Test Set Research Articles

Reviewing Open Data Semantic 3D City Models to Develop Novel 3D Reconstruction Methods

Abstract. Structured semantic 3D city models are pivotal in creating urban 3D digital twins. The wide adoption of such models has been primarily enabled by robust, model-based, and automatic 3D reconstruction methods. However, these methods impose requirements on the reconstruction, mainly restricting the solution space to several model types and relying on accurate 2D footprints. Recent research shows that deep-learning-based methods promise highly generic solution space and are footprint-free. Yet, the current training and test datasets are limited, hindering the methods’ development. In this work, we analyze the ubiquity of already existing, open 3D city model datasets and their potential to serve as a large-scale training and test set for 3D reconstruction, where 27 potential dataset collections have been identified. Our review shows that more than 215 million building models are readily available. We firmly believe that this review will facilitate further research on robust automatic 3D city model reconstruction and serve as a reference for benchmarking 3D city models.

ScCaT: An explainable capsulating architecture for sepsis diagnosis transferring from single-cell RNA sequencing.

Sepsis is a life-threatening condition characterized by an exaggerated immune response to pathogens, leading to organ damage and high mortality rates in the intensive care unit. Although deep learning has achieved impressive performance on prediction and classification tasks in medicine, it requires large amounts of data and lacks explainability, which hinder its application to sepsis diagnosis. We introduce a deep learning framework, called scCaT, which blends the capsulating architecture with Transformer to develop a sepsis diagnostic model using single-cell RNA sequencing data and transfers it to bulk RNA data. The capsulating architecture effectively groups genes into capsules based on biological functions, which provides explainability in encoding gene expressions. The Transformer serves as a decoder to classify sepsis patients and controls. Our model achieves high accuracy with an AUROC of 0.93 on the single-cell test set and an average AUROC of 0.98 on seven bulk RNA cohorts. Additionally, the capsules can recognize different cell types and distinguish sepsis from control samples based on their biological pathways. This study presents a novel approach for learning gene modules and transferring the model to other data types, offering potential benefits in diagnosing rare diseases with limited subjects.

Personalized three-year survival prediction and prognosis forecast by interpretable machine learning for pancreatic cancer patients: a population-based study and an external validation.

The overall survival of patients with pancreatic cancer is extremely low. We aimed to establish machine learning (ML) based model to accurately predict three-year survival and prognosis of pancreatic cancer patients. We analyzed pancreatic cancer patients from the Surveillance, Epidemiology, and End Results (SEER) database between 2000 and 2021. Univariate and multivariate logistic analysis were employed to select variables. Recursive Feature Elimination (RFE) method based on 6 ML algorithms was utilized in feature selection. To construct predictive model, 13 ML algorithms were evaluated by area under the curve (AUC), area under precision-recall curve (PRAUC), accuracy, sensitivity, specificity, precision, cross-entropy, Brier scores and Balanced Accuracy (bacc) and F Beta Score (fbeta). An optimal ML model was constructed to predict three-year survival, and the predictive results were explained by SHapley Additive exPlanations (SHAP) framework. Meanwhile, 101 ML algorithm combinations were developed to select the best model with highest C-index to predict prognosis of pancreatic cancer patients. A total of 20,064 pancreatic cancer patients from SEER database was consecutively enrolled. We utilized eight clinical variables to establish prediction model for three-year survival. CatBoost model was selected as the best prediction model, and AUC was 0.932 [0.924, 0.939], 0.899 [0.873, 0.934] and 0.826 [0.735, 0.919] in training, internal test and external test sets, with 0.839 [0.831, 0.847] accuracy, 0.872 [0.858, 0.887] sensitivity, 0.803 [0.784, 0.825] specificity and 0.832 [0.821, 0.853] precision. Surgery type had the greatest effects on three-year survival according to SHAP results. For prognosis prediction, "RSF+GBM" algorithm was the best prognostic model with C-index of 0.774, 0.722 and 0.674 in training, internal test and external test sets. Our ML models demonstrate excellent accuracy and reliability, offering more precise personalized prognostic prediction to pancreatic cancer patients.

Determining Rock Joint Peak Shear Strength Based on GA-BP Neural Network Method

The peak shear strength of a rock joint is an important indicator in rock engineering, such as mining and sloping. Therefore, direct shear tests were conducted using an RDS-200 rock direct shear apparatus, and the related data such as normal stress, roughness, size, normal loading rate, basic friction angle, and JCS were collected. A peak shear strength prediction model for rock joints was established, by which a predicted rock joint peak shear strength can be obtained by inputting the influencing factors. Firstly, the study used the correlation analysis method to find out the correlation coefficient between the above factors and rock joint peak shear strength to provide a reference for factor selection of the peak shear strength prediction model. Then, the JRC-JCS model and four established GA-BP neural network models were studied to identify the most valuable rock joint peak shear strength prediction method. The GA-BP neural network models used a genetic algorithm to optimize the BP neural network with different input factors to predict rock joint peak shear strength, after dividing the selected data into 80% training set and 20% test set. The results show that the error of the JRC-JCS model is a little bigger, with a value of 11.2%, while the errors of the established GA-BP neural network models are smaller than 6%, which indicates that the four established GA-BP neural network models can well fit the relationship between the peak shear strength and selected input factors. Additionally, increasing the factor number of the input layer can effectively improve the prediction accuracy of the GA-BP neural network models, and the prediction accuracy of the GA-BP neural network models will be higher if factors that have higher correlation with the output results are used as input factors.

A multi-model approach integrating whole-slide imaging and clinicopathologic features to predict breast cancer recurrence risk.

Breast cancer is the most common malignancy affecting women worldwide and is notable for its morphologic and biologic diversity, with varying risks of recurrence following treatment. The Oncotype DX Breast Recurrence Score test is an important predictive and prognostic genomic assay for estrogen receptor positive/HER2 negative breast cancer that guides therapeutic strategies; however, such tests can be expensive, delay care, and are not widely available. The aim of this study was to develop a multi-model approach integrating the analysis of whole-slide images and clinicopathologic data to predict their associated breast cancer recurrence risks and categorize these patients into two risk groups according to the predicted score: low-risk and high-risk. The proposed novel methodology uses convolutional neural networks for feature extraction and vision transformers for contextual aggregation, complemented by a logistic regression model that analyzes clinicopathologic data for classification into two risk categories. This method was trained and tested on 956 hematoxylin and eosin-stained whole-slide images of 950 ER+/HER2- breast cancer patients with corresponding clinicopathological features that had prior Oncotype DX testing. The model's performance was evaluated using an internal test set of 192 patients from Dartmouth Health and an external test set of 405 patients from the University of Chicago. The multi-model approach achieved an AUC of 0.91 (95% CI: 0.87-0.95) on the internal set and an AUC of 0.84 (95% CI: 0.78-0.89) on the external cohort for predicting low- and high-breast cancer recurrence risk categories based on the Oncotype DX recurrence score. With further validation, the proposed methodology could provide an alternative to assist clinicians in personalizing treatment for breast cancer patients and potentially improving their outcomes.

Artificial intelligence applied to development of predictive stability model for intracranial aneurysms

BackgroundWe aimed to develop multiple machine learning models to predict the risk of early intracranial aneurysms (IAs) rupture, evaluate and compare the performance of predictive models.MethodsInformation related to patients diagnosed with IA by CT angiography and clinicians in Central hospital of Dalian University of Technology from January 2010 to June 2022 was collected, including clinical characteristics, blood indicators and IA morphological parameters. IA with rupture or maximum growth ≥ 0.5 mm within 1 month of first diagnosis was considered unstable. The relevant factors affecting IA stability were screened and predictive models were developed based on the above three levels, including random forest (RF), support vector machine (SVM), and artificial neural network (ANN). Sensitivity, specificity, accuracy and area under curve (AUC) value were used to evaluate the predictive models.ResultsA total of 989 IA patients were included in the study, including 561 stable patients and 428 unstable patients. For RF models, the training set showed that sensitivity, specificity, accuracy and the AUC values were 72.8–83.7%, 76.9–86.9%, 75.1–84.1% and 0.748 (0.719–0.778)–0.839 (0.814–0.864), respectively; after test set validation, the results were 71.9–78.8%, 75.0–84.0%, 73.6–81.1% and 0.734 (0.688–0.781)–0.809 (0.768–0.850), respectively. For SVM models, the training set were 66.0–80.2%, 76.5–85.5%, 71.7–82.3%, 0.712 (0.682–0.743)–0.913 (0.884–0.924), respectively; the test set were 44.2–78.3%, 63.4–84.4%, 57.9–80.9%, 0.699 (0.651–0.747)–0.806 (0.765–0.848), respectively. For ANN models, the training set were 66.8–83.0%, 75.3–82.3%, 71.6–82.1%, 0.783 (0.757–0.808)–0.897 (0.879–0.914); the test set were 63.1–76.3%, 65.5–84.0%, 64.4–80.6%, 0.680 (0.593–0.694)–0.860 (0.821–0.899). The results of variable importance showed that age, white blood cell count (WBC) and uric acid (UA) played an important role in predicting the stability of IA.ConclusionsThe predictive stability models of IA based on three artificial intelligence methods shows good clinical application. Age, WBC and UA played an important role in predicting the IA stability, and were potentially important predictors.

A multitask framework with self-attention mechanism for simultaneous detection of axial pressure, shear deformation, and rupture damage in rubber bearings

Base-isolated structures can effectively control structural response, of which rubber bearings are key components. To ensure the proper functioning of rubber bearings, it is important to accurately detect their state. In previous studies, a smart rubber bearing has been proposed to monitor the axial pressure, shear deformation, and rupture damage of rubber bearings. However, owing to the small dataset and limited ability of the prediction model, two problems exist in previous research: (1) the axial pressure, shear deformation, and rupture damage cannot be detected simultaneously and (2) the detection accuracy is not sufficiently high for practical applications. This study solves these problems in three aspects. First, it addresses the issue of limited data by researching and comparing three data augmentation methods. Second, a multitask framework with a self-attention mechanism is established to simultaneously detect axial compression, shear deformation, and rupture damage. Finally, we introduce a two-stage optimization method based on Bayesian optimization and a grid search for the network structure and optimization. The root-mean-square errors for predicting the axial pressure and shear deformation in the test set are 0.19 MPa and 2.04%, which reduced the error by up to 56.8% and 73.0%, respectively, compared to conventional deep neural network models. In addition, the accuracy of rupture damage detection reached 100%.

Can Patient Characteristics and Pre-Treatment MRI Features Predict Survival After Stereotactic Ablative Radiotherapy (SABR) Treatment in Hepatocellular Carcinoma (HCC): Preliminary Assessment.

The study purpose was to develop a machine learning (ML)-based predictive model for event-free survival (EFS) in patients with hepatocellular carcinoma (HCC) undergoing stereotactic ablative radiotherapy (SABR). Patients receiving SABR for HCC at a single institution, between 2017 and 2020, were included in the study. They were split into training and test (85%:15%) cohorts. Events of interest were HCC recurrence or death. Three ML models were trained, the features were selected, and the hyperparameters were tuned. The performance was measured using Harrell's C index with the best-performing model being tested on the unseen cohort. Overall, 41 patients were included (training = 34, test = 7) and 64 lesions were analysed (training = 50, test = 14), resulting in 30 events (60% rate) in the training set (death = 6, recurrence = 24) and 8 events (57% rate) in the test set (death = 5, recurrence = 3). A Cox regression model, using age at treatment, albumin, and intra-lesional fat identified through MRI as variables, had the best performance with a mean training score of 0.78 (standard deviation (SD) 0.02), a mean validation of 0.78 (SD 0.18), and a test score of 0.94. Predicting the outcomes in patients with HCC, following SABR, using a novel model is feasible and warrants further evaluation.

Large-scale structural covariance networks changes relate to executive function deficit in betel quid-dependent chewers.

Previous studies demonstrate deficits in executive function for betel quid-dependent (BQD) patients. Large-scale structural covariance network (SCN) based on gray matter (GM) morphometry may be able to explore the neural mechanism of executive dysfunction in BQD individuals. This study aims to identify spatial covariance patterns of GM volume and to investigate their association with executive dysfunction in BQD individuals. Sixty-four BQD individuals and 48 sex- and age-matched healthy controls (HCs) underwent T1-weighted structural MRI examination and executive function assessments, including the Backward Digit Span (BDS) test and Stroop Color and Word (SCW) test. Seventy SCNs based on GM volume covariance patterns were defined using independent component analysis. An SCN-based classifier was constructed to differentiate between BQD and HC individuals. Receiver operating characteristic (ROC) curves were applied to evaluate the performance of the SCN-based classifier. Linear regression analyses were performed to investigate the association between SCN network indices and executive function indices. Six SCNs had higher classifications for differentiating between BQD and HC individuals. The area under the ROC curve of the SCN-based classifier was 0.812 in the training set and 0.771 in the testing set. Furthermore, linear regression analyses demonstrated that the network indices in the thalamus were associated with BDS scores adjusted for age, sex, and education. Large-scale SCNs could provide potential imaging markers for differentiating BQD and HC groups. The loss of network index in the thalamus was associated with working memory, indicating that SCNs could reveal executive dysfunction in BQD individuals.

Intelligent Adaptive Networks for Chaos Analysis in Stochastic SIS Differential Epidemic Model with the Impact of Unpredictable Environmental Factors

In this study, an intelligent computing framework is presented to explore the influence of unpredictable environmental factors on the spread of infections. A stochastic non-linear SIS epidemic model (SISEM) is examined that incorporates a non-linear incidence rate, and impact of random fluctuations on the disease transmission. The knacks of artificial neural networks Levenberg–Marquardt (ANNLMA) technique are exploited to illustrate the behavioral complexities of stochastic non-linear SISEM, which explores the dynamics of disease transmission in a stochastic environment and provides insights into the interactions between susceptible and infectious populations. The Milstein approach of order 1 generates a dataset for the proposed ANNLMA technique by varying integrated parameters, such as the rates of recovery, death, disease transmission, noise standard deviation and the entire population, for various SISEM scenarios. The referenced dataset is randomly divided into test, validation and train sets for the weight learning of the proposed ANNLMA. The viability of the ANNLMA is measured using intensive simulation-based investigations through analysis of mean squared errors, histograms, regression and autocorrelation studies. The investigations highlighted that the ANNLMA results closely match the reference results, illustrating convergence within the 10[Formula: see text] to 10[Formula: see text] range, indicating the validity of the proposed methodology.

AI-DPAPT: a machine learning framework for predicting PROTAC activity.

Proteolysis Targeting Chimeras are part of targeted protein degradation (TPD) techniques, which are significant for pharmacological and therapy development. Small-molecule interaction with the targeted protein is a complicated endeavor and a challenge to predict the proteins accurately. This study used machine learning algorithms and molecular fingerprinting techniques to build an AI-powered PROTAC Activity Prediction Tool that could predict PROTAC activity by examining chemical structures. The chemical structures of a diverse set of PROTAC drugs and their corresponding activities are selected as a dataset for training the tool. The processes used in this study included data preparation, feature extraction, and model training. Further, evaluation was done for the performance of the various classifiers, such as AdaBoost, Support Vector Machine, Random Forest, Gradient Boosting, and Multi-Layer Perceptron. The findings show that the methods selected here depict accurate PROTAC activities. All the models in this study showed an ROC curve better than 0.9, while the random forest on the test set of the AI-DPAPT had an area under the curve score of 0.97, thus showing accurate results. Furthermore, the study revealed significant insights into the molecular features that can influence the functions of the PROTAC. These findings can potentially increase the understanding of the structure-activity correlations involved in the TPD. Overall, the investigation contributes to computational drug development by introducing this platform powered by artificial intelligence that predicts the function of PROTAC. In addition, it sped up the processes of identifying and improving previously unknown medications. The AI-DPAPT platform can be accessed online using a web server at https://ai-protac.streamlit.app/ .

Coronary heart disease evaluation using PCAT radiomics model based on coronary CT angiography and pericoronary adipose tissue.

To explore the clinical application value of radiomics model based on pericoronary adipose tissue (PCAT) in predicting coronary heart disease. A retrospective analysis was performed for inpatients who had undergone coronary computed tomography angiography from January to December 2023, and 164 cases of coronary artery lesions were screened as the lesion group and 190 cases of normal coronary artery samples were selected as the control group. The clinical data and imaging data of all patients were collected, the radiomics features were extracted by relevant software, and the "region of interest" of pericoronary fat was delineated, and the selection operator and multivariate logistic regression were used to screen the radiomic features of pericoronary fat. A coronary heart disease evaluation model was constructed by the best radiomics features. Area under the curve values of the PCAT radiomics scoring model for predicting the receiver operating characteristic curve of coronary heart disease were 0.863 and 0.851 in training and test sets, respectively. After calibration curve analysis, PCAT radiomics scoring model has a high consistency between the predictive evaluation results and the actual results of coronary heart disease events. In addition, in the training set, the PCAT radiomics scoring model has a net benefit on all threshold probabilities. In the test set, the model has a negative net return with only a small number of threshold probabilities. After combining the clinical characteristics model, the evaluation accuracy of the model for coronary heart disease can reach 0.896. PCAT radiomics model based on coronary computed tomography angiography can effectively predict and evaluate coronary heart disease, which is of great value for the clinical diagnosis of coronary artery disease.

Deep learning-based surgical step recognition for laparoscopic right-sided colectomy.

Understanding the complex anatomy and surgical steps involved in laparoscopic right-sided colectomy (LAP-RC) is essential for standardizing the surgical procedure. Deep-learning (DL)-based computer vision can achieve this. This study aimed to develop a step recognition model for LAP-RC using a dataset of surgical videos with annotated step information and evaluate its recognition performance. This single-center retrospective study utilized a video dataset of laparoscopic ileocecal resection (LAP-ICR) and laparoscopic right-sided hemicolectomy (LAP-RHC) for right-sided colon cancer performed between January 2018 and March 2022. The videos were split into still images, which were divided into training, validation, and test sets using 66%, 17%, and 17% of the data, respectively. Videos were manually classified into eight main steps: 1) medial mobilization, 2) central vascular ligation, 3) dissection of the superior mesenteric vein, 4) retroperitoneal mobilization, 5) lateral mobilization, 6) cranial mobilization, 7) mesocolon resection, and 8) intracorporeal anastomosis. In a simpler version, consecutive surgical steps were combined, resulting in five steps. Precision, recall, F1 scores, and overall accuracy were assessed to evaluate the model's performance in the surgical step classification task. Seventy-eight patients were included; LAP-ICR and LAP-RHC were performed in 35 (44%) and 44 (56%) patients, respectively. The overall accuracy was 72.1% and 82.9% for the eight-step and combined five-step classification tasks, respectively. The automatic surgical step-recognition model for LAP-RCs, developed using a DL algorithm, exhibited a fairly high classification performance. A model that understands the complex steps of LAP-RC will aid the standardization of the surgical procedure.

The efficacy of a food supplement in the treatment of tinnitus with comorbid headache: A Statistical and Machine Learning analysis with a literature review.

Introduction Tinnitus, the perception of sound without an external auditory stimulus, affects approximately 10% to 15% of the population and is often associated with significant comorbidities such as headaches. These conditions can severely impact the quality of life. The aim of this study was to evaluate the efficacy of a food supplement in reducing the symptoms of both tinnitus and headache in patients experiencing these conditions concurrently. Methods This prospective study included 32 patients (21 males and 11 females) aged between 23 and 68 years (mean age 49.38 years) who were experiencing both tinnitus and headache. The study assessed the impact of a food supplement on tinnitus and headache over a 90-day treatment period using three main instruments: the Tinnitus Handicap Inventory (THI), the Headache Impact Test (HIT-6), and a Visual Analog Scale (VAS) for discomfort. Statistical analyses, including paired t-tests, were conducted to compare pre- and post-treatment scores. In the same dataset, Ridge Regression, a linear regression model with L2 regularization, was used to predict post-treatment scores (THI90, HIT90, VAS90). Results The results indicated a statistically significant reduction in all three measures after 90 days of treatment. The mean THI score decreased from 29.81 to 27.06 (p = 0.011), the mean HIT-6 score decreased from 50.41 to 48.75 (p = 0.019), and the mean VAS score for discomfort decreased from 7.63 to 7.13 (p = 0.033). The optimal Ridge Regression model was found with an `alpha` value of approximately 3.73. The performance metrics on the test set were as follows: Mean Squared Error (MSE) of 13.91 and an R-squared score of 0.61, indicating that the model explains approximately 61% of the variance in the post-treatment scores. These results indicate that pre-treatment scores are significant predictors of post-treatment outcomes, and gender plays a notable role in predicting HIT and VAS scores post-treatment. Conclusion This study demonstrates that a food supplement is effective in reducing the symptoms of tinnitus and headache in patients suffering from both conditions. The significant improvements in THI, HIT-6, and VAS scores indicate a positive impact on patient quality of life. Further research with larger sample sizes and more detailed subgroup analyses is recommended to fully understand the differential impacts of treatment across various demographics.

DNA methylation at AHRR as a master predictor of smoke exposure and a biomarker for sleep and exercise

BackgroundDNA methylation profiling may provide a more accurate measure of the smoking status than self-report and may be useful in guiding clinical interventions and forensic investigations. In the current study, blood DNA methylation profiles of nearly 800 Polish individuals were assayed using Illuminia EPIC and the inference of smoking from epigenetic data was explored. In addition, we focused on the role of the AHRR gene as a top marker for smoking and investigated its responsiveness to other lifestyle behaviors.ResultsWe found > 450 significant CpGs associated with cigarette consumption, and overrepresented in various biological functions including cell communication, response to stress, blood vessel development, cell death, and atherosclerosis. The model consisting of cg05575921 in AHRR (p = 4.5 × 10–32) and three additional CpGs (cg09594361, cg21322436 in CNTNAP2 and cg09842685) was able to predict smoking status with a high accuracy of AUC = 0.8 in the test set. Importantly, a gradual increase in the probability of smoking was observed, starting from occasional smokers to regular heavy smokers. Furthermore, former smokers displayed the intermediate DNA methylation profiles compared to current and never smokers, and thus our results indicate the potential reversibility of DNA methylation after smoking cessation. The AHRR played a key role in a predictive analysis, explaining 21.5% of the variation in smoking. In addition, the AHRR methylation was analyzed for association with other modifiable lifestyle factors, and showed significance for sleep and physical activity. We also showed that the epigenetic score for smoking was significantly correlated with most of the epigenetic clocks tested, except for two first-generation clocks.ConclusionsOur study suggests that a more rapid return to never-smoker methylation levels after smoking cessation may be achievable in people who change their lifestyle in terms of physical activity and sleep duration. As cigarette smoking has been implicated in the literature as a leading cause of epigenetic aging and AHRR appears to be modifiable by multiple exogenous factors, it emerges as a promising target for intervention and investment.

Comparative Analysis of Machine Learning Models for Predicting Student Success in Online Programming Courses: A Study Based on LMS Data and External Factors

Early prediction of student performance in online programming courses is essential for implementing timely interventions to enhance academic outcomes. This study aimed to predict academic success by comparing four machine learning models: Logistic Regression, Random Forest, Support Vector Machine (SVM), and Neural Network (Multilayer Perceptron, MLP). We analyzed data from the Moodle Learning Management System (LMS) and external factors of 591 students enrolled in online object-oriented programming courses at the Universidad Estatal de Milagro (UNEMI) between 2022 and 2023. The data were preprocessed to address class imbalance using the synthetic minority oversampling technique (SMOTE), and relevant features were selected based on Random Forest importance rankings. The models were trained and optimized using Grid Search with cross-validation. Logistic Regression achieved the highest Area Under the Receiver Operating Characteristic Curve (AUC-ROC) on the test set (0.9354), indicating strong generalization capability. SVM and Neural Network models performed adequately but were slightly outperformed by the simpler models. These findings suggest that integrating LMS data with external factors enhances early prediction of student success. Logistic Regression is a practical and interpretable tool for educational institutions to identify at-risk students, and to implement personalized interventions.

Ensembling methods for protein-ligand binding affinity prediction

Protein-ligand binding affinity prediction is a key element of computer-aided drug discovery. Most of the existing deep learning methods for protein-ligand binding affinity prediction utilize single models and suffer from low accuracy and generalization capability. In this paper, we train 13 deep learning models from combinations of 5 input features. Then, we explore all possible ensembles of the trained models to find the best ensembles. Our deep learning models use cross-attention and self-attention layers to extract short and long-range interactions. Our method is named Ensemble Binding Affinity (EBA). EBA extracts information from various models using different combinations of input features, such as simple 1D sequential and structural features of the protein-ligand complexes rather than 3D complex features. EBA is implemented to accurately predict the binding affinity of a protein-ligand complex. One of our ensembles achieves the highest Pearson correlation coefficient (R) value of 0.914 and the lowest root mean square error (RMSE) value of 0.957 on the well-known benchmark test set CASF2016. Our ensembles show significant improvements of more than 15% in R-value and 19% in RMSE on both well-known benchmark CSAR-HiQ test sets over the second-best predictor named CAPLA. Furthermore, the superior performance of the ensembles across all metrics compared to existing state-of-the-art protein-ligand binding affinity prediction methods on all five benchmark test datasets demonstrates the effectiveness and robustness of our approach. Therefore, our approach to improving binding affinity prediction between proteins and ligands can contribute to improving the success rate of potential drugs and accelerate the drug development process.

GPT-Driven Radiology Report Generation with Fine-Tuned Llama 3.

The integration of deep learning into radiology has the potential to enhance diagnostic processes, yet its acceptance in clinical practice remains limited due to various challenges. This study aimed to develop and evaluate a fine-tuned large language model (LLM), based on Llama 3-8B, to automate the generation of accurate and concise conclusions in magnetic resonance imaging (MRI) and computed tomography (CT) radiology reports, thereby assisting radiologists and improving reporting efficiency. A dataset comprising 15,000 radiology reports was collected from the University of Medicine and Pharmacy of Craiova's Imaging Center, covering a diverse range of MRI and CT examinations made by four experienced radiologists. The Llama 3-8B model was fine-tuned using transfer-learning techniques, incorporating parameter quantization to 4-bit precision and low-rank adaptation (LoRA) with a rank of 16 to optimize computational efficiency on consumer-grade GPUs. The model was trained over five epochs using an NVIDIA RTX 3090 GPU, with intermediary checkpoints saved for monitoring. Performance was evaluated quantitatively using Bidirectional Encoder Representations from Transformers Score (BERTScore), Recall-Oriented Understudy for Gisting Evaluation (ROUGE), Bilingual Evaluation Understudy (BLEU), and Metric for Evaluation of Translation with Explicit Ordering (METEOR) metrics on a held-out test set. Additionally, a qualitative assessment was conducted, involving 13 independent radiologists who participated in a Turing-like test and provided ratings for the AI-generated conclusions. The fine-tuned model demonstrated strong quantitative performance, achieving a BERTScore F1 of 0.8054, a ROUGE-1 F1 of 0.4998, a ROUGE-L F1 of 0.4628, and a METEOR score of 0.4282. In the human evaluation, the artificial intelligence (AI)-generated conclusions were preferred over human-written ones in approximately 21.8% of cases, indicating that the model's outputs were competitive with those of experienced radiologists. The average rating of the AI-generated conclusions was 3.65 out of 5, reflecting a generally favorable assessment. Notably, the model maintained its consistency across various types of reports and demonstrated the ability to generalize to unseen data. The fine-tuned Llama 3-8B model effectively generates accurate and coherent conclusions for MRI and CT radiology reports. By automating the conclusion-writing process, this approach can assist radiologists in reducing their workload and enhancing report consistency, potentially addressing some barriers to the adoption of deep learning in clinical practice. The positive evaluations from independent radiologists underscore the model's potential utility. While the model demonstrated strong performance, limitations such as dataset bias, limited sample diversity, a lack of clinical judgment, and the need for large computational resources require further refinement and real-world validation. Future work should explore the integration of such models into clinical workflows, address ethical and legal considerations, and extend this approach to generate complete radiology reports.

Diagnostic performance of single-lead electrocardiograms for arterial hypertension diagnosis: a machine learning approach.

Awareness and early identification of hypertension is crucial in reducing the burden of cardiovascular disease (CVD). Artificial intelligence-based analysis of 12-lead electrocardiograms (ECGs) can already detect arrhythmias and hypertension. We performed an observational two-center study in order to develop a machine learning algorithm to proactively detect arterial hypertension from single-lead ECGs. This could serve as proof of concept with an eye towards todays wearables that record single-lead ECGs. In a prospective observational study, we enrolled 1254 consecutive subjects (539 male, aged 60.22 ± 12.46 years), with and without essential hypertension, and no indications of CVD. A 12-lead ECG of 10 seconds duration in resting position was performed on each subject using a digital electrocardiograph and lead I was isolated for analysis using a calibrated Random Forest (RF). Our RF model classified hypertensive from normotensive subjects on a hold-out test set, with 75% accuracy, ROC/AUC 0.831 (95%CI: 0.781-0.871), sensitivity 72%, and specificity 82% (sensitivity and specificity calculated using a threshold of 0.675). Increasing age, larger values of body mass index, the area under the T wave divided by the QRS complex area, and the area under QRS segment adjusted for BMI, were the four most important features that drove the classification decisions of our model. This study demonstrates the potential to opportunistically detect an undiagnosed hypertension, using a single-lead ECG. While studies with data from wearables are required to translate our findings to actual smartwatch settings, our results could pave the way to innovative technologies for hypertension awareness.

MD-HIT: Machine learning for material property prediction with dataset redundancy control

Materials datasets usually contain many redundant (highly similar) materials due to the tinkering approach historically used in material design. This redundancy skews the performance evaluation of machine learning (ML) models when using random splitting, leading to overestimated predictive performance and poor performance on out-of-distribution samples. This issue is well-known in bioinformatics for protein function prediction, where tools like CD-HIT are used to reduce redundancy by ensuring sequence similarity among samples greater than a given threshold. In this paper, we survey the overestimated ML performance in materials science for material property prediction and propose MD-HIT, a redundancy reduction algorithm for material datasets. Applying MD-HIT to composition- and structure-based formation energy and band gap prediction problems, we demonstrate that with redundancy control, the prediction performances of the ML models on test sets tend to have relatively lower performance compared to the model with high redundancy, but better reflect models’ true prediction capability.