on-animal archival tags, spanning 234 samples across 11 individuals from 3 sites. First, we demonstrate that audio embeddings from the BirdNET model can robustly distinguish individual right whales. Then, we simulate the effect of varying ambient noise levels to estimate signal excess for both signal detection and individual identification, finding that an additional ≥7 dB is necessary for the model to distinguish individuals. Altogether, we hope this work provides both a methodological advance for individual identification and a framework for better understanding anthropogenic impacts on vocal wildlife. Roghaye Mohammadbeygi Patrick F. Suthers Fang-Yu Chung Brian F. Pfleger Costas D. Maranas 10.1371/journal.pcbi.1014469 2026-06-29T14:00:00Z 2026-06-29T14:00:00Z <p>by Roghaye Mohammadbeygi, Patrick F. Suthers, Fang-Yu Chung, Brian F. Pfleger, Costas D. Maranas</p> Despite progress in automated gene annotation, many deficiencies and knowledge gaps remain, even for well-studied organisms. Of particular concern is the accuracy and detail of annotations for transporters of various organic substrates and products of metabolism and for enzymes that do not share sequence homology with well-characterized strains. Unfortunately, annotation errors present in earlier genome-scale metabolic (GSM) models propagate to newer models with few opportunities for later correction. Here, we introduce a systematic computational procedure that applies the <i>Escherichia coli</i> genome-scale metabolic model <i>i</i>ML1515, extended with transcriptional regulatory rules, to design auxotrophs that can grow on glucose but fail to grow on different carbon substrate(s) unless rescued with the addition of an ORF encoding a complementation metabolic function (transport and enzymatic reactions). Using the <i>E. coli</i> GSM model supplemented with regulatory rules that quantify growth/no growth outcomes on different organic substrates, we identified 258 distinct auxotrophic designs (97 single-gene, 142 double-gene, and 19 triple-gene knockouts) for which specific single functions can uniquely complement them. Experimental validation of 61 single-knockout strains demonstrated 59% confirmed auxotrophy and 28% partial auxotrophy. We envision that this collection of auxotrophic strains can be used to disambiguate the metabolic role of unannotated or poorly annotated genes. Yunmeng Zhang Xiuhong Wu Qiutong Wang Lin Shi Meiling Liu Guohua Wang 10.1371/journal.pcbi.1014461 2026-06-29T14:00:00Z 2026-06-29T14:00:00Z <p>by Yunmeng Zhang, Xiuhong Wu, Qiutong Wang, Lin Shi, Meiling Liu, Guohua Wang</p> Predicting Herb–Disease Associations (HDA) is pivotal for modernizing Traditional Chinese Medicine (TCM); however, this is impeded by data heterogeneity and the complex, multi-component mechanisms of herbal medicines. Existing drug–disease prediction models often struggle to capture high-order structural patterns and resolve semantic inconsistencies intrinsic to herbs. To overcome these limitations, we present HData, a standardized benchmark dataset that integrates herbal medicinal properties, chemical compositions, and disease associations. We further propose GHF-ACL, a novel multi-order graph contrastive learning framework designed for HDA prediction. Specifically, GHF-ACL explicitly models low-order functional similarities via a herb–disease similarity graph while capturing high-order component interactions through a herb–chemical hypergraph. Furthermore, an adaptive gating-guided structural interaction module aligns heterogeneous graph representations into a unified latent space, and hierarchical contrastive learning enforces consistency across structural views. Extensive experiments on five datasets demonstrate that GHF-ACL achieves superior or competitive performance over six state-of-the-art models across most metrics, with significant improvements over the best-performing baseline model in AUPR (+4.8% on LRSSL, + 3.81% on Cdata), F1 score, and Recall. These results underscore the model’s superior capability in detecting true positive associations within imbalanced biomedical data. By synergizing multi-view graph modeling, semantic fusion, and contrastive regularization, this work establishes a unified framework for HDA prediction, offering valuable insights for computational TCM and data-driven drug discovery. Miruna Cotet David Poensgen Ian Krajbich 10.1371/journal.pcbi.1014459 2026-06-29T14:00:00Z 2026-06-29T14:00:00Z <p>by Miruna Cotet, David Poensgen, Ian Krajbich</p> Learning is fundamental to adaptive behavior. In the typical learning task, each action is associated with only one outcome, which could be immediate or delayed. However, actions often have multiple consequences that unfold over time. Here, we used behavioral and eye-tracking experiments to study how people learn when their choices yield both immediate and delayed reward information. Importantly, the rewards themselves were all delivered at the end of the study so there was no reason to weight immediate and delayed reward information differently. Instead, we found that our subjects overweighted immediate reward information. Moreover, this bias increased over the course of the experiment and was still present when learning from others’ choices. The gaze data reveal mixed evidence that subjects looked more at immediate vs. delayed feedback, and across subjects, the relative dwell proportion did not predict the behavioral bias. Our results indicate that people prioritize not just immediate rewards, but immediate reward information. Unlike temporal discounting, this form of impatience is a clear mistake and leads to objectively worse outcomes. Alon Korngreen 10.1371/journal.pcbi.1014458 2026-06-29T14:00:00Z 2026-06-29T14:00:00Z <p>by Alon Korngreen</p> Biophysically detailed neuron models are often built as a one-way pipeline in which voltage-clamp data are reduced to a single set of best-fit channel parameters, which are then combined into a deterministic spiking model. This practice discards experimentally observed scatter and fitting uncertainty, obscuring the mechanisms by which robustness and degeneracy arise in excitable systems. Here, I reintroduce fitted-parameter uncertainty into the Hodgkin-Huxley model and embed uncertainty and global sensitivity analysis into model construction. I digitized sodium and potassium rate-constant data from the original Hodgkin and Huxley figures and used bootstrap resampling to estimate uncertainty in the fitted voltage-dependent kinetic parameters. I then propagated these uncertainty estimates through a spatially extended squid axon cable model using large-scale Monte Carlo simulations, in which each sample defined a complete set of kinetic, conductance, passive, and structural parameters. At the channel level, first-order Sobol sensitivity indices revealed that all kinetic parameters contribute to output variance in a strongly time-dependent manner, with distinct parameters controlling transient and steady-state behavior for potassium and sodium conductances. At the level of neuronal excitability, the simulations produced a heterogeneous population of firing behaviors, including non-firing, phasic, regular, and spontaneous activity. Across stimulus amplitudes, the dominant firing mode was a single spike at stimulus onset, consistent with the physiological role of the squid giant axon in rapid escape behavior. The canonical 1952 Hodgkin-Huxley parameter set fell within the regularly firing minority subpopulation, rather than representing a unique or dominant solution. In the phasic subpopulation, action potential propagation and conduction velocity varied widely yet remained within experimental ranges. Finally, global sensitivity analysis during spiking showed uniformly small first-order Sobol indices but large total-order indices, indicating that excitability is governed primarily by strong interactions among all parameters rather than by any subset. Together, these results support reframing the Hodgkin-Huxley model as an experimentally constrained ensemble of behaviors rather than a single privileged parameter set, with physiologically relevant firing patterns emerging from structured regions of the parameter space. Yuechen Liu Zishun Wang Chen Qiao Zongben Xu 10.1371/journal.pcbi.1014438 2026-06-29T14:00:00Z 2026-06-29T14:00:00Z <p>by Yuechen Liu, Zishun Wang, Chen Qiao, Zongben Xu</p> Hippocampal formation (HF) supports both the temporary maintenance of task-relevant information and rapid relearning when task structure is preserved. Here we ask what circuit mechanism can link these two functions within a single framework. We propose a model named Generalization and Associative Temporary Encoding (GATE), whose core idea is a self-gating re-entrant EC3–CA1–EC5–EC3 loop. In each lamella, EC3 provides a memory substrate, CA1 selectively reads out the retained information under CA3 gating, and EC5 feeds back to regulate the next EC3 state. Repeating this loop across dorsoventral lamellae yields representational scales that range from local cue-dependent coding to a broader task-related structure. In simple tasks, the single-lamellar model captures selective maintenance and produces place- and splitter-like CA1 activity. In more complex tasks, the multi-lamellar model develops lap, evidence, trace, and other task-relevant representations. Under structure-preserving changes in sensory coding, positional scaffold, or task parameters, the model reuses learned representations and relearns faster. GATE provides a hypothesis-generating computational framework for studying how hippocampal-like circuit motifs may support selective memory gating and structure-preserving relearning. Cristina Colangelo Alberto Muñoz Alberto Antonietti Vishal Sood Alejandro Antón-Fernández Joni Herttuainen Armando Romani Javier DeFelipe Srikanth Ramaswamy 10.1371/journal.pcbi.1014460 2026-06-26T14:00:00Z 2026-06-26T14:00:00Z <p>by Cristina Colangelo, Alberto Muñoz, Alberto Antonietti, Vishal Sood, Alejandro Antón-Fernández, Joni Herttuainen, Armando Romani, Javier DeFelipe, Srikanth Ramaswamy</p> The hindlimb representation in the somatosensory cortex of two-week old Wistar rats has been a valuable model system for dissecting the microcircuitry of neurons and their synaptic connections. In this study, we present a comprehensive experimental dataset quantifying the fiber length per cortical volume and the density of varicosities for cholinergic, catecholaminergic, and serotonergic neuromodulatory systems within the cortical neuropil using immunocytochemical staining and stereological techniques, along with a methodological framework for generating biophysically detailed computational models from these data. Acquired data were integrated into a biophysically detailed computational model of the somatosensory cortex to explore the anatomical organization and functional implications of neuromodulatory innervation. We found that neuromodulatory innervation, although sparse, substantially impacts network activity. Network simulations support the hypothesis that acetylcholine suppresses slow oscillations and promotes the desynchronization of cortical networks, consistent with the extensive findings in existing literature. Additionally, the temporal properties of acetylcholine modulation are consistent with synaptic rather than volume release. Furthermore, we found that the release of dopamine and serotonin in sensory cortices induces network desynchronization by inhibiting delta oscillations and that serotonin also initiates the emergence of theta oscillations, pointing to previously unexplored aspects of their function in governing cortical network activity. The experimental data and the biophysical computational model are available as an open-access community resource. Kexin Ma Ningjing Wang Zai Yang Robert A. Cheke Biao Tang 10.1371/journal.pcbi.1014457 2026-06-26T14:00:00Z 2026-06-26T14:00:00Z <p>by Kexin Ma, Ningjing Wang, Zai Yang, Robert A. Cheke, Biao Tang</p> Adaptive cancer therapy seeks to modulate aggressive treatment to preserve drug-sensitive tumor cells that suppress resistant populations, but existing strategies often rely on frequent treatment decisions enabled by intensive surveillance, limiting clinical feasibility. Here, we propose a clinically motivated alternative that shortens the treatment window within a fixed and relatively long surveillance cycle, thereby avoiding the need for frequent monitoring. Based on this idea, we develop a mechanistic modeling framework for single-threshold-guided adaptive therapy with partial surveillance-cycle treatment (AT-PSC) and benchmark its performance using reinforcement learning. Using clinically calibrated parameters from an individual patient, simulations show that AT-PSC prolongs the time to progression (TTP) by 402 days compared with adaptive therapy using full surveillance-cycle treatment, while substantially reducing treatment exposure (dose reduced by 10.1%). Consequently, AT-PSC achieves significantly larger TTP gains than continuous therapy (1891 days) and two-threshold-guided adaptive therapy AT50 (1123 days). Simulations using data from six additional patients and sensitivity analyses further demonstrate that these benefits are robust across heterogeneous tumor growth profiles, while individual-based treatment should be considered to maximize TTP. Reinforcement learning yields comparable outcomes under the same fixed treatment window and can further extend TTP when the treatment window is adaptively adjusted. Together, these results support AT-PSC as a clinically feasible strategy to improve disease control while reducing treatment burden, and suggest that a practical regimen, such as a 14-day treatment window within a 30-day surveillance cycle, can provide sustained benefits for a broad patient population. Maarten J. Droste Robert Planqué Frank J. Bruggeman 10.1371/journal.pcbi.1014417 2026-06-26T14:00:00Z 2026-06-26T14:00:00Z <p>by Maarten J. Droste, Robert Planqué, Frank J. Bruggeman</p> Many heterotrophic microorganisms gradually replace an energetically-efficient mode of metabolism by an inefficient, more wasteful overflow metabolism above a critical growth rate, even though the energy demand continues to rise with growth rate. For instance, complete respiration of a sugar is replaced by its fermentation. In this paper, we aim to acquire a comprehensive overview of the behaviour of the metabolic fluxes and coarse-grained protein expression as function of the growth rate of the cell, by integrating previously proposed mechanisms and models into one framework. We derive the conditions for a metabolic shift to happen, by using and extending an existing core model of metabolism and growth that is qualitatively in agreement with experimental data. Assuming a fixed cellular protein content, the model shows that protein expression of efficient metabolism and anabolism rises as function of growth rate until a critical value is reached. This growth-associated protein expression is at the expense of proteins associated with future adaptation. At the critical growth rate, this preparatory-protein pool is reduced to zero. Beyond the critical growth rate, the anabolic protein pool and the energy demand continue to rise and therefore less protein remains for catabolism. In this regime, the inefficient metabolism gradually takes over ATP synthesis from the efficient mode. It can do so if it requires less protein per unit of ATP flux. We show that such a metabolic shift can occur only if the maximal growth rate of the inefficient mode is higher than the critical growth rate, and that this is equivalent to the second mode having a higher proteome efficiency than the first. Finally, we reduce a genome-scale model of protein expression in the yeast <i>Saccharomyces cerevisiae</i> to a variant of our core model and show that it is still qualitatively in agreement with the experimental data used to validate the original model. This study provides a synthesis that integrates and unifies existing models (coarse-grained and genome-scale models), that all aim to understand shifts in metabolic strategies.