共找到 20 条结果
The visual system is attuned to the statistical regularities of the visual world, enabling rapid, almost reflexive, and highly accurate recognition of object identities and categories. This article presents recent advances across psychophysics, computational modeling, and cognitive neuroscience, which together suggest that the visual system is just as attuned to the laws of physics governing our physical world. We review psychophysical work showing that vision incorporates intuitive physics as a rapid, spontaneous, and stimulus-driven process. We review the computational framework of physics-based analysis by synthesis, which suggests that the mind's and brain's intuitive physics is in a structure-preserving relationship with entities and their force relations in the physical world. We end with an outline of an integrative program of computational modeling work, along with neuroscientific and psychophysical studies, toward psychologically and neurally refined mechanistic accounts of the visual perception of intuitive physics.
In silico techniques have become essential in contemporary parasitology, offering swift, cost-effective, and scalable methods for exploring parasite biology, host-parasite interactions, drug resistance, the discovery of diagnostic markers, vaccine development, and the prioritization of therapeutic targets. Computational frameworks that encompass genomics, pangenomics, phylogenetics, transcriptomics, proteomics, structural biology, molecular docking, molecular dynamics simulations, artificial intelligence (AI), and immunoinformatics have collectively revolutionized parasite research. They enable the systematic identification of conserved antigens, lineage-specific virulence factors, metabolic weaknesses, and potential therapeutics across the primary parasitic disease categories discussed in this review. Simultaneously, the growth of databases and analytical platforms focused on parasites has enhanced genome annotation, inhibitor discovery, epitope prediction, and systems-level analysis. However, despite these advancements, numerous workflows rely on incomplete or inadequately annotated datasets and biologically oversimplified assumptions that fail to accurately represent the complexity of parasites, variations in their life cycles, and host-dependent factors-further complicated by inconsistent data quality and diminished reproducibility across species. As a result, computational findings necessitate thorough integration with transcriptomic, proteomic, structural, and functional experimental data. When utilized in this collaborative manner, in silico methods expedite hypothesis generation, refine experimental parameters, and bolster rational approaches for drug discovery, vaccine design, and epidemiological surveillance.
Antibody-drug conjugate (ADC) payload discovery remains constrained by reliance on traditional molecular descriptors that inadequately capture three-dimensional geometric features governing target recognition and mechanism of action. Topological data analysis (TDA) offers a mathematical framework for characterizing molecular shape through persistent homology, potentially revealing mechanistic relationships invisible to conventional approaches. We developed a comprehensive TDA framework analyzing 22 FDA-approved ADC payloads across 1,471 clinical trial records, computing 31 topological descriptors encompassing Betti numbers, persistence statistics, and complexity metrics. Hierarchical clustering, principal component analysis, and correlation network analysis were employed for dimensionality reduction and cluster validation, with molecular docking studies validating TDA-derived classifications. TDA-based clustering identified eight distinct payload classes with excellent separation. Principal components captured 79.8% of topological variance, with Betti numbers and persistence lifetime as dominant features. Three major mechanistic clusters emerged: vinca alkaloids (tubulin inhibitors), camptothecins (topoisomerase I poisons), and DNA alkylators. Molecular docking demonstrated high performance within-cluster binding consistency and significant cross-cluster discrimination. We establish the first validated TDA framework for ADC payload discovery, demonstrating that persistent homology captures biologically meaningful mechanistic classifications suitable for rational payload design and mechanism-of-action prediction in precision oncology.
Collecting gold-standard phenotype data via manual extraction is typically labor-intensive and slow, whereas automated computational phenotypes (ACPs) offer a systematic and much faster alternative. However, simply replacing the gold-standard with ACPs, without acknowledging their differences, could lead to biased results and misleading conclusions. Motivated by the complexity of incorporating ACPs while maintaining the validity of downstream analyses, in this paper, we consider a semi-supervised learning setting that consists of both labeled data (with gold-standard) and unlabeled data (without gold-standard), under the covariate shift framework. We develop doubly robust and semiparametrically efficient estimators that leverage ACPs for general target parameters in the unlabeled and combined populations. In addition, we carefully analyze the efficiency gains achieved by incorporating ACPs, comparing scenarios with and without their inclusion. Notably, we identify that ACPs for the unlabeled data, instead of for the labeled data, drive the enhanced efficiency gains. To validate our theoretical findings, we conduct comprehensive synthetic experiments and apply our method to multiple real-world datasets, confirming the practical advantages of our approach.
Robust anomaly and out-of-distribution (OOD) detection in radiology demands learning methods that are accurate, interpretable, computationally efficient, and reliable under real-world distributional shifts. Existing back-propagation-trained models often struggle to meet these requirements simultaneously, while forward-forward learning, despite its conceptual appeal as a resource-efficient and biologically plausible alternative, has so far seen limited adoption in image-based and safety-critical medical applications due to scalability and generalisation limitations. We revisit back-propagation-free learning for the open-world clinical setting and discuss the Convolutional Forward-Forward Algorithm (CFFA), a parameter-efficient reformulation of the Forward-Forward Algorithm tailored to high-dimensional medical image analysis. CFFA incorporates convolutional structure and layer-wise local objectives, overcoming key scalability and generalisation limitations of existing forward-forward approaches while retaining their resource-efficient training paradigm. Building on the observation that the forward-forward objective yields intrinsic and interpretable goodness statistics that directly quantify conformity to the learned data distribution, we introduce SaFF-AD, a self-adaptive forward-forward network explicitly designed for anomaly and OOD detection. SaFF-AD autonomously configures optimisation dynamics, architectural depth, and goodness normalisation, enabling stable learning under constrained computational budgets and in one-shot training regimes. Extensive experiments across multiple medical imaging benchmarks demonstrate that SaFF-AD achieves competitive or superior anomaly detection performance compared to back-propagation-trained models, while requiring substantially fewer parameters and forward evaluations. The forward-forward goodness signal enables self-supervised anomaly and OOD detection without auxiliary networks, post-hoc uncertainty estimation, or heuristically designed scoring functions. These results establish forward-forward learning as a viable and practically attractive alternative to conventional deep learning for safety-critical medical image analysis, particularly in settings characterised by constrained computational budgets, limited labelled data, and distributional uncertainty. By treating anomaly detection as an intrinsic property of the learned model rather than a post-hoc addition, SaFF-AD offers a unified framework that is interpretable, efficient, and well-suited to the open-world conditions encountered in real-world clinical deployment.
To study the potential pathogenic mechanisms of bisphenol A (BPA) in polycystic ovary syndrome (PCOS) using an integrative computational strategy. Integrative computational study combining network toxicology, Mendelian randomization (MR), and molecular docking. For MR analysis, genetic data were sourced from large European-ancestry cohorts, including plasma protein quantitative trait loci data and genome-wide association study summary statistics for PCOS (3,045 cases and 267,780 controls). In silico exposure to BPA for target prediction; genetically predicted plasma protein levels for causal inference. Identification of overlapping targets between BPA and PCOS; functional enrichment pathways; causal effects of prioritized proteins on PCOS risk (odds ratios with 95% confidence intervals); binding affinities between BPA and core targets (kcal/mol). Network toxicology identified 310 overlapping targets between BPA and PCOS. Enrichment analyses revealed significant involvement in endocrine signaling, inflammatory pathways (e.g., IL-17), and cellular processes. MR demonstrated that genetically elevated levels of RET, CXCL8, HTR6, MMP1, MMP9, NTRK1, and TNNI2 were significantly associated with increased PCOS risk, whereas higher PSAP and SHBG levels were protective. Molecular docking confirmed stable binding between BPA and all nine key targets, with strongest affinity for SHBG (-8.4 kcal/mol), followed by NTRK1, TNNI2, and RET. This integrative investigation suggests that BPA may contribute to PCOS pathogenesis through multi-target interactions involving inflammatory mediators, endocrine regulators, and tissue remodeling proteins. The findings provide prioritized targets and mechanistic insights for future experimental validation and environmental risk assessment.
Comparing graphs for structural similarity is one of the most important problems in graph analytics. However, due to the nonlinear nature of graphs, this problem is not straightforward to solve. Most existing graph comparison methods either lack expressiveness, do not provide interpretable measures of similarity, or incur high computational costs, limiting their applicability to large graphs. In this article, we propose novel graph kernels based on quantum Rényi $\alpha $ -entropies of different orders, computed from both the unnormalized and normalized Laplacian matrices. We investigate the properties of these entropies and show that they are determined by the frequencies and degree statistics of substructures of different types and sizes, such as simple paths and cycles ofdifferentlengths. By utilizing quantum Rényi $\alpha $ -entropies of different orders, our approach defines efficient, theoretically grounded, and interpretable graph kernels capable of characterizing the structure of unlabeled graphs. Through extensive experiments on benchmark datasets, we demonstrate that our methods achieve competitive or superior performance compared with state-of-the-art techniques, including deep learning approaches, while remaining computationally efficient.
Population genomic workflows frequently rely on fragmented command-line utilities, custom conversion scripts, and programming language-specific environments, complicating computational reproducibility and obscuring data provenance. As analytical workflows become increasingly automated and computationally intensive, dependence on disparate preprocessing tools can introduce friction between raw genotype files, quality-control decisions, statistical analyses, and downstream workflows. We developed SNPio, a Python-native framework that consolidates single nucleotide polymorphism data parsing, filtering, visualization, numerical genotype encoding, and population genomic summary-statistic calculation within a unified software architecture. VCF file parsing and filtering benchmarks were compared against vcfR and SNPfiltR. SNPio demonstrated faster execution times but used more memory than its R-based comparators, reflecting SNPio's retention of genotype arrays, metadata, and provenance-tracking attributes. Pairwise Weir and Cockerham's FST and Nei's genetic distance estimates aligned with HierFstat expectations based on Pearson correlations and aggregate error metrics. D-statistics conformed to theoretical expectations across eleven simulated datasets spanning a range of introgression signal strengths. SNPio provides a reproducible Python-native workflow for processing, filtering, encoding, visualizing, and analyzing SNP datasets. It integrates common early-stage population genomic operations into a transparent, scriptable framework, which ultimately promotes workflow provenance and reduces reliance on disjointed software tools, unsaved terminal commands, and custom scripts. SNPio is particularly suited for population genomic studies of non-model organisms in ecological, evolutionary, and conservation contexts, where reproducible preprocessing and interoperability with downstream analyses are becoming increasingly important.
In many neural populations, the computationally relevant signals are posited to be a set of "latent factors"-signals shared across many individual neurons. A given brain area may perform many computations, each associated with distinct factors, which can together compose an overall action. The methods for uncovering such structure typically require supervision, which can limit the discovery of novel aspects of activity. Here, we introduce sparse component analysis (SCA), an unsupervised approach. SCA facilitated surprisingly clear parcellations of neural activity across a range of behaviors, when seeking both linear and nonlinear embeddings. We applied SCA to motor cortex activity from reaching and cycling monkeys, single-trial imaging data from C. elegans, and activity from a multitask artificial network. SCA revealed both simple and unexpected instances where the overall population response was built compositionally from sets of factors with distinct computational roles.
Norm, the formal theoretical linguist, and Claudette, the computational language scientist, have a lovely time discussing whether modern language models can inform important questions in the language sciences. Just as they are about to part ways until they meet again, 25 of their closest friends show up - from linguistics, neuroscience, cognitive science, psychology, philosophy, and computer science. We use this discussion to highlight what we see as some common underlying issues: the String Statistics Strawman (the mistaken idea that LMs can't be linguistically competent or interesting because they, like their Markov model predecessors, are statistical models that learn from strings) and the As Good As It Gets Assumption (the idea that LM research as it stands in 2026 is the limit of what it can tell us about linguistics). We clarify the role of LM-based work for scientific insights into human language and advocate for a more expansive research program for the language sciences in the AI age, one that takes on the commentators' concerns in order to produce a better and more robust science of both human language and of LMs.
Continuous monitoring of the Amazon biome demands land cover classification models that are both highly sensitive and computationally feasible. To resolve the inherent trade-off between architectural complexity and predictive performance in spatial deep learning, this study introduces the Vision Transformer-Graph Neural Network with Feature Adaptation (ViT-GNN RFFA). In contrast to conventional end-to-end pixel models, this hybrid architecture operates exclusively on an 11-dimensional vector of extracted color-based vegetation indices (e.g., GRVI) and textural statistics. The dual-branch design isolates global sequence context via the ViT module while leveraging the GNN branch as a structural prior to learn non-linear covariance between specific features. Evaluated against a suite of benchmarks including MiniViT, Baseline CNN, Random Forest, XGBoost, and LightGBM, the proposed algorithm attained the highest Overall Accuracy of 0.930 utilizing merely 16,323 trainable parameters-a nearly 75% reduction in footprint versus pixel-based models. Crucially, a McNemar's statistical test confirmed that the accuracy gain over the strongest classical baseline (XGBoost, OA: 0.927) is statistically significant (p < 0.05). By pairing rigorous spatial cross-validation with an interpretable feature space, this work establishes that intelligent data pre-processing combined with graph-based relational learning offers a robust framework for high-precision environmental mapping under severe resource limitations.
Statistical review is essential for research quality and integrity, yet traditional manual review is inefficient. Large language models (LLMs) offer potential support but are unreliable when used without guidance for precise calculations and raise concerns about accountability. This study evaluated whether a structured, rule-based prompt can reliably constrain an LLM to perform statistical review of comparative categorical data, and characterized both its feasibility and its inherent risks from an accountability perspective. This study employed a two-stage design based on the DeepSeekV3.2. In the first stage, a structured prompt was developed through dozens of "test-fail-iterate" cycles using 20 published medical articles. The prompt assigned the LLM the role of a "statistics expert" and provided a closed set of computational rules and a "recognize data-select calculation formula-calculate" workflow for analyzing categorical data, including Pearson's Chi-square test, continuity correction, and McNemar's tests. In the second stage, the performance of the final prompt was evaluated on a test set of 20 independent manuscripts. The model's output was compared against the results calculated by a senior statistician (the gold standard). The primary outcome measures were the performance in statistical method selection and numerical computation, including accuracy, sensitivity (recall), specificity, positive predictive value, negative predictive value, F1 score, and Cohen's Kappa. Secondary measures included reproducibility and efficiency. The test set consisted of 15 manuscripts with independent samples and 5 with paired samples. In the assessment of the appropriateness of statistical method selection for 148 analysis items, the model achieved an accuracy of 99.3% (147/148), a sensitivity of 96.2% (25/26) (F1=98.0%, κ=0.976). For the test of computational consistency in 97 independent sample tests, the accuracy for χ2 value consistency was 94.8% (92/97) (F1=89.3%, κ=0.859), and for P-value consistency, it was 96.9% (94/97) (F1=90.9%, κ=0.891). In the paired-sample analysis, the model's methods and results were in perfect agreement with the manual review, and prompt optimization eliminated discrepancies in degrees-of-freedom calculation rules. Efficiency analysis showed no statistically significant difference in time consumption between the model (407 s) and manual review (374 s) (P=0.601). In reproducibility tests, the intraclass correlation coefficients for both χ2 values and P-values exceeded 0.91. However, qualitative analysis revealed 3 typical failure modes in the task workflow: (1) Instability: The model's failure to produce identical outputs across repeated runs, manifesting as inconsistent data extraction or the failure to process all designated tasks (scope neglect). (2) Performance degradation/"lazy" behavior: A decline in execution quality on long or complex tasks, often characterized by the model abandoning its reasoning process to copy author-provided values without verification. (3) Anchoring effect: The model's tendency to over-rely on author-provided statistical values (the "anchor"), causing its verification process to be unduly influenced. A structured, rule-based prompt can guide the DeepSeek to achieve high accuracy in standardized statistical review tasks, but its reliability is contingent on operational stability. Inherent failure modes, including performance instability and a strong anchoring effect on author-provided data, persist and can lead to significant errors, particularly when source data are flawed. These findings suggest that the the DeepSeek is not suitable for autonomous auditing. Their most appropriate application is as assistive tools within a human-in-the-loop framework, where rigorous human supervision is essential for risk mitigation and to maintain ultimate accountability.
A recent study by Thomas et al. (2022) showed that 8.5- to 10-month-olds use saliva sharing as an indicator of relationship closeness: infants expect that someone who previously shared saliva with a partner will also comfort that partner during times of distress. The current paper offers a mechanistic explanation for this finding. We propose that through everyday experiences, saliva sharing and comforting become integrated into an enriched representation that contains information about both events. Across five simulations, we show that when this proposal is implemented in a connectionist computational model, the model reproduces infants' looking behaviors in the study by Thomas et al. (2022) (Simulation 1), captures the finding that infants' responses were specific to comfort (Simulation 2), predicts bidirectional expectations about saliva sharing and comforting (Simulation 3), and shows that these expectations strengthen with experience and can shift in response to changes in environmental statistics (Simulations 4 and 5). Taken together, our findings suggest that domain-general associative learning can explain how infants come to understand the social meaning of saliva sharing. How do infants know who will help when someone is upset? Recent research shows that infants expect people who share saliva to comfort one another. The current study uses computer simulations to explain how infants develop this understanding. We propose that infants learn these social rules through everyday observations. By seeing people share saliva and provide support, infants learn to connect these two actions. Our simulations successfully explained infant behavior and showed that these expectations not only grow stronger with more experience but can also be reversed in extreme cases. This suggests that infants use general learning abilities to build an understanding of close relationships based on the patterns they observe in their daily lives.
Computational methods are central to the life sciences. The rapid growth and diversification of software tools and databases make it difficult to find, compare, and reuse methods for a given task. bio.tools is a community-driven registry designed to improve the visibility of research software and allow researchers to simplify access to the software ecosystem through structured, interoperable, and accessible metadata. Tools are annotated using the EDAM ontology and additional controlled vocabularies, enabling users to search and filter by scientific topics, operations, input/output data types, and data formats. bio.tools supports interactive exploration via rich tool landing pages and provides programmatic access through a documented API for search, retrieval, and registry statistics. The registry has expanded to almost 33,000 annotated tools through the combined contributions of thousands of community members and semi-automated literature mining that keep the registry up to date. Recent improvements to the registry include machine-assisted scoring to prioritise curator review, and consolidation of both its standards stack and software architecture. bio.tools has also become a foundational upstream metadata source that is reused by other services in the ELIXIR Research Software Ecosystem and beyond, to support synchronisation, cross-linking, and additional downstream services. bio.tools is freely available at https://bio.tools.
The problem of Human Action Recognition (HAR) continues to be difficult with the intricate temporal interactions, superfluous frames, and minor visual variations that usually define similar actions. A large number of current approaches are based on either transformer or multimodal architectures, which are computationally costly and cannot be used in real-time or resource constrained systems. To overcome these shortcomings, we proposed a novel and lightweight HAR model that integrates spatial, temporal and consistency modeling in an efficient architecture. In our model, we combine EfficientNet-B2, used in efficient spatial feature extraction, with Unitary Temporal Encoder (UTE), to train long-range temporal dependencies, and Adaptive Temporal Consistency Module (ATCM), to improve local temporal consistency. The proposed system is trained and tested on the datasets UCF101 and HMDB51 with a Top-1 accuracy of 97.10% and 87% under RGB input only which is competitive among RGB-based HAR methods while maintaining low computational cost. The 9.3 million parameter model with an inference rate of 9.5 ms per 16-frame video clip is highly accurate and efficient, thus suitable in inference in real-time and at edges. Ablation studies further demonstrate that the proposed components contribute consistent performance improvements across the evaluated benchmarks.
Liver fibrosis is a critical stage in the progression of chronic liver disease to cirrhosis, and effective antifibrotic targets remain lacking. This study aims to investigate the association between protein tyrosine kinase 7 (PTK7) and liver fibrosis through multi-omics analyses, explore its potential role in hepatic stellate cell activation, and evaluate the value of circulating PTK7 as a candidate biomarker for assessing liver fibrosis severity. Genome-wide association study (GWAS) summary statistics, cis-expression quantitative trait locus (cis-eQTL), and plasma protein quantitative trait locus (pQTL) datasets were integrated to identify candidate genes associated with liver fibrosis. Candidate genes were further validated using multiple machine-learning models in three independent Gene Expression Omnibus (GEO) cohorts (GSE84044, GSE25097, and GSE49541). Human and murine liver single-cell transcriptomic datasets were analyzed to characterize the expression profile of PTK7. In an exploratory clinical cohort, plasma PTK7 levels were measured by enzyme-linked immunosorbent assay (ELISA), and PTK7 expression was evaluated in a carbon tetrachloride (CCl4)-induced mouse model of liver fibrosis. In LX-2 cells, PTK7 was silenced to assess changes in β-catenin and fibrosis-related proteins, followed by intervention with SKL2001. Molecular docking and molecular dynamics simulations were performed to evaluate the potential interaction between PTK7 and β-catenin. A total of 18 candidate genes were identified, among which PTK7 consistently showed upregulated expression in fibrotic liver tissues across multiple datasets. Single-cell transcriptomic analysis revealed enriched PTK7 expression in Kupffer cells and fibroblast-related populations. Plasma PTK7 levels were significantly elevated in patients with liver fibrosis and were positively correlated with liver stiffness measurements. In activated LX-2 cells, PTK7 knockdown reduced the expression levels of collagen type I alpha 1 chain (COL1A1), alpha-smooth muscle actin (α-SMA), and β-catenin. Furthermore, rescue experiments using SKL2001 supported the involvement of β-catenin-related signaling pathways in PTK7-mediated profibrotic effects. Computational analyses suggested a stable interaction pattern between PTK7 and β-catenin. PTK7 is closely associated with liver fibrosis at both multi-omics and transcriptomic levels, may participate in hepatic stellate cell activation, and may be associated with β-catenin-related signaling pathways. Circulating PTK7 may serve as a candidate biomarker for reflecting the severity of liver fibrosis; however, further validation in larger clinical cohorts is warranted. 目的: 肝纤维化是慢性肝病进展至肝硬化的关键环节，目前缺乏有效的抗纤维化靶点。本研究旨在通过多组学探讨蛋白酪氨酸激酶7(protein tyrosine kinase 7，PTK7)与肝纤维化的关联，探索其在肝星状细胞活化中的潜在作用，并评估循环PTK7作为肝纤维化严重程度候选指标的价值。方法: 整合全基因组关联研究(genome-wide association study，GWAS)汇总统计数据、顺式表达数量性状位点(cis-expression quantitative trait locus，cis-eQTL)和血浆蛋白数量性状位点(protein quantitative trait locus，pQTL)数据，筛选肝纤维化相关候选基因;并在3个独立基因表达综合(Gene Expression Omnibus，GEO)队列(GSE84044、GSE25097和GSE49541)中结合多种机器学习模型进行验证。利用人和小鼠肝单细胞转录组数据分析PTK7的表达特征;在临床探索性队列中采用酶联免疫吸附实验(enzyme-linked immunosorbent assay，ELISA)检测血浆PTK7水平，并在四氯化碳(carbon tetrachloride，CCl4)诱导的小鼠肝纤维化模型中评估PTK7表达。在LX-2细胞中，通过敲低PTK7观察β-catenin及纤维化相关蛋白质的变化，并进一步采用SKL2001进行干预。采用分子对接和分子动力学模拟评估PTK7与β-catenin的潜在相互作用。结果: 共筛选出18个候选基因，其中PTK7在多个数据集中均表现为在纤维化肝组织中持续上调。单细胞转录组分析提示PTK7在Kupffer细胞和成纤维细胞相关群体中富集表达。肝纤维化患者血浆PTK7水平显著升高，且与肝硬度值呈正相关。在活化的LX-2细胞中，敲低PTK7可降低I型胶原蛋白α1链(collagen type I alpha 1 chain，COL1A1)、α-平滑肌肌动蛋白(alpha‑smooth muscle actin，α-SMA)和β‑连环蛋白(β-catenin)的表达水平，而SKL2001挽救实验支持β-catenin相关信号通路参与PTK7介导的促纤维化过程。计算分析提示PTK7与β-catenin之间存在稳定的相互作用模式。结论: PTK7在多组学和转录组数据层面与肝纤维化密切相关，且可能参与肝星状细胞活化过程，并可能与β-catenin相关信号通路有关。循环PTK7有望作为反映肝纤维化严重程度的候选生物标志物，但仍需在更大样本临床队列中进一步验证。.
Single-cell RNA sequencing methods based on split-pool combinatorial barcoding enable high-throughput profiling, yet sample identity is often encoded during early barcoding steps rather than through the library index. Consequently, reads from multiple biological samples remain pooled, complicating per-sample analysis and selective extraction of samples of interest. Here, I present CapMux, a Snakemake-based pipeline for processing split-pool scRNA-seq data from raw sequencing files to sample-resolved outputs. CapMux supports workflows starting from either BCL files or FASTQ files and reconstructs sample identity by integrating sub-library index information with the experiment-specific barcoding plate layout. The pipeline was developed for the CapSeq method but is configurable for related scRNA-seq combinatorial barcoding designs through specification of barcode positions and experimental layout. In a controlled cell line mixing scRNA-seq experiment, CapMux resolved pooled data into outputs for each sample, enabling independent quality control summaries, mapping statistics, count matrices, and downstream visualizations. Runtime benchmarking indicated that secondary demultiplexing step added only a modest computational overhead. Together, these results show that CapMux provides a practical and adaptable framework for recovering sample-level resolution from split-pool scRNA-seq data.
Putamen dopamine depletion characterizes Parkinson disease (PD). Intraneuronal processes determining dopamine stores have not been systematically examined. This study explored relative contributions of dopamine synthesis, storage, and metabolism to control-PD differences. We updated an intramural tabulation from 2002 to 2024 of postmortem putamen tissue contents of reactants, including the autotoxic dopamine metabolite 3,4-dihydroxyphenylacetaldehyde (DOPAL), from patients with PD and controls. Based on computational models applying first-order kinetics and equilibrium equations, we then compared estimated rates of dopamine synthesis through tyrosine hydroxylase (TH), L-aromatic-amino-acid decarboxylase, vesicular active uptake and passive leakage, exocytotic release and reuptake, and other intraneuronal processes. Results from the modeling were compared with those from in vivo 18F-DOPA PET. Postmortem data were analyzed from 13 patients with PD (median age 77 years, range 73-85 years) and 20 controls (median age 77 years, range 35-91 years). There was approximately a 98% decrease in putamen tissue dopamine in PD, and the concentration ratio of DOPAL/dopamine (DA) was approximately 9 times that of control. Applying the simplest kinetic model, vesicular sequestration was estimated to be decreased by 98.5% (0.073 vs 4.91 nmol/minute). Approximately 3-fold greater in vivo "washout" of putamen 18F-DOPA-derived radioactivity compared with controls also indicated attenuated vesicular storage in PD. According to the complete model, control-PD differences in intraneuronal reaction rates were in descending order of vesicular uptake ≈ vesicular leakage > exocytotic release ≈ neuronal reuptake > L-aromatic-amino-acid decarboxylase activity ≈ TH activity > other reactions. Convergent quantitative evidence points to a substantial vesicular storage defect in residual dopaminergic terminals in PD. This finding challenges the sufficiency of nigrostriatal dopaminergic denervation alone to account for the biochemical phenotype of PD and highlights vesicular dopamine handling as a critical determinant of putamen dopamine deficiency. The reaction rate estimates were drawn from published point values rather than fitted to an experimental data set, and so conventional goodness-of-fit regression statistics were not conducted. Because of the assumption of steady-state conditions for calculating reaction rates based on equilibrium equations, the model does not address the dynamics of disease pathogenesis over years but does provide a platform for further extension to disease progression. We estimated rates of reactions involved with the synthesis, storage, release, reuptake, and metabolism of dopamine in the putamen in PD and found that the main intraneuronal functional abnormality separating PD from controls was attenuated vesicular sequestration, implicating decreased vesicular uptake through the vesicular monoamine transporter and increased vesicular leakiness as key determinants of putamen dopamine deficiency in PD.
Trajectory reconstruction in high-dimensional spatiotemporal chaos is often difficult due to the coupled challenges of structural model misspecification and prohibitive computational costs. We introduce in this work a Low-Rank Extended Rauch-Tung-Striebel Smoother (LR-ExRTSS) for use within the Statistical Finite Element (StatFEM) framework, combining a square-root retrospective smoother with a forward looking low-rank extended Kalman filter. We apply this grid-independent, retrospective scheme to the 2D anisotropic Kuramoto-Sivashinsky equation under a severe dynamical mismatch. We force a predictive model converging to a stable 1D steady state to track a truly chaotic 2D system using sparse, noisy observations. Results demonstrate that the LR-ExRTSS acts as a bridge between misspecified physics and chaotic reality by reintroducing transverse instabilities aggressively damped by the flawed model. Spectral analysis confirms that the efficacy of the scheme is based on its ability to span the unstable-neutral subspace of the true system, regardless of the subspace predicted by the model. This work establishes low-rank StatFEM as a computationally feasible framework for robust state estimation in high-dimensional systems under severe structural error.
Root phenotype is a key agronomic trait affecting maize growth and development. In situ observation and high-precision root phenotypic analysis provide important support for monitoring maize growth. Traditional root phenotyping methods lack in situ monitoring capabilities, and existing models have limited accuracy in root segmentation. To address these issues, we developed a crop root phenotyping system integrating crop cultivation and data collection. We also proposed a DB-UNet model for hydroponic maize root segmentation. DB-UNet builds a CNN-ViT dual-branch parallel structure during encoder downsampling level. The lightweight ViT branch uses sequential downsampling to achieve global topological dependency modeling while reducing computational costs. An attention fusion module dynamically calibrate dual-branch features weights, achieving complementary fusion of local root edge details and global context information. we constructed a mixed loss function combining Dice loss, Focal loss, and structural consistency KL loss to solve class imbalance, hard sample segmentation, and semantic divergence of dual-branch features. On our custom hydroponic maize root dataset, DB-UNet achieved an mIoU of 91.02%, an FG IoU of 82.78%, and a Centerline-Dice of 97.72%.Compared to classic UNet, mIoU, FG IoU, and Centerline-Dice increased by 0.92%, 1.84%, and 1.99%, respectively. Plant-level five-fold cross-validation further showed that DB-UNet maintained stable segmentation performance across different plant-level partitions. Based on DB-UNet segmentation results, we propose a custom skeleton-based algorithm for multi-trait root phenotyping, enabling the extraction of total root length and root branch points. Root area is calculated from binary mask pixel statistics. Compared to the traditional Zhang-Suen algorithm, the average relative error of root length measurement is reduced to 3.14%, which is 8.42 percentage points lower than the traditional method. Furthermore, we analyzed relationships between segmentation accuracy metrics and phenotypic relative errors. Higher segmentation quality generally led to lower phenotypic relative errors and more reliable trait measurements. In particular, Centerline-Dice was closely associated with root length estimation, whereas pixel-level segmentation consistency was more closely related to root area measurement. Pearson and Spearman correlation analyses showed a strong positive correlation between maize plant height and total root length, with coefficients of 0.8466 and 0.8634, respectively.