Childhood dyslexia risk elevated by heavy metal mixtures from e-waste: A machine learning-driven mixture modeling study.

dpGMM: A new R package for efficient and robust Gaussian mixture modeling of 1D and 2D data

Suicidal ideation trajectories among adults following psychiatric hospitalization.

Predicting progression-free survival in hormone-receptor positive (HR+/HER2-) metastatic breast cancer (MBC) treated with CDK4/6 inhibitors: A machine learning approach.

Gender-specific developmental trajectories of anxiety and depression among college students: Risk of attitudes toward suicide and suicidal behavior.

On the duality for real mixture models

With the rapid growth of freight transport in China, ensuring large truck safety on freeways is increasingly critical. However, existing studies focus on isolated behaviours or small datasets, lacking scalable methods to identify high-risk groups in real traffic flows. This study constructs 13 behavioural profiles for large trucks based on China's legal framework, covering speed patterns, load status, travel time and vehicle attributes. A Sample-Weighted Gaussian Mixture Model with Tsallis Entropy (SWGMM-T) is proposed to enhance clustering robustness and interpretability. Compared with four baseline models, SWGMM-T outperformed in clustering performance. Results show that cumulative overspeed travel time, overspeed distance, overspeed rates, abnormal stop rates and cumulative abnormal stop time are crucial in risk assessment. Crash rates per 10,000 trucks confirm this trend. Cluster 3 shows the highest crash rate and severe crash involvement, while Cluster 2 has the lowest crash rate and minimal severity. These findings suggest risks arise from combinations of risky behaviours.

Scalable nonparametric clustering with unified marker gene selection for single-cell RNA-seq data.

Chronic obstructive pulmonary disease (COPD) imposes substantial clinical and economic burdens. Early detection can allow for monitoring and timely treatment to slow its progression. Tracking of patient trajectories prior to their diagnosis can inform the timing and targeting of interventions. We aimed to identify pre-diagnosis healthcare utilisation patterns in a COPD cohort and profile the associated subgroups. We conducted a retrospective cohort study of patients with a new inpatient or specialist outpatient clinic (SOC) diagnosis of COPD from 2018 to 2019 in a regional health system in Singapore. Their healthcare utilisation, expenditure and diagnoses from the 3 years prior to diagnosis were extracted. Patients were classified into subgroups with different expenditure and utilisation patterns using Bayesian mixture modelling and compared against a propensity score-matched non-respiratory control group. 1171 patients with COPD were matched to a control and classified as either chronic (n=688) or transient healthcare users (n=483) prior to a first COPD diagnosis. Chronic users had increasing utilisation over time across all settings, recording multiple SOC visits (median, 25th-75th percentile: 10, 5-20). Transient users had low utilisation throughout, reporting fewer SOC visits than controls (transient: 0, 0-1, control: 2, 0-9). The prevalence of hypertension or hyperlipidaemia was >50% in chronic users, >30% in controls and <30% in transient users. Early detection strategies should focus on case-finding among patients with known risk factors in SOCs for referral to respiratory care and outreach to socially disadvantaged communities to facilitate timely access to healthcare.

We investigate rejection probabilities of statistical tests based on resampling procedures. Our general framework under consideration covers, in particular, bootstrap and permutation techniques. It turns out that specific properties of the p-value distribution play a key role, namely convexity or concavity, the Bernstein property and those of beta mixture models. We provide a detailed analysis and clarify how these properties relate to each other. We derive new bounds for the rejection probability. The results link the number of replications with size and power of the test. Numerical considerations demonstrate the quality of the bounds. An important application is the nested simulation estimator in Monte Carlo simulation studies. Our findings indicate that a moderate or even rather small number of replications is sufficient to obtain useful simulation results. This enables a substantial reduction of the computational effort in Monte Carlo simulation studies.

Design of an optimal planning framework for cryosurgical treatment of brain tumor using CNN segmentation of MRI images.

The neuropeptides oxytocin and vasotocin are predominantly produced in the supraoptic and paraventricular nuclei of the anterior-inferior, anterior-superior and tubular-superior hypothalamic subunits. Evidence suggests that oxytocin and vasotocin signaling play a role in both physiology and behavior, and that dysfunction of these signaling systems may contribute to the co-occurrence of metabolic and psychiatric conditions. The genetic pathways, however, that may underlie the connection between these physiological and behavioral traits are yet to be clearly delineated. We deployed bivariate mixture models and conjunctional FDR to estimate the global and local genetic overlap between three oxytocinergic-vasotocinergic hypothalamus subunits and ten psychiatric and metabolic traits related to oxytocin and vasotocin signaling. We show that these three subunits share moderate-to-extensive genetic overlap with the tested traits, therein stronger overlap with psychiatric than metabolic traits. We found most complete overlap between the anterior subunits and systolic blood pressure. Across all subunit and trait combinations, we pinpoint 95 novel, unique associated loci. The genes associated with these loci were enriched in gene sets linked to neuroimaging and neurodegeneration as well as metabolic markers, and were up-/down-regulated in tissues such as blood vessel and the liver. These findings help shed light on the genetic architecture of the hypothalamic subunits implicated in oxytocin and vasotocin and selected traits, and provide new avenues for future research.

Driven by advances in supercomputing, the scale of scientific simulation data has grown dramatically. In fields such as cosmology, particle data have become a common rep resentation, with state-of-the-art simulations now exceeding the trillion-particle mark. Consequently, the challenge of visually analyzing such massive datasets has become increasingly urgent. The traditional visual analysis workflow typically follows a "compression → storage → reconstruction → visualization" pipeline. However, this process is hampered by an extremely time consuming reconstruction stage, which severely impedes real-time interactive visualization. Moreover, in multi-time-step analyses, the enormous volume of reconstructed data creates significant I/O bottlenecks. In this work, we draw inspiration from 3D Gaussian splatting and compress the simulation data using Gaussian Mixture Models (GMMs), treating the resulting Gaussian kernels as fundamental rendering primitives. Our method renders billion-scale particles for each timestep in approximately 32 ms, requiring only 645 MB of GPU memory per timestep - nearly 20× smaller than the original 12 GB raw data. This eliminates costly reconstruction, accelerates the visual analysis pipeline, and overcomes I/O bottlenecks in multi-time-step analysis. Extensive experiments and comparisons across multiple datasets validate the effectiveness of our method.

During thymocyte development, positive selection produces cells whose T cell receptors (TCRs) bind to self MHC. Then, negative selection culls most thymocytes whose TCRs have too high an affinity for self MHC+peptide. Signal transduction events control these processes. CD8-αβ (via CD8-β) recruits p56lck to the immunological synapse and promotes signaling through the TCR. Conversely, PD-1 attenuates TCR signal transduction. We examined the roles of CD8-β and PD-1 in the survival of thymocytes in H-2k haplotype mice expressing a transgenic BM3 TCR, which has high affinity for the allogeneic H-2Kb MHC I molecule. In transgenic mice expressing both H-2Kb in the thymic medulla and the BM3 TCR, apoptosis eliminates most (but not all) post-selection thymocytes. To analyze the roles of CD8-β and PD-1 in the survival of post-selection thymocytes, we devised a novel probabilistic gating strategy employing Gaussian mixture models and statistical methods using sliding windows and changepoint detection. We found that at high levels of CD8-β and therefore high levels of CD8-αβ), thymocytes are prone to apoptosis, regardless of the PD-1 level. At intermediate levels of CD8-β, thymocyte survival increases concordantly with increasing PD-1 levels. At low levels of CD8-β, thymocyte survival is high regardless of the PD-1 level. Surviving DPlo post-selection thymocytes give rise to PD-1+CCR7+DN and PD-1+CCR7-DN post-selection thymocytes, which appear to become DN T cells and IELs, respectively. Thus, PD-1 appears to promote the survival of both IEL precursors and thymocytes destined for other fates. More strikingly, downregulation of CD8-β is a hallmark of autoreactive MHC I-restricted thymocytes that survive negative selection.

Accurate phylogenetic inference requires models that account for heterogeneity in molecular evolution. Mitochondrial protein-coding genes, which encode membrane-bound proteins composed of multiple transmembrane α-helices, exhibit considerable compositional and functional variation across structural regions, variation that is often overlooked in standard partitioning strategies. Here, we introduce TRAMPO (TRAnsMembrane Protein Order), a novel pipeline that incorporates predicted secondary structural features (i.e. matrix-facing, transmembrane, and intermembrane-facing domains) into phylogenetic partitioning schemes. We applied TRAMPO to seven mitochondrial datasets spanning crustaceans, hexapods, and vertebrates, and evaluated eight partitioning strategies based on combinations of codon position, strand, and secondary structure. Transmembrane helices, especially at second codon positions, showed pronounced thymine enrichment and hydrophobic amino-acid composition, reflecting domain-specific evolutionary constraints. To assess whether these structural patterns influence phylogenetic reconstruction, we performed maximum likelihood analyses under standard and Lie Markov models, General Heterogeneous evolution On a Single Topology, and profile mixture models. We also evaluated different models of among-site rate variation (including the proportion of invariant sites, gamma distributions, and FreeRates, which approximates rate heterogeneity using flexible discrete rate categories) to examine their interaction with partitioning strategies and overall model performance. Incorporating structural information into partitioning schemes consistently improved model fit and reduced apparent heterogeneity, as reflected in lower AIC values and more compositionally homogeneous partitions. These improvements translated into more consistent and topologically congruent phylogenetic trees across most datasets, while also reducing computational time. Notably, second codon positions within transmembrane helices were consistently retained as distinct partitions during model optimization, even in Mammals and Vertebrates, where secondary structure contributed little to overall model performance, underscoring their strong and conserved evolutionary signal. Surveys of tree space using quartet distances further supported these findings, with structurally informed models yielding more tightly clustered and internally consistent tree topologies. The benefits of structural partitioning were most pronounced in lineages of intermediate evolutionary depth and declined in ancient vertebrate and mammalian clades, where substitutional saturation accumulates with evolutionary time and strand asymmetry tends to emerge more frequently. In some cases, models with the lowest AIC did not yield the most congruent topologies, underscoring the limitations of information criteria when comparing models of different complexity. Overall, our findings demonstrate that secondary structural features, particularly the repetitive architecture of transmembrane helices, harbour meaningful phylogenetic signal. Incorporating this information into partitioning schemes improves tree reconstruction and mitigates underlying heterogeneity. TRAMPO provides a scalable, open-source tool to implement this approach in mitochondrial phylogenetics.

Super-resolution fluorescence microscopy, and specifically DNA-PAINT, provides localization precision down to ~2 nm enabling molecular-resolution imaging. To produce molecular maps of single biomolecules, their positions must be inferred from localizations stemming from single fluorescent molecules. Current clustering methods fail to exploit the full potential of the imaging method. Here, we introduce G5M, a modified Gaussian Mixture Modeling algorithm tailored to DNA-PAINT data. By incorporating prior knowledge of localization precision, spatial constraints, and DNA hybridization kinetics, G5M accurately infers true molecular positions while avoiding overfitting. In realistic simulations of dimers, G5M resolves molecules at the Rayleigh limit with a 27-fold higher recovery rate than current methods and <0.1% false positives. Applied to experimental datasets, G5M recovers full nuclear pore complex structures and detects higher-order CD20 oligomers induced by antibody treatment, outperforming conventional DNA-PAINT analysis. G5M is implemented in the open-source Picasso platform, offering an accessible solution for high-resolution, high-accuracy molecular mapping in super-resolution microscopy.

Understanding the complex microbial interactions and their implications for host health is a critical endeavor in biomedical research. In this paper, we propose a transformation-free Bayesian inference approach for estimating microbial and metabolomic association networks based on a latent Ising model. Our method addresses the challenges posed by the compositionality and zero-inflation of microbiome data, offering computational efficiency and versatility for mixed data types. By integrating two-component mixture models tailored to microbiome and metabolome data, along with spike-and-slab priors for sparse graph estimation and a pseudolikelihood approximation for efficient Bayesian computation, we provide a unified framework for joint microbial and metabolomic network inference. Simulation studies demonstrate the superior performance of our method compared to existing approaches, and an application to a real bacterial vaginosis microbiome-metabolome dataset reveals intriguing interaction patterns. Our proposed approach offers a promising avenue for uncovering biological insights from complex microbiome data and holds potential for advancing our understanding of microbiome-associated diseases and therapeutic interventions.

Longitudinal triglyceride-glucose index trajectories and kidney outcomes in patients with metabolic dysfunctionassociated fatty liver disease.

This study focuses on statistical inference for the class of quasi-infinitely divisible (QID) distributions, which was recently introduced by Lindner et al. [On quasi-infinitely divisible distributions. Trans Am Math Soc. 2018;370(12):8483–8520]. The paper presents a Fourier approach, based on the analogue of the Lévy–Khintchine theorem with a signed spectral measure. In particular, this method allows to recover the components in a mixture model μ = p μ ~ σ + ( 1 − p ) μ ∘ , which is a QID distribution, where 1/2 $ ]]> p > 1 / 2 , μ ~ σ is a centred normal distribution with unknown variance σ 2 , and the measure μ ∘ satisfies a certain nonparametric condition. We prove that for some subclasses of QID distributions, the considered estimates have polynomial rates of convergence. This is an interesting finding when compared to the logarithmic convergence rates of similar methods for infinitely divisible distributions, which cannot be improved in general. We demonstrate the numerical performance of the algorithm using simulated examples.

To address the challenge that existing vehicle chassis coordinated control methods struggle to balance the nonlinear couplings and control conflicts among Four-Wheel Steering (4WS), Direct Yaw-moment Control (DYC), and Active Suspension Systems (ASS), this paper proposes a Cooperative Distributed Model Predictive Control (Co-DMPC) strategy. First, the 4WS, DYC, and ASS are modeled as three interacting agents that effectively mitigate inter-subsystem control conflicts through information exchange and coupling compensation. Second, a Gaussian Mixture Model (GMM) is utilized to extract features from vehicle state data to enable the real-time grading of instability risks, which dynamically adjusts the control weights of the 4WS, DYC, and ASS agents. Finally, a distributed iterative optimization algorithm is designed to ensure that all agents converge to a global Pareto-optimal solution through rapid negotiation, achieving a balance between control performance and computational burden. Simulation results demonstrate that compared with No-Control and CMPC, the proposed Co-DMPC strategy significantly enhances the comprehensive performance of the vehicle. In terms of path tracking accuracy, the maximum tracking errors under high- and low-adhesion road conditions are reduced by 32.73% and 17%, respectively. Regarding roll stability, the peak roll angles of the vehicle are 0.27 rad and 0.01 rad under the respective conditions. For lateral stability, the proposed method maintains a more compact sideslip angle-yaw rate phase plane envelope, effectively achieving the coordinated optimization of chassis subsystems. Hardware-in-the-Loop (HIL) experiments further validate the performance and effectiveness of the controller.