ROC-AUC Score Research Articles

Multisensor fusion-based digital twin for localized quality prediction in robotic laser-directed energy deposition

Early detection of defects, such as keyhole pores and cracks is crucial in laser-directed energy deposition (L-DED) additive manufacturing (AM) to prevent build failures. However, the complex melt pool behaviour cannot be adequately captured by conventional single-modal process monitoring approaches. This study introduces a multisensor fusion-based digital twin (MFDT) for localized quality prediction in the robotic L-DED process. The data used in multisensor fusion includes features extracted from a coaxial melt pool vision camera, a microphone, and an off-axis short wavelength infrared thermal camera. The key novelty of this work is a spatiotemporal data fusion method that synchronizes multisensor features with the real-time robot motion data to achieve localized quality prediction. Optical microscope (OM) images of the printed part are used to locate defect-free and defective regions (i.e., cracks and keyhole pores), which serve as ground truth labels for training supervised machine learning (ML) models for quality prediction. The trained ML model is then used to generate a virtual quality map that registers quality prediction outcomes within the 3D volume of the printed part, thus eliminating the need of physical inspections by destructive methods. Experiments show that the virtual quality map closely matches the actual quality observed by OM. Compared to traditional single-sensor-based quality prediction, the MFDT has achieved a significantly higher quality prediction accuracy (96%), a higher ROC-AUC score (99%), and a lower false alarm rate (4.4%). As a result, the MFDT is a more reliable method for defect prediction. The proposed MFDT also lays the groundwork for our future development of a self-adaptive hybrid processing strategy that combines machining with AM for defect removal and quality improvement.

Machine Learning based Intelligent System for Breast Cancer Prediction (MLISBCP)

Risks of death from Breast Cancer (BC) are drastically rising in recent years. The diagnosis of breast cancer is time-consuming due to the limited availability of diagnostic systems such as dynamic MRI, X-rays etc. Early detection and diagnosis of breast cancer significantly impacts life expectancy as current medical technologies are not advanced enough to treat patients in later stages effectively. Even though researchers have created many expert systems for early detection of BC such as WNBC, AR + NN system, AdaBoost ELM etc., but still most expert systems frequently lack adequate handling of the class imbalance problem, proper data pre-processing, and systematic feature selection. To overcome these limitations, this work proposes an expert system named “Machine Learning Based Intelligent System for Breast Cancer Prediction (MLISBCP)” for better prediction of breast cancer using machine learning analytics. The suggested system utilises the ‘K-Means SMOTE’ oversampling method to handle the class imbalance problem and ‘Boruta’ feature selection technique to select the most relevant features of the BC dataset. To understand the effectiveness of the proposed model – MLISBCP, its performance is compared with various single classifier based models, ensemble models and various models present in literature in terms of performance metrics- accuracy, precision, recall, F1-score and RoC AUC Score. The results reveal that the MLISBCP obtained the highest accuracy of 97.53 % with respect to existing models present in the literature.

Improving the classification of veterinary thoracic radiographs through inter-species and inter-pathology self-supervised pre-training of deep learning models

The analysis of veterinary radiographic imaging data is an essential step in the diagnosis of many thoracic lesions. Given the limited time that physicians can devote to a single patient, it would be valuable to implement an automated system to help clinicians make faster but still accurate diagnoses. Currently, most of such systems are based on supervised deep learning approaches. However, the problem with these solutions is that they need a large database of labeled data. Access to such data is often limited, as it requires a great investment of both time and money. Therefore, in this work we present a solution that allows higher classification scores to be obtained using knowledge transfer from inter-species and inter-pathology self-supervised learning methods. Before training the network for classification, pretraining of the model was performed using self-supervised learning approaches on publicly available unlabeled radiographic data of human and dog images, which allowed substantially increasing the number of images for this phase. The self-supervised learning approaches included the Beta Variational Autoencoder, the Soft-Introspective Variational Autoencoder, and a Simple Framework for Contrastive Learning of Visual Representations. After the initial pretraining, fine-tuning was performed for the collected veterinary dataset using 20% of the available data. Next, a latent space exploration was performed for each model after which the encoding part of the model was fine-tuned again, this time in a supervised manner for classification. Simple Framework for Contrastive Learning of Visual Representations proved to be the most beneficial pretraining method. Therefore, it was for this method that experiments with various fine-tuning methods were carried out. We achieved a mean ROC AUC score of 0.77 and 0.66, respectively, for the laterolateral and dorsoventral projection datasets. The results show significant improvement compared to using the model without any pretraining approach.

An end-to-end approach to combine attention feature extraction and Gaussian Process models for deep multiple instance learning in CT hemorrhage detection

Intracranial hemorrhage (ICH) is a serious life-threatening emergency caused by blood leakage inside the brain. Radiologists usually confirm the presence of ICH by analyzing computed tomography (CT) scans, so, developing an automated diagnosis system that can process this type of images has become an important research problem. One of the main challenges to apply AI algorithms in this setting is the lack of labeled data. To mitigate the labeling burden, Multiple Instance Learning (MIL) algorithms group instances into bags, relying solely on bag-level labels for model training. Due to their capacity to handle uncertainty and deliver accurate predictions, Gaussian Processes (GPs) stand out as promising classifiers for MIL problems. Recent research has also demonstrated the effectiveness of combining attention mechanisms with GPs for ICH detection. Nonetheless, existing methods have a notable limitation: they train the attention mechanism and the GP separately, resulting in suboptimal feature extraction for GP-based classification. In this study, we introduce an innovative end-to-end MIL model that concurrently trains the CNN backbone and attention mechanism along with the GP classifier. Our approach enhances the robustness and accuracy of bag predictions by optimizing feature extraction for GP-based classification. We validate our method experimentally by focusing on two ICH detection datasets. Our results reveal a significant performance advantage in terms of accuracy, F1-score, precision, and ROC-AUC score over existing MIL approaches, especially two-stage GP approaches. Additionally, we offer empirical insights into the functionality and effectiveness of our novel model.

Label-Free Cell Detection of Acute Leukemia Using High-Content Morphological Profiling in Flow

Background: Early diagnosis and prompt initiation of appropriate treatment are critical for improving the prognosis of acute leukemia. Acute leukemia is typically diagnosed through microscopic morphological examination of bone marrow smears and flow cytometric immunophenotyping of bone marrow cells stained with fluorophore-conjugated antibodies. However, these diagnostic processes require trained professionals and are time and resource-intensive. Here, we present a novel diagnostic approach using ghost cytometry (GC), a recently developed high-content flow cytometric approach that leverages machine vision-based, stain-free, high-speed morphological characterization and analysis of cells. Methods: Label-free morphological information of cells was acquired as compressive imaging temporal waveforms. These waveforms were directly analyzed using the support vector machine (SVM) algorithm without computational image production, enabling high-speed and accurate information processing. Conventional flow cytometric data modalities, such as forward-scattering (FSC), backscattering (BSC), and fluorescence signals from biological labeling, were used as ground truth labels to identify and annotate blast and normal cells in bone marrow (BM) based on flow cytometric gating schemes. Subsequently, we trained a leukemia cell classifier using the annotated GC (compressive imaging) waveforms and evaluated its classification ability by comparing the predicted results with the ground truth labels. Finally, we applied the trained classifier to unidentified GC waveforms (a test dataset not used for classifier development) to demonstrate the detection of blast cells in BM based on their morphology, without relying on biological labels. Following this protocol, we first trained and tested a machine classifier using a human leukemia NB-4 cell line spiked into BM from a healthy donor. Next, we trained and tested two classifiers using BM aspirates from patients diagnosed with acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML), where CD45 was used as the ground truth marker of leukemic cells. Results: The leukemia cell classifier, developed using a human NB-4 cell line spiked into BM cells from a healthy donor, demonstrated excellent classification accuracy with an ROC-AUC score of 0.9998. The high classification accuracy was reproducible across different spike-in concentrations ranging from 1% to 95%. The classifiers trained using BM cells from patients diagnosed with ALL and AML showed high ROC-AUC scores of 0.986 and 0.941, respectively. When we used the classifiers to evaluate blast cell proportions in test datasets that were not used to develop the classifiers, they were also highly accurate in identifying the percentage of blast cells: 56.8% for ALL with a decision boundary of SVM scores at 0, and 30.4% for AML with a decision boundary of SVM scores at 0.37, when compared to May-Grunwald-Giemsa staining of the same cases by medical technologists at diagnosis (60.0% for ALL and 31.2% for AML). Conclusions: Here, we present a novel diagnostic approach that leverages machine vision-based, stain-free, high-speed morphological characterization and analysis of cells. We demonstrate that the approach can detect leukemic cells from the bone marrow cells of patients diagnosed with ALL and AML, without requiring biological staining. The method presented here holds promise as a rapid, precise, simple, and cost-effective method for the diagnosis of acute leukemia in clinical settings.

Improving serious illness conversations in oncology: A machine learning approach that integrates natural language processing for mortality prediction.

590 Background: ML-based mortality prediction tools in oncology can optimize clinical decisions and prompt end-of-life care discussions. Patients with advanced cancer who have engaged in Goals of Care (GoC) conversations report improved quality of life and better care alignment. However, oncologists often have overly optimistic prognoses and miss timely GoC discussions. Clinical notes are a valuable source of information, but processing and extracting data from them is time-consuming and labor-intensive. To address this issue, we have developed a machine learning application that ingests clinical notes and structured data from electronic health records (EHRs) to generate a 180-day mortality risk, prompting oncologists for GoC conversations. Methods: A predictive machine learning model was developed using data from cancer patients aged 21 and above, diagnosed between January 2016 and December 2021. Data was collected from various sources, including cancer and death registry and the EHR. By analyzing structured and unstructured data from ambulatory progress notes, a clinical profile was created for each patient. The model utilized Spark-NLP for preprocessing, applying word2vec embedding and pre-trained NER models to extract information on diseases, symptoms, procedures, treatments, and medications. Feature engineering techniques were used to select the best NLP features, combined with structured data. The model was trained using 894 patients, employing Random Forest Classifier with 10-fold cross-validation, and tested on a separate set of 43,274 patients. Performance evaluation included ROC AUC, PR AUC, and F1 Score metrics. Results: After the fine tuning, the best model showed an AUC-ROC of 0.88 on the train set and 0.75 on the test set. At a threshold of 0.44, the model achieved a balanced performance with a sensitivity of 0.70 and specificity of 0.71 on the testing set. Conclusions: Our team pioneered the development of an automated multi-modality pipeline that combines unstructured real-world data with structured data, allowing for training and testing of a fusion model. This automation opens doors for scaling and dissemination, to enhance mortality prediction. Future works will involve qualitative analysis of implementation and acceptance in clinical practice.

Machine Learning-Based Screening Solution for COVID-19 Cases Investigation: Socio-Demographic and Behavioral Factors Analysis and COVID-19 Detection

The COVID-19 pandemic has unleashed an unprecedented global crisis, releasing a wave of illness, mortality, and economic disarray of unparalleled proportions. Numerous societal and behavioral aspects have conspired to fuel the rampant spread of COVID-19 across the globe. These factors encompass densely populated areas, adherence to mask-wearing protocols, inadequate awareness levels, and various behavioral and social practices. Despite the extensive research surrounding COVID-19 detection, an unfortunate dearth of studies has emerged to meticulously evaluate the intricate interplay between socio-demographic and behavioral factors and the likelihood of COVID-19 infection. Thus, a comprehensive online-based cross-sectional survey was methodically orchestrated, amassing data from a substantial sample size of 500 respondents. The precisely designed survey questionnaire encompassed various variables encompassing socio-demographics, behaviors, and social factors. The Bivariate Pearson’s Chi-square association test was deftly employed to unravel the complex associations between the explanatory variables and COVID-19 infection. The feature importance approach was also introduced to discern the utmost critical features underpinning this infectious predicament. Four distinct Machine Learning (ML) algorithms, specifically Decision Tree, Random Forest, CatBoost, and XGBoost, were employed to accurately predict COVID-19 infection based on a comprehensive analysis of socio-demographic and behavioral factors. The performance of these models was rigorously assessed using a range of evaluation metrics, including accuracy, recall, precision, ROC-AUC score, and F1 score. Pearson’s Chi-square test revealed a statistically significant association between vaccination status and COVID-19 infection. The use of sanitizer and masks, the timing of infection, and the interval between the first and second vaccine doses were significantly correlated with the likelihood of contracting the COVID-19 virus. Among the ML models tested, the XGBoost classifier demonstrated the highest classification accuracy, achieving an impressive 97.6%. These findings provide valuable insights for individuals, communities, and policymakers to implement targeted strategies aimed at mitigating the impact of the COVID-19 pandemic.

Design of Primal Soft-margin SVM Leveraging Piecewise Linear Approximation based Linear Programming Technique

This research introduces a new approach for training soft-margin Support Vector Machines (SVMs) using the primal formulation. The method, called soft-margin Piecewise Linear Approximation based SVM (Soft-margin PLA-SVM), streamlining the optimization of soft-margin SVM hyperparameters in linear programming framework using well-known GUROBI Optimizer Solver. It eliminates the need for an initial hyperparameter guess and uses an adaptable initial search domain. The study uses the Wisconsin Breast Cancer Original dataset from UCI machine learning to validate the effectiveness of proposed soft-margin PLA-SVM. Comparative analysis shows that proposed PLA-SVM outperforms other classifiers in terms of training speed, accuracy, precision, and ROC-AUC scores. The scalability and computational efficiency of soft-margin PLA-SVM make it suitable for high-dimensional and large-scale datasets. The research demonstrates the effectiveness of the primal perspective in solving the soft-margin SVM design problem.

Smoker Detection Using Machine Learning

Smoking remains a persistent global public health challenge, posing severe health risks and socioeconomic burdens. Early detection of smokers is crucial for implementing targeted interventions, public health campaigns, and personalized support to mitigate smoking-related illnesses. The dataset used in this research comprises extensive information on individual health, encompassing age, height, weight, waist circumference, visual acuity in both eyes, hearing capability in both ears, systolic and diastolic blood pressure, cholesterol levels (HDL, LDL, and triglycerides), haemoglobin levels, urine protein content, liver enzymes (AST, ALT, and GTP), presence of dental caries, and fasting blood sugar. Four For the smoking identification job, cutting-edge machine learning methods including logic regression, Gaussian Naive Bayes, Random Forest Classifier, and XGBoost Classifier are used. Each algorithm is tested using a variety of efficiency indicators after being trained on the data set used for training., including accuracy, ROC-AUC scores, and confusion matrices.The outcomes show the efficiency of the Random Forest Classifier, demonstrating an excellent precision of around 79.4% in correctly identifying smokers. This model outperforms the other algorithms and proves to be a robust approach for smoker detection based on the provided health parameters. Furthermore, the study delves into the interpretability of the models, analyzing the significance of different health parameters in predicting smoking behaviour. Insights gained from the feature importance analysis offer valuable guidance for public health practitioners and policymakers in designing targeted interventions.

A Novel Approach for Improving the Security of IoT–Medical Data Systems Using an Enhanced Dynamic Bayesian Network

IoT (Internet of Things) devices are increasingly being used in healthcare to collect and transmit patient data, which can improve patient outcomes and reduce costs. However, this also creates new challenges for data security and privacy. Thus, the major demand for secure and efficient data-sharing solutions has prompted significant attention due to the increasing volume of shared sensor data. Leveraging a data-fusion-based paradigm within the realm of IoT-protected healthcare systems enabled the collection and analysis of patient data from diverse sources, encompassing medical devices, electronic health records (EHRs), and wearables. This innovative approach holds the potential to yield immediate benefits in terms of enhancing patient care, including more precise diagnoses and treatment plans. It empowers healthcare professionals to devise personalized treatment regimens by amalgamating data from multiple origins. Moreover, it has the capacity to alleviate financial burdens, elevate healthcare outcomes, and augment patient satisfaction. Furthermore, this concept extends to fortifying patient records against unauthorized access and potential misuse. In this study, we propose a novel approach for secure transmission of healthcare data, amalgamating the improved context-aware data-fusion method with an emotional-intelligence-inspired enhanced dynamic Bayesian network (EDBN). The findings indicated that F1 score, accuracy, precision, recall, and ROC-AUC score using DCNN were 89.3%, 87.4%, 91.4%, 92.1%, and 0.56, respectively, which was second-highest to the proposed method. On the other hand, the F1 score, accuracy, precision, recall, and ROC-AUC scores of FRCNN and CNN were low in accuracy at 83.2% and 84.3%, respectively. Our experimental investigation demonstrated superior performance compared with existing methods, as evidenced by various performance metrics, including recall, precision, F measures, and accuracy.

Enhancement of Backward Feature Elimination as a Pre-Processing Method for K Nearest Neighbor Algorithm Applied to Insurance Fraud Detection

This study presents novel enhancements to the Backward Feature Elimination (BFE) method for improved insurance fraud detection using the K-Nearest Neighbor (KNN) algorithm. The research addresses issues inherent in the baseline BFE process, such as the over-reliance on p-values, the potential for misleading results, and suboptimal feature selection leading to overfitting. To address these, the study integrates confidence intervals and feature importance into the BFE process, establishing a more robust and reliable criterion for feature selection. Moreover, feature engineering techniques are introduced during preprocessing to enhance model performance. The modified BFE method demonstrates superior performance over the baseline model regarding the recall, precision, and F1 score. Stratified K-Fold Cross-Validation, ROC-AUC Score, and Coefficient of Variation (CV) confirm the consistency and robustness of the enhanced model across varying data subsets. These innovations offer a comprehensive and reliable solution to feature selection in the BFE method, applied to the KNN model for effective insurance fraud detection. The study mitigates the issues related to p-value dependence and boosts model performance, paving the way for more accurate and robust fraud detection systems.

Automated Alzheimer’s Disease Diagnosis using Convolutional Neural Networks and Magnetic Resonance Imaging

Alzheimer’s disease is a debilitating neuro-logical condition affecting millions globally; therefore, correct diagnosis plays a significant role in treating or managing it effectively. Convolutional neural networks (CNNs), which are popular deep learning algorithms are applied to image processing tasks, offer a good technique to study and investigate images processing. In this study, a CNN model for classifying Alzheimer’s patients is proposed. The research yielded impressive results: recall and precision scores as high as 0.9958 which indicate trustworthy identification of true positives while maintaining few false positives; test accuracy exceeding 99% confirming desirable generalization capabilities from the training dataset to live scenarios; ROC AUC score at an astronomical height of 0.9999 signifying great potential in distinguishing between afflicted individuals from their non-affected counterparts accurately. The proposed network achieved a classification accuracy of 99.94% on LMCI vs EMCI, 99.87% on LMCI vs MCI, 99.95% on LMCI vs AD, 99.94% on LMCI vs CN, 99.99% on CN vs AD, 99.99% on CN vs EMCI, 99.99% on CN vs MCI, 99.99% on AD vs EMCI, 99.98% on AD vs MCI, and 99.96% on MCI vs EMCI. The proposed CNNs model is compared with two ultramodern models such as VGG19 and ResNet50. The results show that the proposed model achieved a superior performance in diagnostic precision and effectiveness of Alzheimer’s disease, leading to early detection, enhanced treatment plans, and enriching the quality of life for those affected.

Prediction of short-acting beta-agonist usage in patients with asthma using temporal-convolutional neural networks.

Changes in short-acting beta-agonist (SABA) use are an important signal of asthma control and risk of asthma exacerbations. Inhaler sensors passively capture SABA use and may provide longitudinal data to identify at-riskpatients. We evaluate the performance of several ML models in predicting daily SABA use for participants with asthma and determine relevant features for predictive accuracy. Participants with self-reported asthma enrolled in a digital health platform (Propeller Health, WI), which included a smartphone application and inhaler sensors that collected the date and time of SABA use. Linear regression, random forests, and temporal convolutional networks (TCN) were applied to predict expected SABA puffs/person/day from SABA usage and environmental triggers. The models were compared with a simple baseline model using explained variance (R2), as well as using average precision (AP) and area under the receiving operator characteristic curve (ROC AUC) for predicting days with ≥1-10 puffs. Data included 1.2 million days of data from 13 202 participants. A TCN outperformed other models in predicting puff count (R2 = 0.562) and day-over-day change in puff count (R2 = 0.344). The TCN predicted days with ≥10 puffs with an ROC AUC score of 0.952 and an AP of 0.762 for predicting a day with ≥1 puffs. SABA use over the preceding 7 days had the highest feature importance, with a smaller but meaningful contribution from air pollutant features. Predicted SABA use may serve as a valuable forward-looking signal to inform early clinical intervention and self-management. Further validation with known exacerbation events is needed.

Identification of cancerlectin proteins using hyperparameter optimization in deep learning and DDE profiles

This study focuses on the development, metastasis, and spread of cancer diseases. It is therefore very desirable to establish deep learning method that classify cancerlectin proteins function efficiently and effectively. We used feature extraction model for physicochemical properties, such as Cancerlectins protein structure, functions, and other compounds. We propose a computational method, namely, cancerlectin two-dimensional convolutional neural networks (Lectin2D-CNN), for predicting cancerlectin proteins. Additionally, we conduct the cross-validation experiments. In addition to this approach, our paper proposes using cancerlectin two-dimensional convolutional neural networks (Lectin2D-CNN) to do image-based classification. The results indicate the proposed method Lectin2D-CNN achieved high accuracy and satisfactory specificity for comparison data sets and was superior to the compared methods. Various classifiers were used to predict cancerlectin protein functions. We developed a prediction model based on the 2D-CNN architecture to increase the recognition sensitivity and accuracy for cancerlectins. Results provide a basis for the estimation of cancer lectins and demonstrate deep learning approaches in in computational proteomics. When the Cross-validation using 2D-CNN random number generator has produced accuracy score obtain 0.7169%, Sensitivity score obtain 0.7012%, Specificity score obtain 0.7326%, MCC score obtain 0.4428%, ROC-AUC score is 0.76%, respectively, then we know we've attained a reliable result. When the Independent datasets using 2D-CNN random number generator has produced accuracy score obtain 0.6375%, Sensitivity score obtain 0.6160%, Specificity score obtain 0.6589%, MCC score obtain 0.2851%, and ROC (auc) score is 0.76%, respectively.

Hyperparameters Optimization in XGBoost Model for Rainfall Estimation: A Case Study in Pontianak City

Estimating rainfall accurately is crucial for both the community and various institutions involved in managing water resources and preventing disasters. The XGBoost model has demonstrated its effectiveness in predicting rainfall, but it still requires fine-tuning of hyperparameters to enhance its performance. This study seeks to determine the optimal learning rate for rainfall prediction while keeping the max_depth and n_estimator parameters fixed. The hyperparameter optimization process was carried out using a two-step approach: an initial coarse search using RandomizedSearchCV followed by a more detailed fine-tuning using GridSearchCV. The model's foundation relied on historical rainfall data gathered over three months from the Automated Weather Observed System (AWOS) at the Pontianak Meteorological Station, recorded on an hourly basis. To assess the model's performance, several metrics were employed, including accuracy, precision, recall, F1 score, and ROC-AUC. The model demonstrated promising results, with accuracy, precision, recall, and F1 score all reaching 95%, indicating its ability to effectively predict rainfall. However, the ROC-AUC score was somewhat lower at 62%. After conducting the hyperparameter search, the optimal learning rate determined for the model, utilizing the 2040 dataset, was found to be 0.204.

Improving Patient Safety Event Report Classification with Machine Learning and Contextual Text Representation.

Adverse events caused by medical errors pose a significant threat to patient safety, with estimates of 251,454 deaths and a cost of $17.1 billion to the healthcare system annually in the United States. Patient safety event (PSE) reports play a vital role in identifying measures to prevent adverse events, but their utility is dependent on the accurate classification of PSE reports. Recent studies have used static natural language processing (NLP) and machine learning (ML) techniques to automate PSE report classification. However, the use of static NLP has limitations in differentiating the meaning of words in disparate contexts, which can lead to inferior classification results. Thus, this study proposes to utilize contextual text representation produced from neural NLP methods to improve the accuracy of PSE report classification. The results suggest that the contextual text representation can further improve the performance of PSE classifiers. The best-performing classifier, a support vector machine trained with contextual text representation (Roberta-base) reaches an accuracy of 0.75 and a ROCAUC score of 0.94, surpassing all ML classifiers trained with static text representations. Furthermore, the confusion matrix of the best classifier exposes latent deficiencies in the PSE reports' classification taxonomy, such as the multi-class nature of PSE and conceptually related event types. The study's findings can save time for PSE reclassification, enhance the learning capabilities of the reporting system, ultimately improve patient safety.

Development and validation of machine learning models for the prediction of SH-2 containing protein tyrosine phosphatase 2 inhibitors.

Discovery and development of a new drug to the market is a highly challenging and resource consuming process. Although, modern drug discovery technologies have enabled the rapid identification of lead compounds, translation of the lead compounds into successful clinical candidates remains a big challenge. In recent years, the availability of massive structural and biological data of diverse small molecules and macromolecules has helped the researchers to deep mine the multidimensional data with the help of artificial intelligence-based predictive tools to draw useful insights on the structural features of biological or therapeutic significance. The aim of this study was to utilize the available data on small molecule (SH2)-containing protein tyrosine phosphatase 2 (SHP2) inhibitors to build and develop machine learning (ML) models that can predict the SHP2 inhibitory potential of new compounds. The dataset contained 2739 unique small molecule SHP2 inhibitors obtained from the BindingDB, ChEMBL and recent literature. After curation of the data, the predictive models such as XGBoost, K nearest neighbours, neural networks were developed and validated through a tenfold cross-validation testing procedure. Out of the seven models developed, the XGBoost model showed an excellent performance with ROC AUC score of 0.96 and accuracy of 0.97 on the test data. Moreover, the Shapley Additive Explanations method was applied to assess a more in-depth understanding of the influence of variables on the model's predictions. In summary, the XGBoost model developed in this study can be useful in the identification of novel SHP2 inhibitors and therefore, can accelerate the discovery of novel therapeutics for cancer therapy.

New drugs and stock market: a machine learning framework for predicting pharma market reaction to clinical trial announcements

Pharmaceutical companies operate in a strictly regulated and highly risky environment in which a single slip can lead to serious financial implications. Accordingly, the announcements of clinical trial results tend to determine the future course of events, hence being closely monitored by the public. Most works focus on retrospective analysis of announcement impact on company stock prices, bypassing the consideration of the problem in the predictive paradigm. In this work, we aim to close this gap by proposing a framework that allows predicting the numerical values of announcement-induced changes in stock prices. In fact, it is a problem of the impact prediction of the specific event on the corresponding time series. Our framework includes a BERT model for extracting the sentiment polarity of announcements, a Temporal Fusion Transformer for forecasting the expected return, a graph convolution network for capturing event relationships, and gradient boosting for predicting the price change. We operate with one of the biggest FDA (the Food and Drug Administration) datasets, consisting of 5436 clinical trial announcements from 681 companies for the years 2018–2022. During the study, we get several significant outcomes and domain-specific insights. Firstly, we obtain statistical evidence for the clinical result promulgation influence on the public pharma market value. Secondly, we witness inherently different patterns of responses to positive and negative announcements, reflected in a stronger and more pronounced reaction to negative clinical news. Thirdly, we discover two factors that play a crucial role in a predictive framework: (1) the drug portfolio size of the company, indicating the greater susceptibility to an announcement in the case of low diversification among drug products and (2) the announcement network effect, manifesting through an increase in predictive power when exploiting interdependencies of events belonging to the same company or nosology. Finally, we prove the viability of the forecast setting by getting ROC AUC scores predominantly greater than 0.7 for the classification of price change on historical data. We emphasize the transferability and generalizability of the developed framework on other datasets and domains but on the condition of the presence of two key entities: events and the associated time series.

A Fine-Tuned Hybrid Stacked CNN to Improve Bengali Handwritten Digit Recognition

Recognition of Bengali handwritten digits has several unique challenges, including the variation in writing styles, the different shapes and sizes of digits, the varying levels of noise, and the distortion in the images. Despite significant improvements, there is still room for further improvement in the recognition rate. By building datasets and developing models, researchers can advance state-of-the-art support, which can have important implications for various domains. In this paper, we introduce a new dataset of 5440 handwritten Bengali digit images acquired from a Bangladeshi University that is now publicly available. Both conventional machine learning and CNN models were used to evaluate the task. To begin, we scrutinized the results of the ML model used after integrating three image feature descriptors, namely Binary Pattern (LBP), Complete Local Binary Pattern (CLBP), and Histogram of Oriented Gradients (HOG), using principal component analysis (PCA), which explained 95% of the variation in these descriptors. Then, via a fine-tuning approach, we designed three customized CNN models and their stack to recognize Bengali handwritten digits. On handcrafted image features, the XGBoost classifier achieved the best accuracy at 85.29%, an ROC AUC score of 98.67%, and precision, recall, and F1 scores ranging from 85.08% to 85.18%, indicating that there was still room for improvement. On our own data, the proposed customized CNN models and their stack model surpassed all other models, reaching a 99.66% training accuracy and a 97.57% testing accuracy. In addition, to robustify our proposed CNN model, we used another dataset of Bengali handwritten digits obtained from the Kaggle repository. Our stack CNN model provided remarkable performance. It obtained a training accuracy of 99.26% and an almost equally remarkable testing accuracy of 96.14%. Without any rigorous image preprocessing, fewer epochs, and less computation time, our proposed CNN model performed the best and proved the most resilient throughout all of the datasets, which solidified its position at the forefront of the field.

Drug Toxicity Prediction by Machine Learning Approaches

Drug property prediction, especially toxicity, helps reduce risks in a range of real-world applications. In this paper, we aim to apply various machine-learning models for solving the drug toxicity prediction problem. Among various machine-learning approaches, we select five suitable representatives: random forest, multi-layer perceptron, logistic regression, graph convolutional neural network, and graph isomorphism network (GIN) for conducting experiments on six datasets for toxicity prediction, including Tox 21, ClinTox, ToxCast, SIDER, HIV, and BACE. We design the GIN with four hidden layers and select the Adam optimizer with the learning rate [Formula: see text] and the batch size [Formula: see text]. Furthermore, we use a batch norm layer inside each of the GIN hidden layers. Experimental results show that the designed GIN model is most efficient in distinguishing between safe and toxic drugs and outperforms the others under the supervision of ROC AUC score and recall.