An efficient framework for high-accuracy sorting of retired Li-ion batteries of electric vehicles

Risk-aware model predictive control for autonomous vehicle platoons under uncertain cut-in scenarios based on Gaussian mixture models

ACAT1-mediated lactylation reprogramming governs immune-stromal crosstalk in ulcerative colitis.

Exploring e-scooters as first- and last-mile with Gaussian mixture and machine-learning models

Mesh-agnostic prediction of unsteady flow dynamics using graph U-Nets

Early recurrence prediction model for DLBCL based on Gaussian mixture model bidirectional clustering resampling and improved deep forest

Mechanical and microstructural evolution of Ni-YSZ electrodes in solid oxide cells after long-term operation: insights from high-speed nanoindentation mapping and Gaussian mixture model clustering

Abstract Accurate prediction of the remaining useful life (RUL) of rolling bearings is crucial for ensuring the reliable operation of rotating machinery. However, existing studies often overlook the dynamic evolution of feature sensitivity throughout the complete degradation lifecycle, and insufficiently explore the deep coupling between local and global temporal features, which limits prediction accuracy. To address this, this paper proposes a hybrid prediction framework integrating bidirectional temporal convolution, convolutional block attention module, and bidirectional gated recurrent unit (BiConv-CBAM-BiGRU). First, a multi-dimensional weighted metric based on monotonicity, correlation, and robustness is constructed to screen key degradation features. Gaussian mixture model (GMM) and an adaptive criterion are employed to determine the first prediction time (FPT), resolving the ambiguity in the initial prediction moment. Second, a bidirectional convolutional structure is designed to deeply extract bidirectional contextual features from local waveforms. The CBAM mechanism is introduced to adaptively recalibrate feature weights, thereby enhancing key degradation information, while BiGRU is coupled to capture long-term nonlinear temporal dependencies. Experimental results show that the prediction accuracy of the proposed model on the PHM2012/XJTU dataset is improved by 2.06% and 15.65% compared with the mainstream model. This method demonstrates excellent generalization capability and robustness, providing a new technical pathway for high-precision predictive maintenance of industrial equipment.

Hierarchical structures play a key role in governing the electronic properties of peptides. Despite recent advances, establishing clear structure-property relationships that connect the solvent environment, molecular conformation, and electron transport at the single-molecule level remains challenging. Here, we use a combination of single-molecule experiments, molecular dynamics (MD) simulations, and machine learning (ML) analysis to understand how electron transport in peptides depends on solvent conditions for several different environments including water, 2,2,2-trifluoroethanol, acetonitrile, and glycerol. Our results reveal two distinct conductance populations for peptides in water or 2,2,2-trifluoroethanol: a high-conductance state associated with defined secondary structures (β turns or 310 helices) and a low-conductance state corresponding to extended primary structures. Peptides show a diminished high-conductance state in acetonitrile, which is known to weakly stabilize secondary structures and denature peptides. Interestingly, the high-conductance state is diminished in glycerol for tetrapeptides but not for pentapeptides. Unsupervised ML analysis using silhouette clustering and Gaussian mixture modeling suggests that solvent-dependent conductance behavior is mediated by peptide conformation. Complementary MD simulations, time-lagged independent component analysis of intramolecular hydrogen-bonding (H-bonding) distances, and Pearson correlation coefficients further reveal how solvent-peptide interactions and secondary structures govern electron transport pathways. Overall, our results show that the solvent environment significantly influences electron transport in peptides mediated by secondary structure and H-bonding interactions.

Accurately quantifying protein-protein binding at the single-molecule level is essential for understanding the mechanisms of viral infection and therapeutic targeting. Here, we use solid-state nanopores (SSNs) to detect complex formation between the SARS-CoV-2 Spike receptor-binding domain (Spike RBD) and the alternative host receptors KREMEN1 and Asialoglycoprotein Preceptor 1 (ASGR1). Single-molecule translocation events were analyzed using unsupervised Gaussian mixture modeling and a control-anchored semisupervised classification framework to resolve overlapping free-protein and complex populations. This approach enabled direct identification of receptor-Spike RBD complexes and calculation of apparent dissociation constants under experimental conditions. The inferred affinities were 261.1 nM for ASGR1 and 56.6 nM for KREMEN1, in good agreement with reported literature values and can be used as rough estimates on Spike and its receptor affinities. A negative control using human serum transferrin and Spike RBD showed no emergent high-ΔI population, supporting the specificity of the observed interactions. These results establish SSNs as a scalable and quantitative platform for single-molecule affinity measurements.

Acute heart failure (AHF) is commonly regarded as a worsening stage of chronic heart failure (CHF), yet the hemodynamic transition to decompensation remains incompletely defined. This study evaluated whether a noninvasive physiology-guided approach based on venous return (VR)-cardiac output (CO) balance can provide a framework to characterize transitional circulatory states preceding overt AHF. A total of 193 patients with CHF, hypertensive heart disease, or AHF were analyzed. Mean circulatory filling pressure and right atrial pressure were estimated noninvasively using inferior vena cava diameter indexed to body surface area (IVC/BSA) and extracellular-to-total body water ratio (ECW/TBW), respectively. VR-CO imbalance was quantified using a composite mismatch index. The distribution of VR-CO mismatch demonstrated a bimodal pattern on Gaussian mixture modeling, with two components (means ± SD: 2.25 ± 0.96 and 5.25 ± 2.61). All overt AHF cases were localized within the higher-mismatch component. Intermediate states exhibited increasing mismatch while maintaining relatively preserved forward flow, consistent with buffered circulatory states. A subset of patients exhibited marked hemodynamic mismatch without clinical congestion, representing a highly buffered pre-decompensated state. AHF may represent a failure of circulatory compensation in which venous-return loading exceeds the capacity of forward flow. A noninvasive VR-CO framework reveals a continuum from compensated to buffered to overtly decompensated states and identifies a previously unrecognized highly buffered state preceding clinical deterioration. This physiology-guided approach may enable earlier detection of circulatory instability and provide a quantitative framework for risk stratification and therapeutic monitoring.

This study aims to characterize differences during fine-motor execution between late-life depression (LLD) and healthy control (HC) older adults, and to evaluate the feasibility and robustness of classification models based on cluster-derived features. We enrolled 95 individuals in 2024 at The First People's Hospital of Yunnan Province (LLD, n=58; HC, n=37). Participants wore AX6 triaxial accelerometers on both wrists (100Hz) and completed four Purdue Pegboard subtests and the Clock Drawing Test. Preprocessing included three-axis standardization, computation of signal vector magnitude (SVM), sliding-window segmentation, principal component analysis, and Gaussian mixture model clustering. On scales, HAMD-17 was significantly higher in LLD than HC, and MMSE also differed. LLD performed worse than HC on all four Purdue Pegboard subtests with significant differences, while the raw Clock Drawing score was not significant. Task-level comparisons of "pattern composition" were significant across all five tasks (Purdue subtests p=0.001 each; Clock Drawing test p=0.010). Of 50 clusterwise comparisons, 31 showed significant between-group differences with replication across tasks. Classification performance was as follows: Lasso-logistic, accuracy 0.89 (95% CI 0.81-0.95), sensitivity 0.81, specificity 0.95, AUC 0.95; XGBoost, accuracy 0.89 (95% CI 0.81-0.95), sensitivity 0.91, specificity 0.89, AUC 0.94. Overall, LLD more often exhibited intermittent, higher-peak, rhythm-unstable kinematic patterns during task execution. Wrist-worn accelerometry combined with standardized fine-motor tasks can distill interpretable, reproducible "motor pattern fingerprints," objectively delineating differences between LLD and HC in rhythmic organization and motor stability.

MRI-based radiomics has shown potential for differentiating HER2-zero, HER2-low, and HER2-positive breast cancers; however, the added value of integrating intratumor heterogeneity (ITH) and peritumoral habitat features remains unclear. In this retrospective retrospective study, 515 patients from five hospitals were included and divided into training, internal testing, and external validation cohorts. Radiomic features were extracted from intratumor, peritumor, and habitat regions. Habitat subregions were generated using a Gaussian mixture model, and an ITH index was constructed. Three radiomics models (intratumor+peritumor, ITH, and peritumor + ITH) and a combined clinical-radiomics model were developed. Model performance was evaluated using receiver operating characteristic analysis and decision curve analysis. Among all models, the peritumor + ITH model showed comparatively better performance among the evaluated models. For Task 1 (HER2-positive vs. HER2-negative), the model yielded AUCs ranging from 0.713 to 0.791 across different cohorts. For Task 2 (HER2-low vs. HER2-zero), it achieved an AUC of 0.839 in the training cohort, 0.831 in the internal testing cohort, and 0.799 and 0.734 in the external validation cohorts, respectively. The combined model further improved predictive performance, achieving an AUC of 0.802 in external validation cohort 1 and 0.760 in cohort 2. Decision curve analysis demonstrated superior clinical utility of the combined model compared with other approaches. MRI-based habitat radiomics integrating intratumor heterogeneity and peritumoral features provides a promising non-invasive complementary tool for predicting HER2 status and supporting clinical decision-making.

Functional diversity is a fundamental aspect of community structure and composition, reflecting diversity and redundancy in ecological niches, functional roles, and environmental responses among species within a community. Despite its growing importance for quantifying ecosystem-level biodiversity, existing functional diversity metrics remain difficult to calculate and interpret, hindering their adoption and application beyond the scientific realm. One potential solution to this problem is to categorize species into functional groups based on their traits, which provides a simple, intuitive categorization of functional diversity that allows for the application of traditional species-based metrics. The functional-group approach, however, has several challenges that have limited its adoption, namely, the difficulty in identifying robust functional clusters, which can vary substantially due to trait variability, measurement error, and trait correlation. Here, to address these challenges, we present a multi-step consensus clustering method that integrates trait uncertainty and correlation into the classification of species into functional groups. Our approach proceeds in four main steps: (1) (re)sample trait data from an underlying distribution or with measurement error, (2) fit a Gaussian Mixture Model (to account for correlation) to each resample, (3) build a consensus matrix quantifying how often species pairs are grouped together across the noisy trait sample, and (4) apply traditional hierarchical clustering to this matrix and select the final groups. As a case study of this approach, we apply this method to a global dataset of 47,828 tree species using 18 traits, identifying 42 functional groups with distinct trait patterns and varying degrees of stability. We show how the resulting groups reflect underlying ecological trade-offs and phylogenetic structure, and we demonstrate how traditional diversity metrics (richness and Simpson's Index) can be applied to these functional groups to provide intuitive measures of functional group richness and functional redundancy. Collectively, this framework presents a scalable, interpretable approach for quantifying functional groups that embraces trait correlation and trait uncertainty, allowing for repeatable and intuitive quantification of functional biodiversity that can aid its adoption in biodiversity assessments by conservation and restoration organisations.

Objective.Adaptive deep brain stimulation (aDBS) for Parkinson's disease is a recently-approved therapy that adjusts stimulation in response to neurophysiologic biomarkers of motor-symptom state. Most real-time implementations of aDBS rely on instantaneous, noise-susceptible classifiers that apply simple thresholds to neurophysiologic biomarkers. We examined whether incorporating temporal history through Bayesian state-space modeling improved motor-state classification compared to instantaneous discriminant classifiers.Approach. We analyzed naturalistic neural data from three patients with Parkinson's disease chronically implanted with investigational sensing-enabled DBS systems, recording from both the subthalamic nucleus (STN) and sensorimotor cortex. Biomarkers were extracted across multiple window lengths and labeled using wearable-derived bradykinesia and dyskinesia scores. Classifier behavior was evaluated using two biomarkers (cortical stimulation-entrained gamma and STN beta oscillations) across a factorial combination of two conditions: (1) instantaneous discriminant analysis vs Bayesian time-history modeling via hidden Markov models (HMMs), and (2) single Gaussian vs Gaussian mixture modeling of each motor state's biomarker distribution. Performance metrics includedF1 scores, accuracy, prediction smoothness, latency, and computational load.Main Results. Using entrained-gamma biomarkers, incorporating time history via HMMs significantly improved hyperkinetic-state detection (F1: +12.9 ± 1.8%; accuracy: +30.0 ± 2.7%; bothpadj< 0.001) with modest decreases in hypokinetic-state performance, yielding a net increase in averageF1 (+4.7 ± 0.9%,p< 0.001). HMMs also yielded smoother and more accurate predictions for a given latency compared to simply increasing the window length used to extract neurophysiologic biomarkers. Entrained-gamma biomarkers outperformed STN beta biomarkers across all classifiers (averageF1: +12.9% ± 0.5%,p< 0.001). All methods operated within sub-millisecond prediction times and demonstrated sublinear empirical computational scaling.Significance. Bayesian time-history modeling enhanced motor-state classification while preserving the low latency and computational efficiency required for real-time aDBS. These findings, derived from chronic at-home recordings, support the translational potential of Bayesian state-space models for next-generation aDBS systems.

Intratumor heterogeneity (ITH) impacts cancer progression, and its characterization is crucial. Clustering algorithms applied to the variant allele frequency (VAF) of mutations can facilitate the exploratory analysis of ITH. This study comparatively evaluated six clustering algorithms to characterize ITH by clustering mutations based on their VAFs. We utilized data from The Cancer Genome Atlas to analyze three cancer types by examining the distribution of clusters in the results from various methods and four internal validation metrics. The results indicated that the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and Variational Bayesian Gaussian Mixture Model methods identified an insufficient number of clusters in most tumor samples. The Hierarchical DBSCAN (HDBSCAN) and Ordering Points to Identify the Clustering Structure (OPTICS) algorithms exhibited greater variability in the number of clusters, while Affinity Propagation (AP) showed controlled behavior, and Mean-Shift demonstrated greater consistency. The Mean-Shift and AP methods were consistently superior in the validation metrics, in contrast to HDBSCAN and OPTICS, which had inferior performance. We conclude that Mean-Shift and AP are promising and accessible alternatives for the initial exploratory analysis of ITH by VAFs. A computational pipeline is provided on the Google Colab platform to facilitate future studies.

Proteins fold through dynamic intermediates that dictate their routes to functional structures, with ensembles predominantly displaying heterogeneity across nanosecond-to-microsecond timescales. Directly observing these states in solution remains challenging as single-molecule methods often require technically demanding microfluidics, surface attachment that alters behavior, or denaturants that distort natural energy landscapes. Here we introduce NEXT-FRET, a solution-based single-molecule platform combining single-molecule FRET (smFRET) with time-varying Gaussian mixture modeling to resolve how diffusing proteins populate and interconvert between conformations under near-native conditions. By incorporating prior equilibrium information into time-dependent analysis, NEXT-FRET requires a few molecules per condition and accessible instrumentation, enabling application in the presence of chaperones and aggregation-prone precursors. We apply NEXT-FRET to the Escherichia coli Maltose-Binding Protein (MBP) and pre-MBP to reveal a long-sought closed on-pathway intermediate that exchanges with both native and unfolded states. The signal peptide raises the barrier selectively for the intermediate-to-native transition. Profiling interactions with chaperones shows that each stabilizes nonnative conformations distinctively, generating kinetic traps. These findings demonstrate that sequence features and proteostasis factors actively reshape the folding landscape. By following molecules out of equilibrium, NEXT-FRET reveals intermediates invisible at equilibrium. This reflects the inherent nonequilibrium character of cells, which maintain order through ongoing energy exchange and dissipation, with fluctuations governing the kinetics and connectivity of biomolecular states. By exposing transient intermediates and quantifying kinetic flows, NEXT-FRET offers a scalable strategy to interrogate nonequilibrium dynamics, providing mechanistic insights into protein (mis)folding, enzyme catalysis, ligand binding and broader biomolecular reactions with implications for biotechnology and therapeutics.

Label-free optical absorption microscopy techniques continue to evolve as promising tools for label-free histopathological imaging of cells and tissues. However, critical challenges relating to specificity and contrast, as compared to current gold-standard methods continue to hamper adoption. This work introduces Photon Absorption Remote Sensing (PARS), a new absorption microscope modality, which simultaneously captures the dominant de-excitation processes following an absorption event. In PARS, radiative (auto-fluorescence) and non-radiative (photothermal and photoacoustic) relaxation processes are collected simultaneously, providing enhanced specificity to a range of biomolecules. As an example, a multiwavelength PARS system featuring UV (266nm) and visible (532nm) excitation is applied to imaging human skin, and murine brain tissue samples. It is shown that PARS can directly characterize, differentiate, and unmix, clinically relevant biomolecules inside complex tissues samples using established statistical processing methods. Gaussian mixture models (GMM) are used to characterize clinically relevant biomolecules (e.g., white, and gray matter) based on their PARS signals, while non-negative least squares (NNLS) is applied to map the biomolecule abundance in murine brain tissues, without stained ground truth images or deep-learning methods. PARS unmixing and abundance estimates are directly validated and compared against chemically stained ground truth images, and deep learning based-image transforms. Overall, it is found that the PARS unique and rich contrast may provide comprehensive, and otherwise inaccessible, label-free characterization of molecular pathology, representing a new source of data to develop AI and machine learning methods for diagnostics and visualization.

Stunting remains a significant public health issue in Indonesia, particularly in Aceh Province, where considerable disparities continue to exist across districts and municipalities. Identifying regional prevalence patterns is crucial for developing evidence-based intervention strategies. This study assesses the performance of four unsupervised learning algorithms, namely K-Means, Hierarchical Clustering, Gaussian Mixture Model (GMM), and Fuzzy C-Means (FCM), for clustering district-level stunting data in Aceh Province across five observation periods. Algorithm performance was evaluated using the Calinski-Harabasz Index, convergence efficiency, and cluster interpretability. The findings demonstrate that Fuzzy C-Means outperformed the other methods, achieving the highest Calinski-Harabasz score of 49.75, followed by GMM with 42.61, Hierarchical Clustering with 36.48, and K-Means with 25.30. In addition, FCM showed the fastest convergence, requiring only three iterations. Three stable regional clusters were identified, representing high, moderate, and low prevalence levels. High-prevalence areas included Aceh Barat, Aceh Utara, Aceh Tenggara, Pidie Jaya, Aceh Barat Daya, Simeulue, and Bener Meriah, whereas Subulussalam constituted the low-prevalence cluster. These findings indicate that Fuzzy C-Means provides a reliable approach for regional stunting classification and may contribute to more targeted policy interventions in Aceh Province.

Exploring a Subpopulation of MASLD Associated With New Onset of CKD Using Supervised Clustering Techniques.