Robust Prediction Research Articles

Robust metabolomic age prediction based on a wide selection of metabolites

Abstract Chronological age is a major risk factor for numerous diseases. However, chronological age does not capture the complex biological aging process. the difference between the chronological age and biologically driven aging could be more informative in reflecting health status. Here, we set out to develop a metabolomic age prediction model by applying ridge regression and bootstrapping with 826 metabolites (678 endogenous and 148 xenobiotics) measured by an untargeted platform in relatively healthy blood donors aged 18-75 years from the INTERVAL study (N=11,977;50.2% men). After bootstrapping internal validation, the metabolomic age prediction models demonstrated high performance with an adjusted R2 of 0.83 using all metabolites and 0.82 using only endogenous metabolites. The former was significantly associated with obesity and cardiovascular disease (CVD) in the NEO study (N=599;47.0% men; age range=45-65) due to the contribution of medication derived metabolites—namely salicylate and ibuprofen—and environmental exposures such as cotinine. Additional metabolomic age prediction models using all metabolites were developed for men and women separately. The models had high performance (R²=0.85 and 0.86) but shared a moderate correlation of 0.72. Furthermore, we observed 163 sex-dimorphic metabolites, including threonine, glycine, cholesterol, and androgenic and progesterone-related metabolites. Our strongest predictors across all models were novel and included hydroxyasparagine (Model Endo+Xeno β=4.74), vanillylmandelate (β=4.07), and 5,6-dihydrouridine (β=-4.2). Our study presents a robust metabolomic age model that reveals distinct sex-based age-related metabolic patterns and illustrates the value of including xenobiotic to enhance metabolomic prediction accuracy.

PnP-epipolar optimization approach for VSLAM using robust pose prediction

Abstract Visual Simultaneous Localization and Mapping (VSLAM) is easily influenced by aggressive motion and low overlap images. Based on the challenge, we propose a VSLAM system with the robust tracking method. The feature tracking model is constructed by the robust pose prediction and PnP-Epipolar optimization. Firstly, as conventional VSLAM systems struggle with aggressive motion, we improve the constant acceleration prediction to estimate a dependable initial pose. The pose prediction takes into account the variation of the rotation axis, providing a more reliable initial guess for the pose optimization. Secondly, a novel global feature matching with BoW acceleration and map-point projection is performed to increase the number of matches. The matching method considers the points without map-point observation. Thirdly, according to the matches, we constitute the PnP-Epipolar constraint to optimize the pose with low overlap images. Finally, we compare our algorithm with several state-of-the-art VSLAM systems on KITTI dataset and EuRoC MAV dataset. The experimental results indicate that the proposed VSLAM reduces the RMSE by about 30% on average. For some sequences containing aggressive angle turning scene, the RMSE reduction even reaches 50%.

EnrichRBP: an automated and interpretable computational platform for predicting and analyzing RNA-binding protein events.

Predicting RNA-binding proteins (RBPs) is central to understanding post-transcriptional regulatory mechanisms. Here, we introduce EnrichRBP, an automated and interpretable computational platform specifically designed for the comprehensive analysis of RBP interactions with RNA. EnrichRBP is a web service that enables researchers to develop original deep learning and machine learning architectures to explore the complex dynamics of RNA-binding proteins. The platform supports 70 deep learning algorithms, covering feature representation, selection, model training, comparison, optimization, and evaluation, all integrated within an automated pipeline. EnrichRBP is adept at providing comprehensive visualizations, enhancing model interpretability, and facilitating the discovery of functionally significant sequence regions crucial for RBP interactions. In addition, EnrichRBP supports base-level functional annotation tasks, offering explanations and graphical visualizations that confirm the reliability of the predicted RNA binding sites. Leveraging high-performance computing, EnrichRBP provides ultra-fast predictions ranging from seconds to hours, applicable to both pre-trained and custom model scenarios, thus proving its utility in real-world applications. Case studies highlight that EnrichRBP provides robust and interpretable predictions, demonstrating the power of deep learning in the functional analysis of RBP interactions. Finally, EnrichRBP aims to enhance the reproducibility of computational method analyses for RNA-binding protein sequences, as well as reduce the programming and hardware requirements for biologists, thereby offering meaningful functional insights. EnrichRBP is available at https://airbp.aibio-lab.com/. The source code is available at https://github.com/wangyb97/EnrichRBP, and detailed online documentation can be found at https://enrichrbp.readthedocs.io/en/latest/. Supplementary data are available at Bioinformatics online.

THE COMPARISON OF LONG SHORT-TERM MEMORY AND BIDIRECTIONAL LONG SHORT-TERM MEMORY FOR FORECASTING COAL PRICE

Coal remains vital for global energy despite recent demand fluctuations due to the COVID-19 pandemic and geopolitical tensions. The International Energy Agency (IEA) projected a decline in global coal demand starting in early 2024, driven by increasing renewable energy adoption. As one of the top coal exporters, Indonesia must adjust to these changes. This study aims to forecast future coal prices using historical data from Indonesia's Ministry of Energy and Mineral Resources (KESDM), applying and comparing Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM) models. While BiLSTM has shown advantages in other contexts and studies, its effectiveness for coal price forecasting remains underexplored. To ensure robust predictions, we employ walk-forward validation, which divides the data into six segments and evaluates 90 hyperparameter combinations across all segments. The BiLSTM model consistently outperforms the LSTM model, achieving lower average RMSE and MAPE values. Specifically, BiLSTM records an average MAPE of 7.847 and RMSE of 10.485, compared to LSTM's 10.442 and 11.993, respectively. The Diebold-Mariano (DM) test using squared error and absolute error loss functions further corroborates these findings, with most segments showing significant improvements in favor of BiLSTM, indicated by negative DM-test statistics and p-values below 0.01 or 0.10. This superior performance continues into the testing data, where BiLSTM maintains lower error metrics and a significant result of the DM test, underscoring its reliability for forecasting. In the final stage, the forecasts from both models indicate a nearly linear downward trend in coal prices over the next 18 months, aligning with the International Energy Agency's 2023 projection of a structural decline in coal demand driven by the sustained growth of clean energy technologies.

Development of immune-derived molecular markers for preeclampsia based on multiple machine learning algorithms

Preeclampsia (PE) is a major pregnancy-specific cardiovascular complication posing latent life-threatening risks to mothers and neonates. The contribution of immune dysregulation to PE is not fully understood, highlighting the need to explore molecular markers and their relationship with immune infiltration to potentially inform therapeutic strategies. We used bioinformatics tools to analyze gene expression data from the Gene Expression Omnibus (GEO) database using the GEOquery package in R. Differential expression analysis was performed using the DESeq2 and limma packages, followed by analysis of variance to identify immune-related differentially expressed genes (DEGs). Several machine learning algorithms, including least absolute shrinkage and selection operator (LASSO), bagged trees, and random forest (RF), were used to select immune-related signaling genes closely associated with the occurrence of PE. Our analysis identified 34 immune source–related DEGs. Using the identified PE- and immune source–related genes, we constructed a diagnostic forecasting model employing several ML algorithms. We identified six types of statistically significant immune cells in patients with PE and discovered a strong relationship between biomarkers and immune cells. Moreover, the immune-derived hub genes for PE exhibited strong binding capabilities with drugs, such as alitretinoin, tretinoin, and acitretin. This study presents a robust prediction model for PE that integrates multiple machine learning–derived immune-related biomarkers. Our results indicate that these biomarkers may outperform previously reported molecular signatures in predicting PE and provide insights into the mechanisms underlying immune dysregulation in PE. Further validation in larger cohorts could lead to their clinical application in PE prediction and treatment.

Meta-Learning Enables Complex Cluster-Specific Few-Shot Binding Affinity Prediction for Protein-Protein Interactions.

Predicting protein-protein interaction (PPI) binding affinities in unseen protein complex clusters is essential for elucidating complex protein interactions and for the targeted screening of peptide- or protein-based drugs. We introduce MCGLPPI++, a meta-learning framework designed to improve the adaptability of pretrained geometric models in such scenarios. To effectively boost the meta-learning optimization by injecting prior intersample distribution knowledge, three specially designed training sample cluster splitting patterns based on protein interaction interfaces are introduced. Additionally, MCGLPPI++ is equipped with an independent energy component which explicitly models interface nonbonded interaction energies closely related to the strengths of PPIs. To validate our approach, we curate a new data set featuring a challenging test cluster of T-cell receptors binding to antigenic peptide-MHC molecules (TCR-pMHC). Experimental results show that geometric models enhanced by the MCGLPPI++ framework achieve significantly more robust binding affinity predictions after fine-tuning on a few samples from this novel cluster compared to their vanilla counterparts, which demonstrates the effectiveness of the framework.

Machine Learning for Automated Identification of Anatomical Landmarks in Ultrasound Periodontal Imaging.

To identify landmarks in ultrasound periodontal images and automate the image-based measurements of gingival recession (iGR), gingival height (iGH), and alveolar bone level (iABL) using machine learning. We imaged 184 teeth from 29 human subjects. The dataset included 1580 frames for training and validating the U-Net CNN machine learning model, and 250 frames from new teeth that were not used in training for testing the generalization performance. The predicted landmarks including the tooth, gingiva, bone, gingival margin (GM), cementoenamel junction (CEJ), and alveolar bone crest (ABC), were compared to manual annotations. We further demonstrated automated measurements of the clinical metrics iGR, iGH, and iABL. Over 98% of predicted GM, CEJ, and ABC distances are within 200 µm of the manual annotation. Bland-Altman analysis revealed biases (bias of machine learning versus ground truth) of -0.1 µm, -37.6 µm, and -40.9 µm, with 95% limits of agreement of [-281.3, 281.0] µm, [-203.1, 127.9] µm, and [-297.6, 215.8] µm for iGR, iGH, and iABL, respectively, when compared to manual annotations. On the test dataset, the biases were 167.5 µm, 40.1 µm, and 78.7 µm with 95% CIs of [-1175, 1510] µm, [-910.3, 990.4] µm, and [-1954, 1796] µm for iGR, iGH, and iABL, respectively. The proposed machine learning model demonstrates robust prediction performance, with the potential to enhance the efficiency of clinical periodontal diagnosis by automating landmark identification and clinical metrics measurements.

Robust RNA secondary structure prediction with a mixture of deep learning and physics-based experts.

A mixture-of-experts (MoE) approach has been developed to mitigate the poor out-of-distribution (OOD) generalization of deep learning (DL) models for single-sequence-based prediction of RNA secondary structure. The main idea behind this approach is to use DL models for in-distribution (ID) test sequences to leverage their superior ID performances, while relying on physics-based models for OOD sequences to ensure robust predictions. One key ingredient of the pipeline, named MoEFold2D, is automated ID/OOD detection via consensus analysis of an ensemble of DL model predictions without requiring access to training data during inference. Specifically, motivated by the clustered distribution of known RNA structures, a collection of distinct DL models is trained by iteratively leaving one cluster out. Each DL model hence serves as an expert on all but one cluster in the training data. Consequently, for an ID sequence, all but one DL model makes accurate predictions consistent with one another, while an OOD sequence yields highly inconsistent predictions among all DL models. Through consensus analysis of DL predictions, test sequences are categorized as ID or OOD. ID sequences are subsequently predicted by averaging the DL models in consensus, and OOD sequences are predicted using physics-based models. Instead of remediating generalization gaps with alternative approaches such as transfer learning and sequence alignment, MoEFold2D circumvents unpredictable ID-OOD gaps and combines the strengths of DL and physics-based models to achieve accurate ID and robust OOD predictions.

Comparative Analysis of Stock Price Prediction Using Deep Learning with Data Scaling Method

The dynamic and unpredictable nature of stock prices makes accurate forecasting an important challenge in financial analysis. This study aims to compare the performance of three deep learning models, namely, Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM) in predicting stock prices on historical daily banking data from Yahoo Finance. The main objective is to determine the model that is best able to capture sequential patterns and temporal dependencies in stock price movements. Each model was trained and op-timized through data scaling, namely MinMax Scaler and Standard Scaler, with performance evaluated using Root Mean Square Error (RMSE) as the primary metric. Results show that while the RNN provides a basic approach, the GRU and LSTM models produce higher prediction accuracy, with GRU achieving the lowest RMSE thanks to its better ability to maintain long-term depend-encies. The RMSE achieved by RNN, GRU, and LSTM were 211.47, 158.89, and 197.45, respectively. The lowest error results were achieved when using MinMax Scaler. The use of MinMax Scaler here shows a better performance improvement with an average improvement of 22.57% compared to using Standard Scaler. This comparative analysis contributes to providing empirical insight into the relative effectiveness of the tested architectures. The findings suggest that the combination of GRU and MinMax Scaler can be a more reliable tool for financial forecasting, with the potential to develop more robust stock prediction applications under fluctuating market conditions.

Uncertainty-Aware Multimodal Trajectory Prediction via a Single Inference from a Single Model.

In the domain of autonomous driving, trajectory prediction plays a pivotal role in ensuring the safety and reliability of autonomous systems, especially when navigating complex environments. Unfortunately, trajectory prediction suffers from uncertainty problems due to the randomness inherent in the driving environment, but uncertainty quantification in trajectory prediction is not widely addressed, and most studies rely on deep ensembles methods. This study presents a novel uncertainty-aware multimodal trajectory prediction (UAMTP) model that quantifies aleatoric and epistemic uncertainties through a single forward inference. Our approach employs deterministic single forward pass methods, optimizing computational efficiency while retaining robust prediction accuracy. By decomposing trajectory prediction into velocity and yaw components and quantifying uncertainty in both, the UAMTP model generates multimodal predictions that account for environmental randomness and intention ambiguity. Evaluation on datasets collected by CARLA simulator demonstrates that our model not only outperforms Deep Ensembles-based multimodal trajectory prediction method in terms of accuracy such as minFDE and miss rate metrics but also offers enhanced time to react for collision avoidance scenarios. This research marks a step forward in integrating efficient uncertainty quantification into multimodal trajectory prediction tasks within resource-constrained autonomous driving platforms.

A novel sensor-independent convolutional neural network for structural damage detection：illustrated by a case study on gantry crane

Structural damage detection is a critical technology for ensuring the safety of engineering structures. In recent years, intelligent structural damage detection algorithms based on machine learning have garnered widespread attention globally, thanks to their efficient and reliable detection performance. However, existing damage detection methods largely depend on the use of sensors, which not only increases operational and maintenance costs but also limits the applicability of these methods—especially in parts of structures where sensors are difficult to install. To overcome this barrier, this paper introduces a novel sensor-independent convolutional neural network (CNN) approach for predicting the maximum response of structures under seismic excitation. This method utilizes IDA method to collect a large number of samples (these samples have been uploaded to Mendeley Data for reference). It employs innovative data processing and grouping methods for feature alignment as well as the division of the test and training sets. By automatically extracting the signal characteristics of seismic excitation and optimizing feature combinations through multi-layer fusion, this method achieves high-precision and robust prediction without relying on sensors. To validate the effectiveness of this method, this study conducted a performance evaluation on a typical gantry crane metal structure. The results show that the proposed sensor-independent CNN algorithm achieves high-precision structural response prediction, with over 99% accuracy for peak responses, demonstrating the robustness and reliability of the model. It also performs excellently in key metrics such as root mean square error, error bias, and learning rate progression, making it a reliable and cost-effective solution for real-time seismic damage detection.

Data Driven-Based Machine Learning Modelling and Empirical Correlations for Predicting Snow-Covered Area in the Swat Region, Pakistan

In recent decades, global and regional climate change has emerged as a significant challenge with potential catastrophic consequences, including hurricanes, floods, sea level rise, and temperature shifts. Snow-covered area (SCA) serves as a crucial climatic parameter reflecting climate changes, yet accurately determining SCA proves to be a challenging and time-consuming task. This study aims to develop robust prediction models for SCA by employing three machine learning (ML) approaches using readily available climatic data from Swat, Pakistan, spanning two decades. The climate data encompass precipitation, daily maximum/minimum temperatures, and SCA measurements. Three ML methods—artificial neural networks (ANN), functional networks (FN), and adaptive neuro-fuzzy inference systems (ANFIS)—were employed to model SCA. Accuracy measures, including coefficient of determination (R2), average absolute percentage error (AAPE), and root mean squared error (RMSE) were utilized to evaluate model performance. All three ML models exhibited superior performance, with high R2 values and low errors. Accuracy indicators of the ANN model are better than FN and ANFIS models, yielding the highest R2 (0.956) and minimum RMSE and AAPE values (0.61 and 0.91). ANFIS demonstrated slightly better performance than FN, with RMSE, AAPE, and R2 values of 0.65, 1.1, and 0.950, respectively. FN yielded RMSE, AAPE, and R values of 1.14, 1.72, and 0.85, respectively. Additionally, two empirical correlations were derived from the optimized FN and ANN models for SCA prediction using the same input variables. This study underscores the efficacy of ML techniques in accurately and consistently predicting SCA parameters, offering valuable insights into climate change and its consequences.

Geochemical speciation and activation risks of Cd, Ni, and Zn in soils with naturally high background in karst regions of southwestern China.

Agricultural soils in karst regions present a remarkable paradox where high geochemical background levels of heavy metals correspond with unexpectedly low crop uptake, challenging traditional risk assessment frameworks and limiting agricultural development. To decode this paradox, we investigated the geochemical speciation of cadmium (Cd), nickel (Ni), and zinc (Zn) in soil-rice systems in southwestern China, which collectively constitute the world's largest continuous karst region and represent diverse soil weathering stages. We employed three chemical extraction methods that revealed reactive pools ranking as Cd (58.74 %) > Zn (7.31 %) > Ni (4.65 %) and risk patterns varying with soil type (Andosols > Cambisols/Gleysols > Lithosols), while multi-surface speciation model (MSM) elucidated the underlying mechanisms. We identified a stage activation-contamination model (SACM) that demonstrates how pH-dependent weathering controlled heavy metal distribution among dissolved, surface-active, semi-stable carbonate, and nonactive species, thereby explaining the observed risk patterns. Specifically, in alkaline soils (pH > 7.50), Cd and Zn were primarily humic acid (HA)- and carbonate-bound, while Ni was goethite-bound. As weathering intensified, the reactive pool shifted to more active HA- and hydrous ferric oxide (HFO)-bound species. In acidic soils (pH < 6.50), dissolved and HA-bound species dominated. Both random forest, offering robust predictions using readily available data, and MSM stepwise regression models, providing high accuracy with mechanistic insights, effectively predicted rice risks. This study from speciation and activation model to prediction model clarifies why standardized risk assessments fail in karst regions and offers practical tools for accurate risk evaluation and management in these agricultural environments.

Forecasting A Major Banking Corporation Stock Prices Using LSTM Neural Networks

The increasing complexity of stock market predictions necessitates advanced computational techniques to address the unique challenges posed by financial data's non-linear and volatile nature. This study aims to leverage Long Short-Term Memory (LSTM) neural networks to accurately forecast stock prices, using historical data collected from a major banking corporation as a primary source. The LSTM model excels at processing sequential time-series data, allowing it to predict monthly stock closing prices over a one-year horizon with a high degree of precision. Our findings indicate a Root Mean Squared Error (RMSE) of 3.2, underscoring the model's efficiency and reliability in financial forecasting tasks. The novelty of this research lies in the systematic incorporation of preprocessing techniques and fine-tuned hyperparameters to optimize model performance. Furthermore, this study explores the practical implications of implementing LSTM models in real-world trading scenarios, analyzing their adaptability to dynamic market conditions and their potential integration into automated trading systems. These findings contribute to the growing body of knowledge in financial analytics and demonstrate the viability of machine learning-based solutions for accurate and robust market predictions.

Identification of RNA-binding protein genes associated with renal rejection and graft survival

Rejection is one of the major factors affecting the long-term prognosis of kidney transplantation, and timely recognition and aggressive treatment of rejection is essential to prevent disease progression. RBPs are proteins that bind to RNA to form ribonucleoprotein complexes, thereby affecting RNA stability, processing, splicing, localization, transport, and translation, which play a key role in post-transcriptional gene regulation. However, their role in renal transplant rejection and long-term graft survival is unclear. The aim of this study was to comprehensively analyze the expression of RPBs in renal rejection and use it to construct a robust prediction strategy for long-term graft survival. The microarray expression profiles used in this study were obtained from GEO database. In this study, a total of eight hub RBPs were identified, all of which were upregulated in renal rejection samples. Based on these RBPs, the renal rejection samples could be categorized into two different clusters (cluster A and cluster B). Inflammatory activation in cluster B and functional enrichment analysis showed a strong association with rejection-related pathways. The diagnostic prediction model had a high diagnostic accuracy for T cell mediated rejection (TCMR) in renal grafts (area under the curve = 0.86). The prognostic prediction model effectively predicts the prognosis and survival of renal grafts (p < .001) and applies to both rejection and non-rejection situations. Finally, we validated the expression of hub genes, and patient prognosis in clinical samples, respectively, and the results were consistent with the above analysis.

Enhancing AI Through Statistical Methods for Improved Decision-Making

This study examines the crucial relationship between Artificial Intelligence (AI) and statistics, which has grown in importance due to the information boom and computational advances. Modern data-intensive applications have led these domains to merge, despite their parallel history. This document explores AI's fundamentals, focusing on machine learning and deep learning, and how statistical methods help AI perform tasks like decision-making and pattern identification. We integrate AI with robust statistical modelling and prediction to improve AI transparency and effectiveness. This multidisciplinary approach emphasizes theoretical advances, practical applications, ethical considerations, and future problems at the interface of AI and statistics. We encourage AI and statistics communities to collaborate to promote innovation and responsible AI development and deployment. This collection of publications is a comprehensive resource for researchers and practitioners using AI and statistics to better decision-making and predictive analytics

Video-Based Lifting Action Recognition Using Rank-Altered Kinematic Feature Pairs.

To identify lifting actions and count the number of lifts performed in videos based on robust class prediction and a streamlined process for reliable real-time monitoring of lifting tasks. Traditional methods for recognizing lifting actions often rely on deep learning classifiers applied to human motion data collected from wearable sensors. Despite their high performance, these methods can be difficult to implement on systems with limited hardware resources. The proposed method follows a five-stage process: (1) BlazePose, a real-time pose estimation model, detects key joints of the human body. (2) These joints are preprocessed by smoothing, centering, and scaling techniques. (3) Kinematic features are extracted from the preprocessed joints. (4) Video frames are classified as lifting or nonlifting using rank-altered kinematic feature pairs. (5) A lifting counting algorithm counts the number of lifts based on the class predictions. Nine rank-altered kinematic feature pairs are identified as key pairs. These pairs were used to construct an ensemble classifier, which achieved 0.89 or above in classification metrics, including accuracy, precision, recall, and F1 score. This classifier showed an accuracy of 0.90 in lifting counting and a latency of 0.06ms, which is at least 12.5 times faster than baseline classifiers. This study demonstrates that computer vision-based kinematic features could be adopted to effectively and efficiently recognize lifting actions. The proposed method could be deployed on various platforms, including mobile devices and embedded systems, to monitor lifting tasks in real-time for the proactive prevention of work-related low-back injuries.

A novel data-driven bridge monitoring data prediction framework using improved intelligent optimization-assisted deep learning

Bridge monitoring data prediction plays a crucial role in maintaining the serviceability and safety of bridges. To assess the performance of bridge structures in advance, an accurate and robust prediction model for measurement data prediction is needed. In this context, this paper presents a novel hybrid prediction model called IGWO-VMD-IGWO-LSTM, which aims to improve the accuracy and generalization ability of bridge monitoring data prediction. First, the proposed model incorporates variational mode decomposition (VMD) and an improved gray wolf optimization (IGWO) algorithm to decompose the measured data into several intrinsic mode functions (IMFs) to capture the signal’s features comprehensively. Then, long short-term memory (LSTM) models are established individually for each IMF, with hyperparameters optimization using the IGWO algorithm. Finally, the integrated reconstructed subsequences yield the prediction results of the proposed IGWO-VMD-IGWO-LSTM model. Thus, the proposed framework avoids exploring the complex internal mechanism of bridge behavior evolution. The measurement data collected from a real bridge demonstrates the feasibility of the proposed prediction method, and different deep learning models are used for comparison. The results demonstrate a significant improvement in prediction performance comparison with the other concerned models, highlighting the strong generalization ability and robustness of the proposed IGWO-VMD-IGWO-LSTM model.

Evaluating Photosynthetic Light Response Models for Leaf Photosynthetic Traits in Paddy Rice (Oryza sativa L.) Under Field Conditions.

Accurate photosynthetic parameters obtained from photosynthetic light-response curves (LRCs) are crucial for enhancing our comprehension of plant photosynthesis. However, the task of fitting LRCs is still demanding due to diverse variations in LRCs under different environmental conditions, as previous models were evaluated based on a limited number of leaf traits and a small number of LRCs. This study aimed to compare the performance of nine LRC models in fitting a set of 108 LRCs measured from paddy rice (Oryza sativa L.) grown in field across 3 years under different leaf positions, leaf ages, nitrogen levels, irrigation levels, and varieties. The shape of 108 LRCs varies significantly under a range of leaf traits, which can be typed into three leaf light-acclimation types-high-light leaves (HL-1 and HL-2), and low-light leaves (LL). The accuracy of these models was evaluated by (1) LRCs from three acclimation types: HL-1 and HL-2, and LL; and (2) LRCs across three irradiance stages: light-limited, light-saturated, and photoinhibition. Results indicate that the Ye model emerged as the top performance among the nine models, particularly in the photoinhibition stage of LL leaves, with median values of R2, SSE, and AIC of 0.99, 2.39, and -14.03, respectively. Furthermore, the Ye model produced the most accurate predictions of key photosynthetic parameters, including dark respiration (RD), light-compensation point (Icomp), maximum net photosynthetic rate (PNmax), and light-saturation point (Isat). Results also suggest that PNImax and Imax were the most appropriate parameters to describe photosynthetic activity at the light-saturation point. These findings have significant implications for improving the accuracy of fitting LRCs, and thus robust predictions of photosynthetic parameters in rice under different environmental conditions.

Development and validation of a novel model to predict post-stroke cognitive impairment within 6 months after acute ischemic stroke.

Cognitive decline following acute ischemic stroke (AIS), termed post-stroke cognitive impairment (PSCI), is a prevalent phenomenon that significantly elevates disability and mortality rates among affected patients. The objective of this investigation was to develop a robust clinical prediction model capable of forecasting PSCI within six months post-AIS and subsequently validate its effectiveness. A cohort of 573 AIS patients was stratified into two groups: those with PSCI (260 cases) and those who remained cognitively normal (CN) (313 cases). These patients were further subdivided into three distinct cohorts: a development cohort comprising 193 AIS patients, an internal validation cohort with 193 AIS patients, and an external validation cohort encompassing 187 AIS patients. A thorough multifactor logistic regression analysis was conducted to identify independent predictors of PSCI, which were subsequently incorporated into the prediction model for comprehensive analysis and validation. The discriminatory power, calibration accuracy, and clinical net benefits of the prediction model were rigorously evaluated using the area under the receiver operating characteristic curve (AUC-ROC), calibration plots, and decision curve analyses, respectively. Utilizing a meticulously selected panel of variables, including smoking status, alcohol consumption, female gender, low educational attainment, NIHSS score at admission, stroke progression, diabetes mellitus, atrial fibrillation, stroke localization, HCY levels, and Lp-PLA2 levels, a clinical prediction model was formulated to predict the occurrence of PSCI within six months of AIS. The model demonstrated AUC-ROC values of 0.898 (95%CI, 0.853-0.942), 0.847 (95%CI, 0.794-0.901), and 0.849 (95%CI, 0.7946-0.9031) in the development, internal validation, and external validation cohorts, respectively. Further validation through calibration curve analyses, Hosmer-Lemeshow goodness-of-fit tests, and additional metrics confirmed the model's impressive predictive performance. The proposed model exhibits strong discriminative ability for predicting PSCI and holds considerable promise for guiding clinical decision-making. However, ongoing optimization with multicenter data is necessary to bolster its robustness and broaden its applicability.