Matthews Correlation Research Articles

Predictive modeling of gestational weight gain: a machine learning multiclass classification study

BackgroundGestational weight gain (GWG) is a critical factor influencing maternal and fetal health. Excessive or insufficient GWG can lead to various complications, including gestational diabetes, hypertension, cesarean delivery, low birth weight, and preterm birth. This study aims to develop and evaluate machine learning models to predict GWG categories: below, within, or above recommended guidelines.MethodsWe analyzed data from the Araraquara Cohort, Brazil, which comprised 1557 pregnant women with a gestational age of 19 weeks or less. Predictors included socioeconomic, demographic, lifestyle, morbidity, and anthropometric factors. Five machine learning algorithms (Random Forest, LightGBM, AdaBoost, CatBoost, and XGBoost) were employed for model development. The models were trained and evaluated using a multiclass classification approach. Model performance was assessed using metrics such as area under the ROC curve (AUC-ROC), F1 score and Matthew’s correlation coefficient (MCC).ResultsThe outcomes were categorized as follows: GWG within recommendations (28.7%), GWG below (32.5%), and GWG above recommendations (38.7%). The XGBoost presented the best overall model, achieving an AUC-ROC of 0.79 for GWG within, 0.76 for GWG below, and 0.65 for GWG above. The LightGBM also performed well with an AUC-ROC of 0.79 for predicting GWG within recommendations, 0.76 for GWG below, and 0.624 for GWG above. The most important predictors of GWG were pre-gestational BMI, maternal age, glycemic profile, hemoglobin levels, and arm circumference.ConclusionMachine learning models can effectively predict GWG categories, offering a valuable tool for early identification of at-risk pregnancies. This approach can enhance personalized prenatal care and interventions to promote optimal pregnancy outcomes.

An Artificial Intelligence Approach for Test‐Free Identification of Sarcopenia

ABSTRACTBackgroundThe diagnosis of sarcopenia relies extensively on human and equipment resources and requires individuals to personally visit medical institutions. The objective of this study was to develop a test‐free, self‐assessable approach to identify sarcopenia by utilizing artificial intelligence techniques and representative real‐world data.MethodsThis multicentre study enrolled 11 661 middle‐aged and older adults from a national survey initialized in 2011. Follow‐up data from the baseline cohort collected in 2013 (n = 9403) and 2015 (n = 10 356) were used for validation. Sarcopenia was retrospectively diagnosed using the Asian Working Group for Sarcopenia 2019 framework. Baseline age, sex, height, weight and 20 functional capacity (FC)–related binary indices (activities of daily living = 6, instrumental activities of daily living = 5 and other FC indices = 9) were considered as predictors. Multiple machine learning (ML) models were trained and cross‐validated using 70% of the baseline data to predict sarcopenia. The remaining 30% of the baseline data, along with two follow‐up datasets (n = 9403 and n = 10 356, respectively), were used to assess model performance.ResultsThe study included 5634 men and 6027 women (median age = 57.0 years). Sarcopenia was identified in 1288 (11.0%) individuals. Among the 20 FC indices, the running/jogging 1 km item showed the highest predictive value for sarcopenia (AUC [95%CI] = 0.633 [0.620–0.647]). From the various ML models assessed, a 24‐variable gradient boosting classifier (GBC) model was selected. This GBC model demonstrated favourable performance in predicting sarcopenia in the holdout data (AUC [95%CI] = 0.831 [0.808–0.853], accuracy = 0.889, recall = 0.441, precision = 0.475, F1 score = 0.458, Kappa = 0.396 and Matthews correlation coefficient = 0.396). Further model validation on the temporal scale using two longitudinal datasets also demonstrated good performance (AUC [95%CI]: 0.833 [0.818–0.848] and 0.852 [0.840–0.865], respectively). The model's built‐in feature importance ranking and the SHapley Additive exPlanations method revealed that lifting 5 kg and running/jogging 1 km were relatively important variables among the 20 FC items contributing to the model's predictive capacity, respectively. The calibration curve of the model indicated good agreement between predictions and actual observations (Hosmer and Lemeshow p = 0.501, 0.451 and 0.374 for the three test sets, respectively), and decision curve analysis supported its clinical usefulness. The model was implemented as an online web application and exported as a deployable binary file, allowing for flexible, individualized risk assessment.ConclusionsWe developed an artificial intelligence model that can assist in the identification of sarcopenia, particularly in settings lacking the necessary resources for a comprehensive diagnosis. These findings offer potential for improving decision‐making and facilitating the development of novel management strategies of sarcopenia.

Leveraging Mixture of Experts and Deep Learning-Based Data Rebalancing to Improve Credit Fraud Detection

Credit card fraud detection is a critical challenge in the financial sector due to the rapidly evolving tactics of fraudsters and the significant class imbalance betweenegitimate and fraudulent transactions. Traditional models, while effective to some extent, often suffer from high false positive rates and fail to generalize well to emerging fraud patterns. In this paper, we propose a novel approach that integrates a Mixture of Experts (MoE) model with a Deep Neural Network-based Synthetic Minority Over-sampling Technique (DNN-SMOTE) to enhance fraud detection performance. The MoE modeleverages multiple specialized expert networks, each trained to detect specific types of fraud, while the DNN-SMOTE generates high-quality synthetic samples to address the class imbalance. Our experimental results on a publicly available dataset demonstrate that the proposed method achieves a classification accuracy of 99.93%, a true positive rate of 84.69%, and a true negative rate of 99.95%. The Matthews Correlation Coefficient (MCC) of 0.7883 further highlights the model’s balanced performance in detecting fraudulent transactions. These results underscore the effectiveness of combining MoE with DNN-SMOTE, offering a robust solution for real-world credit card fraud detection scenarios.

Diagnosis of Custard Apple Disease Based on Adaptive Information Entropy Data Augmentation and Multiscale Region Aggregation Interactive Visual Transformers

Accurate diagnosis of plant diseases is crucial for crop health. This study introduces the EDA–ViT model, a Vision Transformer (ViT)-based approach that integrates adaptive entropy-based data augmentation for diagnosing custard apple (Annona squamosa) diseases. Traditional models like convolutional neural network and ViT face challenges with local feature extraction and large dataset requirements. EDA–ViT overcomes these by using a multi-scale weighted feature aggregation and a feature interaction module, enhancing both local and global feature extraction. The adaptive data augmentation method refines the training process, boosting accuracy and robustness. With a dataset of 8226 images, EDA–ViT achieved a classification accuracy of 96.58%, an F1 score of 96.10%, and a Matthews Correlation Coefficient (MCC) of 92.24%, outperforming other models. The inclusion of the Deformable Multi-head Self-Attention (DMSA) mechanism further enhanced feature capture. Ablation studies revealed that the adaptive augmentation contributed to a 0.56% accuracy improvement and a 0.34% increase in MCC. In summary, EDA–ViT presents an innovative solution for custard apple disease diagnosis, with potential applications in broader agricultural disease detection, ultimately aiding precision agriculture and crop health management.

AI4ACEIP: A Computing Tool to Identify Food Peptides with High Inhibitory Activity for ACE by Merged Molecular Representation and Rich Intrinsic Sequence Information Based on an Ensemble Learning Strategy.

Hypertension is a common chronic disorder and a major risk factor for cardiovascular diseases. Angiotensin-converting enzyme (ACE) converts angiotensin I to angiotensin II, causing vasoconstriction and raising blood pressure. Pharmacotherapy is the mainstay of traditional hypertension treatment, leading to various negative side effects. Some food-derived peptides can suppress ACE, named ACEIP with fewer undesirable effects. Therefore, it is crucial to seek strong dietary ACEIP to aid in hypertension treatment. In this article, we propose a new model called AI4ACEIP to identify ACEIP. AI4ACEIP uses a novel two-layer stacked ensemble architecture to predict ACEIP relying on integrated view features derived from sequence, large language models, and molecular-based information. The analysis of feature combinations reveals that four selected integrated feature pairs exhibit enhancing performance for identifying ACEIP. For finding meta models with strong abilities to learn information from integrated feature pairs, PowerShap, a feature selection method, is used to select 40 optimal feature and meta model combinations. Compared with seven state-of-the-art methods on the source and clear benchmark data sets, AI4ACEIP significantly outperformed by 8.47 to 20.65% and 5.49 to 14.42% for Matthew's correlation coefficient. In brief, AI4ACEIP is a reliable model for ACEIP prediction and is freely available at https://github.com/abcair/AI4ACEIP.

MIPPIS: protein–protein interaction site prediction network with multi-information fusion

BackgroundThe prediction of protein–protein interaction sites plays a crucial role in biochemical processes. Investigating the interaction between viruses and receptor proteins through biological techniques aids in understanding disease mechanisms and guides the development of corresponding drugs. While various methods have been proposed in the past, they often suffer from drawbacks such as long processing times, high costs, and low accuracy.ResultsAddressing these challenges, we propose a novel protein–protein interaction site prediction network based on multi-information fusion. In our approach, the initial amino acid features are depicted by the position-specific scoring matrix, hidden Markov model, dictionary of protein secondary structure, and one-hot encoding. Simultaneously, we adopt a multi-channel approach to extract deep-level amino acids features from different perspectives. The graph convolutional network channel effectively extracts spatial structural information. The bidirectional long short-term memory channel treats the amino acid sequence as natural language, capturing the protein’s primary structure information. The ProtT5 protein large language model channel outputs a more comprehensive amino acid embedding representation, providing a robust complement to the two aforementioned channels. Finally, the obtained amino acid features are fed into the prediction layer for the final prediction.ConclusionCompared with six protein structure-based methods and six protein sequence-based methods, our model achieves optimal performance across evaluation metrics, including accuracy, precision, F1, Matthews correlation coefficient, and area under the precision recall curve, which demonstrates the superiority of our model.

An Accurate and Efficient Approach to Knowledge Extraction from Scientific Publications Using Structured Ontology Models, Graph Neural Networks, and Large Language Models

The rapid growth of biomedical literature makes it challenging for researchers to stay current. Integrating knowledge from various sources is crucial for studying complex biological systems. Traditional text-mining methods often have limited accuracy because they don’t capture semantic and contextual nuances. Deep-learning models can be computationally expensive and typically have low interpretability, though efforts in explainable AI aim to mitigate this. Furthermore, transformer-based models have a tendency to produce false or made-up information—a problem known as hallucination—which is especially prevalent in large language models (LLMs). This study proposes a hybrid approach combining text-mining techniques with graph neural networks (GNNs) and fine-tuned large language models (LLMs) to extend biomedical knowledge graphs and interpret predicted edges based on published literature. An LLM is used to validate predictions and provide explanations. Evaluated on a corpus of experimentally confirmed protein interactions, the approach achieved a Matthews correlation coefficient (MCC) of 0.772. Applied to insomnia, the approach identified 25 interactions between 32 human proteins absent in known knowledge bases, including regulatory interactions between MAOA and 5-HT2C, binding between ADAM22 and 14-3-3 proteins, which is implicated in neurological diseases, and a circadian regulatory loop involving RORB and NR1D1. The hybrid GNN-LLM method analyzes biomedical literature efficiency to uncover potential molecular interactions for complex disorders. It can accelerate therapeutic target discovery by focusing expert verification on the most relevant automatically extracted information.

Rapid Convolutional Algorithm for the Discovery of Blueberry Honey Authenticity Markers via Nontargeted LC-MS Analysis.

Bees produce honey through the collection and transformation of nectar, whose botanical origin impacts the taste, nutritional value, and, therefore, the market price of the resulting honey. This phenomenon has led some to mislabel their honey so that it can be sold at a higher price. Metabolomics has been gaining popularity in food authentication, but rapid data mining algorithms are needed to facilitate the discovery of new authenticity markers. A nontargeted high-resolution liquid chromatography-mass spectrometry (HR/LC-MS) analysis of 262 monofloral honey samples, of which 50 were blueberry honey, was performed. Data mining methods were demonstrated for the discovery of binary single-markers (compound was only detected in blueberry honey), threshold single-markers (compound had the highest concentration in blueberry honey), and interval ratio-markers (the ratio of two compounds was within a unique interval in blueberry honey). A novel convolutional algorithm was developed for the discovery of interval ratio-markers, which trained 14× faster and achieved a 0.2 Matthews correlation coefficient (MCC) units higher classification score than existing open-source implementations. The convolutional algorithm also had classification performance similar to that of a brute-force search but trained 1521× faster. A pipeline for shortlisting candidate authenticity markers from the LC-MS spectra that may be suitable for chemical structure identification was also demonstrated and led to the identification of niacin as a blueberry honey threshold single-marker. This work demonstrates an end-to-end approach to mine the honey metabolome for novel authenticity markers and can readily be applied to other types of food and analytical chemistry instruments.

Exploring the Application of Machine Learning for Automatic Inbound Email Classification in CRM System at XYZ Company

Customer service has become increasingly crucial in today's business landscape, necessitating companies to provide fast, responsive, and personalized assistance to their clientele. However, amidst the challenges posed by surges in email volume, manual categorization and response strategies often lead to performance declines. To address this, we propose a system leveraging Machine Learning techniques for automated email classification. Our evaluation reveals promising results, with SVM achieving the highest accuracy of 96.59%, followed by XGB (96.02%) and RF (95.27%). These models exhibit commendable precision, recall, F1 scores, and Matthews Correlation Coefficient (MCC), showcasing their effectiveness in improving customer service efficiency and responsiveness. This integration of technology not only enhances operational efficiency but also fosters harmonious customer relationships, ultimately leading to increased loyalty and profitability for companies.

A deep learning model for syncope recognition in the emergency department

Abstract Introduction Syncope is a common presenting symptom in the emergency departments (EDs) and poses challenges in its recognition and differentiation from alternative causes of transient loss of consciousness. Natural Language Processing (NLP) is an artificial intelligence technique that enables the analysis of natural language, such as clinical notes and medical history within electronic medical records (EMRs). The application of NLP into the medical field is surging but its use for the recognition of syncope is still underexplored. Purpose We aimed to develop a NLP-based deep learning model for syncope recognition in the ED. Methods Two models, based on the Italian and Multilingual Bidirectional Encoder Representations from Transformers (BERT) language model, were developed and evaluated using consecutive EMRs from Humanitas Research Hospital ED. The first "triage" model was only based on textual data contained in the "triage" section of the EMR. The second "anamnesis" model also considered textual data contained in the "medical history" section. Performance metrics including area under the curve (AUC), Matthew’s correlation coefficient (MCC) and Brier’s score (BS) were used to compare the discriminative capabilities of different BERT models. Results The models were developed and tested on 15,098 and 15,222 EMRs, respectively. The Multilanguage BERT discriminated patients with syncope with an AUC of 0.85 in the triage model and of 0.92 in the anamnesis model. The Italian BERT identified patients with syncope with a MCC of 0.50 and a BS of 0.02 in the triage model and of 0.64 and 0.01, respectively, in the anamnesis model. Conclusions We propose two NLP-based models that identify syncope patients with a high discriminative capability. The presented algorithm may provide a starting point for the management of syncope patients using machine learning techniques.ROC and calibration curves of modelsFramework for model implementation

Incident atrial fibrillation prediction using ECG-based deep learning at a specialized tertiary cardiac care center

Abstract Background Deep learning (DL) applied to electrocardiograms (ECG) is an emerging modality in the prediction of incident atrial fibrillation or atrial flutter (AF). The generalizability of such methods to a tertiary heart center (THC) population has not yet been formally explored. Purpose Evaluate the performance of DL in predicting 5-year incident AF using a resting 12-lead ECG in sinus rhythm acquired at a THC. Methods In a retrospective study, we examined 1.4 million ECGs acquired from over 250,000 adults at a THC between 2004 and 2022. ECGs were excluded if they were performed within 30 days of cardiac surgery, captured a rhythm other than sinus rhythm, or were acquired in patients with pre-existing AF. Incident AF at 5 years was modelled as a binary outcome and was determined on the basis of available outpatient and inpatient clinical databases and ECG diagnoses. Included ECGs were randomly split by distributing patients into training (70%), validation (10%), and test (20%) sets. A ResNet-50 model was trained using the training set. Hyperparameters were optimized using the validation set. The tuned model's performance is reported on the test set. The results at the patient level were derived by averaging the model's probability outputs for ECGs grouped according to both their AF outcome and the patient's identity. Bootstrapping was used to report confidence intervals (CI). Sensitivity and specificity are reported at a classification threshold based on the Matthews correlation coefficient. Saliency maps were used to enhance the model’s explainability. Results A total of 669,782 ECGs (47% of the screened ECGs) were included among 145,323 patients. Mean age was 63±15 years and 62% were male. The 5-year incident AF outcome was observed in 12% of ECGs and 16% of patients. The performance of the tuned model was first evaluated at the ECG level on the test set demonstrating an area under the receiver operating curve (AUC) of 0.75 (95% CI: 0.745-0.753) and an integrated calibration index of 0.009 (95% CI: 0.008-0.011). When testing the model at the patient level to simulate a deployment scenario, the AUC improved to 0.78 (95% CI: 0.768-0.783) with a sensitivity of 50% (95% CI: 49-52) and specificity of 87% (95% CI: 86.5-87.3). Stratified results by sex showed a better AUC in females, 0.81 (95% CI: 0.80-0.82), compared to males, 0.75 (95% CI: 0.74-0.76). Subgroup analyses revealed a trend of improved performance at the patient level with an increasing number of ECGs. Saliency maps highlighted the P-wave area as having the highest influence on the model’s prediction, thereby enhancing the model’s plausibility and explainability. Conclusion A unimodal ECG-based deep learning model showed promising 5-year incident AF prediction performance in a cohort of all-comer patients at a tertiary heart center. Further studies could explore the use of multimodal prediction models that integrate ECGs with other clinical and imaging data.

Deep Convolutional Transformer Network for Stock Movement Prediction

The prediction and modeling of stock price movements have been shown to possess considerable economic significance within the finance sector. Recently, a range of artificial intelligence methodologies, encompassing both traditional machine learning and deep learning approaches, have been introduced for the purpose of forecasting stock price fluctuations, yielding numerous successful outcomes. Nonetheless, the identification of effective features for predicting stock movements is considered a complex challenge, primarily due to the non-linear characteristics, volatility, and inherent noise present in financial data. This study introduces an innovative Deep Convolutional Transformer (DCT) model that amalgamates convolutional neural networks, Transformers, and a multi-head attention mechanism. It features an inception convolutional token embedding architecture alongside separable fully connected layers. Experiments conducted on the NASDAQ, Hang Seng Index (HSI), and Shanghai Stock Exchange Composite (SSEC) employ Mean Absolute Error (MAE), Mean Square Error (MSE), Mean Absolute Percentage Error (MAPE), accuracy, and Matthews Correlation Coefficient (MCC) as evaluation metrics. The findings reveal that the DCT model achieves the highest accuracy of 58.85% on the NASDAQ dataset with a sliding window width of 30 days. In terms of error metrics, it surpasses other models, demonstrating the lowest average prediction error across all datasets for MAE, MSE, and MAPE. Furthermore, the DCT model attains the highest MCC values across all three datasets. These results suggest a promising capability for classifying stock price trends and affirming the DCT model’s superiority in predicting closing prices.

A Strategy for Predicting the Performance of Supervised and Unsupervised Tabular Data Classifiers

AbstractMachine Learning algorithms that perform classification are increasingly been adopted in Information and Communication Technology (ICT) systems and infrastructures due to their capability to profile their expected behavior and detect anomalies due to ongoing errors or intrusions. Deploying a classifier for a given system requires conducting comparison and sensitivity analyses that are time-consuming, require domain expertise, and may even not achieve satisfactory classification performance, resulting in a waste of money and time for practitioners and stakeholders. This paper predicts the expected performance of classifiers without needing to select, craft, exercise, or compare them, requiring minimal expertise and machinery. Should classification performance be predicted worse than expectations, the users could focus on improving data quality and monitoring systems instead of wasting time in exercising classifiers, saving key time and money. The prediction strategy uses scores of feature rankers, which are processed by regressors to predict metrics such as Matthews Correlation Coefficient (MCC) and Area Under ROC-Curve (AUC) for quantifying classification performance. We validate our prediction strategy through a massive experimental analysis using up to 12 feature rankers that process features from 23 public datasets, creating additional variants in the process and exercising supervised and unsupervised classifiers. Our findings show that it is possible to predict the value of performance metrics for supervised or unsupervised classifiers with a mean average error (MAE) of residuals lower than 0.1 for many classification tasks. The predictors are publicly available in a Python library whose usage is straightforward and does not require domain-specific skill or expertise.

Prospective and External Validation of Machine Learning Models for Short- and Long-Term Mortality in Acutely Admitted Patients Using Blood Tests

Background: Predicting mortality in emergency departments (EDs) using machine learning models presents challenges, particularly in balancing simplicity with performance. This study aims to develop models that are both simple and effective for predicting short- and long-term mortality in ED patients. Our approach uses a minimal set of variables derived from one single blood sample obtained at admission. Methods: Data from three cohorts at two large Danish university hospitals were analyzed, including one retrospective and two prospective cohorts where prognostic models were applied to predict individual mortality risk, spanning the years 2013–2022. Routine biochemistry analyzed in blood samples collected at admission was the primary data source for the prediction models. The outcomes were mortality at 10, 30, 90, and 365 days after admission to the ED. The models were developed using Light Gradient Boosting Machines. The evaluation of mortality predictions involved metrics such as Area Under the Receiver Operating Characteristic Curve (AUC), sensitivity, specificity, negative predictive values, positive predictive values, and Matthews correlation coefficient (MCC). Results: A total of 43,648 unique patients with 65,484 admissions were analyzed. The models showed high accuracy, with very good to excellent AUC values between 0.87 and 0.93 across different time intervals. Conclusions: This study demonstrates that a single assessment of routine clinical biochemistry upon admission can serve as a powerful predictor for both short-term and long-term mortality in ED admissions.

Predicting the Pathway Involvement of All Pathway and Associated Compound Entries Defined in the Kyoto Encyclopedia of Genes and Genomes

Background/Objectives: Predicting the biochemical pathway involvement of a compound could facilitate the interpretation of biological and biomedical research. Prior prediction approaches have largely focused on metabolism, training machine learning models to solely predict based on metabolic pathways. However, there are many other types of pathways in cells and organisms that are of interest to biologists. Methods: While several publications have made use of the metabolites and metabolic pathways available in the Kyoto Encyclopedia of Genes and Genomes (KEGG), we downloaded all the compound entries with pathway annotations available in the KEGG. From these data, we constructed a dataset where each entry contained features representing compounds combined with features representing pathways, followed by a binary label indicating whether the given compound is associated with the given pathway. We trained multi-layer perceptron binary classifiers on variations of this dataset. Results: The models trained on 6485 KEGG compounds and 502 pathways scored an overall mean Matthews correlation coefficient (MCC) performance of 0.847, a median MCC of 0.848, and a standard deviation of 0.0098. Conclusions: This performance on all 502 KEGG pathways represents a roughly 6% improvement over the performance of models trained on only the 184 KEGG metabolic pathways, which had a mean MCC of 0.800 and a standard deviation of 0.021. These results demonstrate the capability to effectively predict biochemical pathways in general, in addition to those specifically related to metabolism. Moreover, the improvement in the performance demonstrates additional transfer learning with the inclusion of non-metabolic pathways.

An Integrated Multi-Model Framework Utilizing Convolutional Neural Networks Coupled with Feature Extraction for Identification of 4mC Sites in DNA Sequences

N4-methylcytosine (4mC) is a chemical modification that occurs on one of the four nucleotide bases in DNA and plays a vital role in DNA expression, repair, and replication. It also actively participates in the regulation of cell differentiation and gene expression. Consequently, it is important to comprehend the role of 4mC in the epigenetic regulation for revealing the complications of the gene expression and their associated governing cellular operations. However, the inherent resource requirements and time constraints of the experimental procedure, present challenges to the cellular culture process. While data-driven methodologies present promising solutions to mitigate the demand for extensive experimental efforts, their performance relies on the suitability and existence of high-quality data. This study presents a multi-model framework that integrates convolutional neural network (CNN) with the distributed k-mer and embedding feature extraction techniques to enhance the identification of 4mC sites in DNA sequences. The integration of k-mers ensures the effective representation of the local sequence patterns, while the utilization of embedding enables a more holistic encoding by considering the broader context and semantics of the sequence data. Following the initial step, the obtained distributed representation of the DNA sequence seamlessly enters the CNN, triggering a crucial convolution operation wherein a set of adaptable filters systematically convolves across the sequence to detect vital local patterns. The proposed integrated multi-model framework was applied to six publicly available datasets and evaluated against the cutting-edge 4mCPred, 4mCCNN, iDNA4mC, Meta-4mCpred, DeepTorrent, 4mCPred-SVM, and DMKL-HFIS methods. The evaluation was based on accuracy, specificity, sensitivity, and Matthews Correlation Coefficient. The results demonstrated that the proposed multi-model framework outperformed the state-of-the-art methods, as well as one-hot encoding and the hybrid of one-hot & TNC features, in accurately identifying 4mC sites.

Prediction of human O-linked glycosylation sites using stacked generalization and embeddings from pre-trained protein language model.

O-linked glycosylation, an essential post-translational modification process in Homo sapiens, involves attaching sugar moieties to the oxygen atoms of serine and/or threonine residues. It influences various biological and cellular functions. While threonine or serine residues within protein sequences are potential sites for O-linked glycosylation, not all serine and/or threonine residues undergo this modification, underscoring the importance of characterizing its occurrence. This study presents a novel approach for predicting intracellular and extracellular O-linked glycosylation events on proteins, which are crucial for comprehending cellular processes. Two base multi-layer perceptron models were trained by leveraging a stacked generalization framework. These base models respectively employ ProtT5 and Ankh O-linked glycosylation site-specific embeddings whose combined predictions are used to train the meta-multi-layer perceptron model. Trained on extensive O-linked glycosylation datasets, the stacked-generalization model demonstrated high predictive performance on independent test datasets. Furthermore, the study emphasizes the distinction between nucleocytoplasmic and extracellular O-linked glycosylation, offering insights into their functional implications that were overlooked in previous studies. By integrating the protein language model's embedding with stacked generalization techniques, this approach enhances predictive accuracy of O-linked glycosylation events and illuminates the intricate roles of O-linked glycosylation in proteomics, potentially accelerating the discovery of novel glycosylation sites. Stack-OglyPred-PLM produces Sensitivity, Specificity, Matthews Correlation Coefficient, and Accuracy of 90.50%, 89.60%, 0.464, and 89.70%, respectively on a benchmark NetOGlyc-4.0 independent test dataset. These results demonstrate that Stack-OglyPred-PLM is a robust computational tool to predict O-linked glycosylation sites in proteins. The developed tool, programs, training, and test dataset are available at https://github.com/PakhrinLab/Stack-OglyPred-PLM. Supplementary data are available at Bioinformatics online.

G4 & the balanced metric family – a novel approach to solving binary classification problems in medical device validation & verification studies

BackgroundIn medical device validation and verification studies, the area under the receiver operating characteristic curve (AUROC) is often used as a primary endpoint despite multiple reports showing its limitations. Hence, researchers are encouraged to consider alternative metrics as primary endpoints. A new metric called G4 is presented, which is the geometric mean of sensitivity, specificity, the positive predictive value, and the negative predictive value. G4 is part of a balanced metric family which includes the Unified Performance Measure (also known as P4) and the Matthews’ Correlation Coefficient (MCC). The purpose of this manuscript is to unveil the benefits of using G4 together with the balanced metric family when analyzing the overall performance of binary classifiers.ResultsSimulated datasets encompassing different prevalence rates of the minority class were analyzed under a multi-reader-multi-case study design. In addition, data from an independently published study that tested the performance of a unique ultrasound artificial intelligence algorithm in the context of breast cancer detection was also considered. Within each dataset, AUROC was reported alongside the balanced metric family for comparison. When the dataset prevalence and bias of the minority class approached 50%, all three balanced metrics provided equivalent interpretations of an AI’s performance. As the prevalence rate increased / decreased and the data became more imbalanced, AUROC tended to overvalue / undervalue the true classifier performance, while the balanced metric family was resistant to such imbalance. Under certain circumstances where data imbalance was strong (minority-class prevalence < 10%), MCC was preferred for standalone assessments while P4 provided a stronger effect size when evaluating between-groups analyses. G4 acted as a middle ground for maximizing both standalone assessments and between-groups analyses.ConclusionsUse of AUROC as the primary endpoint in binary classification problems provides misleading results as the dataset becomes more imbalanced. This is explicitly noticed when incorporating AUROC in medical device validation and verification studies. G4, P4, and MCC do not share this limitation and paint a more complete picture of a medical device’s performance in a clinical setting. Therefore, researchers are encouraged to explore the balanced metric family when evaluating binary classification problems.

Identifying cardiovascular disease risk in the U.S. population using environmental volatile organic compounds exposure: A machine learning predictive model based on the SHAP methodology

BackgroundCardiovascular disease (CVD) remains a leading cause of mortality globally. Environmental pollutants, specifically volatile organic compounds (VOCs), have been identified as significant risk factors. This study aims to develop a machine learning (ML) model to predict CVD risk based on VOC exposure and demographic data using SHapley Additive exPlanations (SHAP) for interpretability. MethodsWe utilized data from the National Health and Nutrition Examination Survey (NHANES) from 2011 to 2018, comprising 5098 participants. VOC exposure was assessed through 15 urinary metabolite metrics. The dataset was split into a training set (70 %) and a test set (30 %). Six ML models were developed, including Random Forest (RF), Light Gradient Boosting Machine (LightGBM), Decision Tree (DT), Extreme Gradient Boosting (XGBoost), Multi-Layer Perceptron (MLP), and Support Vector Machines (SVM). Model performance was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUROC), accuracy, balanced accuracy, F1 score, J-index, kappa, Matthew's correlation coefficient (MCC), positive predictive value (PPV), negative predictive value (NPV), sensitivity (sens), specificity (spec) and SHAP was applied to interpret the best-performing model. ResultsThe RF model exhibited the highest predictive performance with an ROC of 0.8143. SHAP analysis identified age and ATCA as the most significant predictors, with ATCA showing a protective effect against CVD, particularly in older adults and those with hypertension. The study found a significant interaction between ATCA levels and age, indicating that the protective effect of ATCA is more pronounced in older individuals due to increased oxidative stress and inflammatory responses associated with aging. E-values analysis suggested robustness to unmeasured confounders. ConclusionsThis study is the first to utilize VOC exposure data to construct an ML model for predicting CVD risk. The findings highlight the potential of combining environmental exposure data with demographic information to enhance CVD risk prediction, supporting the development of personalized prevention and intervention strategies.

Protein allosteric site identification using machine learning and per amino acid residue reported internal protein nanoenvironment descriptors

Allosteric regulation plays a crucial role in modulating protein functions and represents a promising strategy in drug development, offering enhanced specificity and reduced toxicity compared to traditional active site inhibition. Existing computational methods for predicting allosteric sites on proteins often rely on static protein surface pocket features, normal mode analysis or extensive molecular dynamics simulations encompassing both the protein function modulator and the protein itself. In this study, we introduce an innovative methodology that employs a per amino acid residue classifier to distinguish allosteric site-forming residues (AFRs) from non-allosteric, or free residues (FRs). Our model, STINGAllo, exhibits robust performance, achieving Distance Center Center (DCC) success rate when all AFRs were predicted within pockets identified by FPocket, overall DCC, F1 score and a Matthews correlation coefficient (MCC) of 78 %, 60 %, 64 % and 64 % respectively. Furthermore, we identified key descriptors that characterize the internal protein nanoenvironment of AFRs, setting them apart from FRs. These descriptors include the sponge effect, distance to the protein centre of geometry (cg), hydrophobic interactions, electrostatic potentials, eccentricity, and graph bottleneck features.