Build Machine Learning Models Research Articles

Molecular Generation, QSAR, and Molecular Dynamic Simulation Studies for Virtual Screening of DNA Polymerase Theta Inhibitors.

The machine learning-based QSAR modeling procedure, molecular generations, and molecular dynamic simulations were applied to virtually screen the DNA polymerase theta inhibitors. The DNA polymerase theta (Polθ or POLQ) is an attractive target for treatments of homologous recombination deficient (such as BRCA deficient) cancers. There are no approved drugs for targeting POLQ, and only one inhibitor is in Phase Ⅱclinical trials; thus, it is necessary to develop novel POLQ inhibitors. To build machine learning models that predict the bioactivities of POLQ inhibitors. To build molecular generation models that generate diverse molecules. To virtually screen the generated molecules by the machine learning models. To analyze the binding modes of the screening results by molecular dynamic simulations. In the present work, 325 inhibitors with POLQ polymerase domain bioactivities were Collected. Two machine learning methods, random forest and deep neural network, were used for building the ligand- and structure-based quantitative structure-activity relationship (QSAR) models. The substructure replacement-based method and transfer learning-based deep recurrent neural network method were used for molecular generations. Molecular docking and consensus QSAR models were carried out for virtual screening. The molecular dynamic simulations and MM/GBSA binding free energy calculation and decomposition were used to further analyze the screening results. The MCC values of the best ligand- and structure-based consensus QSAR models reached 0.651 and 0.361 for the test set, respectively. The machine learning-based docking scores had better-predicted ability to distinguish the highly and weakly active poses than the original docking scores. The 96490 molecules were generated by both molecular generation methods, and 10 molecules were retained by virtual screening. Four favorable interactions were concluded by molecular dynamic simulations. We hope that the screening results and the binding modes are helpful for designing the highly active POLQ polymerase inhibitors and the models of the molecular design workflow can be used as reliable tools for drug design.

Incorporating Synthetic Accessibility in Drug Design: Predicting Reaction Yields of Suzuki Cross-Couplings by Leveraging AbbVie's 15-Year Parallel Library Data Set.

Despite the increased use of computational tools to supplement medicinal chemists' expertise and intuition in drug design, predicting synthetic yields in medicinal chemistry endeavors remains an unsolved challenge. Existing design workflows could profoundly benefit from reaction yield prediction, as precious material waste could be reduced, and a greater number of relevant compounds could be delivered to advance the design, make, test, analyze (DMTA) cycle. In this work, we detail the evaluation of AbbVie's medicinal chemistry library data set to build machine learning models for the prediction of Suzuki coupling reaction yields. The combination of density functional theory (DFT)-derived features and Morgan fingerprints was identified to perform better than one-hot encoded baseline modeling, furnishing encouraging results. Overall, we observe modest generalization to unseen reactant structures within the 15-year retrospective library data set. Additionally, we compare predictions made by the model to those made by expert medicinal chemists, finding that the model can often predict both reaction success and reaction yields with greater accuracy. Finally, we demonstrate the application of this approach to suggest structurally and electronically similar building blocks to replace those predicted or observed to be unsuccessful prior to or after synthesis, respectively. The yield prediction model was used to select similar monomers predicted to have higher yields, resulting in greater synthesis efficiency of relevant drug-like molecules.

DeepSP: Deep learning-based spatial properties to predict monoclonal antibody stability

Therapeutic antibody development faces challenges due to high viscosities and aggregation tendencies. The spatial charge map (SCM) and spatial aggregation propensity (SAP) are computational techniques that aid in predicting viscosity and aggregation, respectively. These methods rely on structural data derived from molecular dynamics (MD) simulations, which are computationally demanding. DeepSCM, a deep learning surrogate model based on sequence information to predict SCM, was recently developed to screen high-concentration antibody viscosity. This study further utilized a dataset of 20,530 antibody sequences to train a convolutional neural network deep learning surrogate model called Deep Spatial Properties (DeepSP). DeepSP directly predicts SAP and SCM scores in different domains of antibody variable regions based solely on their sequences without performing MD simulations. The linear correlation coefficient between DeepSP scores and MD-derived scores for 30 properties achieved values between 0.76 and 0.96 with an average of 0.87. DeepSP descriptors were employed as features to build machine learning models to predict the aggregation rate of 21 antibodies, and the performance is similar to the results obtained from the previous study using MD simulations. This result demonstrates that the DeepSP approach significantly reduces the computational time required compared to MD simulations. The DeepSP model enables the rapid generation of 30 structural properties that can also be used as features in other research to train machine learning models for predicting various antibody stability using sequences only. DeepSP is freely available as an online tool via https://deepspwebapp.onrender.com and the codes and parameters are freely available at https://github.com/Lailabcode/DeepSP.

Abstract PO3-07-02: Radiomics Models for B-mode Breast Ultrasound and Strain Elastography to improve Breast Cancer Diagnosis (INSPiRED 005): An International, Multicenter Analysis

Abstract Background: Breast Elastography, a technique that quantifies tissue stiffness, has been evaluated to objectify and improve the performance of B-mode breast ultrasound. However, large prospective trials showed benefits in BI-RADS 4a breast masses only and a high operator dependency. Modern Artificial Intelligence techniques for automated image processing like radiomics, a technique where quantified features are extracted from images, may overcome these limitations. We aimed to develop and validate radiomics models based on B-mode and Strain Elastography (SE) images for patients with BI-RADS 3 or 4 breast masses and compare their performance to the respective human experts. Methods: This is a secondary analysis of an international, multicenter trial (NCT02638935), evaluating the performance of SE in women with BI-RADS 3 or 4 breast masses. Women were recruited at 12 institutions in 7 countries and underwent B-mode breast ultrasound as well as SE. B-mode images were saved and re-assed by three ultrasound readers ( >10 years of experience), resulting in three independent assessments and a final consensus assessment. SE was interpreted using the E-/B ratio. B-mode and strain images were manually segmented and quantitative radiomics features were extracted using pyradiomics. We used 10-fold cross-validation to build machine learning models (XGBoostTree, MARS) based on data of 11 of 12 study sites. The data of the 12th (largest) study site was used as external validation set. Performance metrics included sensitivity, specificity and area under the receiver operator characteristic curve (AUROC). Results: The study included a total of 1288 patients, 1206 with evaluable B-mode images and 1190 with evaluable Strain images. Mean age was 46.6 years (SD 16.02) and a total number of 29.0% (350 of 1206) and 28.9% (344 of 1190) breast masses were malignant in the B-mode and Strain cohort, respectively. Distribution of BI-RADS categories was 33.0%, 34.5%, 14.5%, and 18.0% for BI-RADS 3, 4a, 4b, and 4c, respectively. In the external validation set (n = 342), the B-mode radiomics model (XGBoostTree) achieved an AUROC of 0.86 (95% CI 0.82 to 0.90), with a sensitivity of 97.4% (95% CI 0.93 to 1.00, 113 of 116) and a specificity of 27.0% (95% CI 0.21 to 0.33, 61 of 226). The model showed equivalent performance compared to the three ultrasound readers (P = 0.133); see also Table 1. In the external validation set (n = 333), the Strain radiomics model (MARS) achieved an AUROC of 0.84 (95% CI 0.79 to 0.88), with a sensitivity of 100% (95% CI 47.0 to 58.0, 115 of 115) and a specificity of 25.5% (95% CI 0.22 to 0.34, 60 of 218). The model showed equivalent performance compared to the three ultrasound readers (P = 0.696) and performed significantly better compared to SE (P = 0.002); see also Table 1. Sensitivity of the strain model was descriptively higher (100% vs. 97.4%, see Table 1). Both models were well-calibrated. Conclusion: This is the largest development and validation study for radiomics models based on B-mode breast ultrasound and SE, to date. The radiomics models performed on par with human readers, with the strain radiomics model showing potential to identify initially missed carcinomas in BI-RADS 3 breast masses. Future implementation studies may evaluate the performance of these image analysis algorithms in clinical routine and their integration into the multi-modal breast cancer diagnostics process, including mammography and MRI. Table 1. Diagnostic performance metrics Citation Format: André Pfob, Tanja He, Lie Cai, Richard G. Barr, Volker Duda, Zaher Alwafai, Corinne Balleyguier, Dirk-André Clevert, Sarah Fastner, Christina Gomez, Manuela Goncalo, Ines Gruber, Markus Hahn, André Hennigs, Panagiotis Kapetas, Sheng-Chieh Lu, Juliane Nees, Ralf Ohlinger, Fabian Riedel, Matthieu Rutten, Benedikt Schaefgen, Anne Stieber, Riku Togawa, Mitsuhiro Tozaki, Sebastian Wojcinski, Cai Xu, Geraldine Rauch, Jörg Heil, Chris Sidey-Gibbons, Michael Golatta. Radiomics Models for B-mode Breast Ultrasound and Strain Elastography to improve Breast Cancer Diagnosis (INSPiRED 005): An International, Multicenter Analysis [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO3-07-02.

FXAM: A unified and fast interpretable model for predictive analytics

Predictive analytics aims to build machine learning models to predict behavior patterns and use predictions to guide decision-making. Predictive analytics is human involved, thus the machine learning model is preferred to be interpretable. In literature, Generalized Additive Model (GAM) is a standard for interpretability. However, due to the one-to-many and many-to-one phenomena which appear commonly in real-world scenarios, existing GAMs have limitations to serve predictive analytics in terms of both accuracy and training efficiency. In this paper, we propose FXAM (Fast and eXplainable Additive Model), a unified and fast interpretable model for predictive analytics. FXAM extends GAM’s modeling capability with a unified additive model for numerical, categorical, and temporal features. FXAM conducts a novel training procedure called Three-Stage Iteration (TSI). TSI corresponds to learning over numerical, categorical, and temporal features respectively. Each stage learns a local optimum by fixing the parameters of other stages. We design joint learning over categorical features and partial learning over temporal features to achieve high accuracy and training efficiency. We prove that TSI is guaranteed to converge to the global optimum. We further propose a set of optimization techniques to speed up FXAM’s training algorithm to meet the needs of interactive analysis. Thorough evaluations conducted on diverse data sets verify that FXAM significantly outperforms existing GAMs in terms of training speed, and modeling categorical and temporal features. In terms of interpretability, we compare FXAM with the typical post-hoc approach XGBoost+SHAP on two real-world scenarios, which shows the superiority of FXAM’s inherent interpretability for predictive analytics.

Development of a Novel In Silico Classification Model to Assess Reactive Metabolite Formation in the Cysteine Trapping Assay and Investigation of Important Substructures.

Predicting whether a compound can cause drug-induced liver injury (DILI) is difficult due to the complexity of drug mechanism. The cysteine trapping assay is a method for detecting reactive metabolites that bind to microsomes covalently. However, it is cumbersome to use 35S isotope-labeled cysteine for this assay. Therefore, we constructed an in silico classification model for predicting a positive/negative outcome in the cysteine trapping assay. We collected 475 compounds (436 in-house compounds and 39 publicly available drugs) based on experimental data performed in this study, and the composition of the results showed 248 positives and 227 negatives. Using a Message Passing Neural Network (MPNN) and Random Forest (RF) with extended connectivity fingerprint (ECFP) 4, we built machine learning models to predict the covalent binding risk of compounds. In the time-split dataset, AUC-ROC of MPNN and RF were 0.625 and 0.559 in the hold-out test, restrictively. This result suggests that the MPNN model has a higher predictivity than RF in the time-split dataset. Hence, we conclude that the in silico MPNN classification model for the cysteine trapping assay has a better predictive power. Furthermore, most of the substructures that contributed positively to the cysteine trapping assay were consistent with previous results.

Research on a machine learning-based adaptive and efficient screening model for psychological symptoms of community correctional prisoners.

Community correction institutions in China frequently employ the Symptom Checklist-90 (SCL-90) and the health survey brief (SF-12) as primary tools for psychological assessment of community correctional prisoners. However, in practical application, the SCL-90 Checklist faces issues such as complex item numbers, overall low cultural level of the subjects, and insufficient professional level of the administrators. The SF-12 health survey brief, as a preliminary screening tool, although only has 12 questions, to some extent simplifies the evaluation process and improves work efficiency, it is prone to missed screening. The research team collected 17-dimensional basic characteristic data and corresponding SCL-90 and SF-12 data from 25,480 samples of community correctional prisoners in Zhejiang Province, China. This study explored the application of multi-label multi-classification algorithms and oversampling techniques in building machine learning models to delve into the correlation between the psychological health risks of community correctional prisoners and their characteristic data. Inspired by computerized adaptive testing (CAT), we constructed an adaptive and efficient screening model for community correctional prisoners through experimental comparisons, based on the binary relevance algorithm with sample oversampling. This screening model personalize the assessment process by dynamically matching participants with the most relevant subset (s) of the nine dimensions of the SCL-90 based on their individual characteristics. Thus, adaptive dynamic simplification and personalized recommendation of the SCL-90 scale between question groups were achieved for the specific group of community correctional prisoners. As a screening tool for psychological symptoms of community correctional prisoners, this model significantly simplifies the number of questions compared to SCL-90, with a simplification rate of up to 65%. However, it achieves this simplification while maintaining excellent performance. The accuracy reached 0.66, with a sensitivity of 0.754, and an F1 score of 0.649. This innovation simplified the assessment process, reduced the assessment time, improved work efficiency, and enhanced the ability to judge the specificity of community correctional prisoners population. Compared to the SF-12, although the simplification rate and accuracy of the model are slightly lower than those of the SF-12, the sensitivity increased by 42.26%, and the F1 score improved by 15.28%. This means the model greatly reduces the possibility of missed screening, effectively preventing prisoners with abnormal psychological or mental states from losing control due to missed screening, and even committing suicide, self injury, or injuring others.

Analysis of Job Recommendations in Vocational Education Using the Intelligent Job Matching Model

Vocational high schools are one of the educational stages impacted by Indonesia's low quality of education. Vocational High Schools play a crucial role in improving human resources. Graduates of Vocational High Schools can continue their education at universities or enter the workforce directly. Many students are found to have not yet considered their career path after graduation. At the same time, graduates are still expected to find mismatched employment with their expertise and skills. This research uses CRISP-DM, or Cross Industry Standard Process for Data Mining, to build machine learning models. The approach used is content-based filtering. This model recommends items similar to previously liked or selected items by the user. Item similarity can be calculated based on the features of the items being compared. After students receive job recommendations through intelligent job matching, they can use these recommendations as references when applying for jobs that align with their results. This process helps students direct their steps toward finding jobs that match their profiles, ultimately increasing their chances of success in the job market. These recommendations are crucial in guiding students toward career paths that align with their abilities and interests. The Intelligent Job Matching Model developed in this research provides recommendations for the job-matching process. This model benefits graduates by providing job recommendations aligned with their profiles and offers advantages to the job market. By implementing the Model of Intelligent Job Matching in the recruitment process, applicants with job qualifications can be matched effectively.

Utilizing ultra-early continuous physiologic data to develop automated measures of clinical severity in a traumatic brain injury population

Determination of prognosis in the triage process after traumatic brain injury (TBI) is difficult to achieve. Current severity measures like the Trauma and injury severity score (TRISS) and revised trauma score (RTS) rely on additional information from the Glasgow Coma Scale (GCS) and the Injury Severity Score (ISS) which may be inaccurate or delayed, limiting their usefulness in the rapid triage setting. We hypothesized that machine learning based estimations of GCS and ISS obtained through modeling of continuous vital sign features could be used to rapidly derive an automated RTS and TRISS. We derived variables from electrocardiograms (ECG), photoplethysmography (PPG), and blood pressure using continuous data obtained in the first 15 min of admission to build machine learning models of GCS and ISS (ML-GCS and ML-ISS). We compared the TRISS and RTS using ML-ISS and ML-GCS and its value using the actual ISS and GCS in predicting in-hospital mortality. Models were tested in TBI with systemic injury (head abbreviated injury scale (AIS) ≥ 1), and isolated TBI (head AIS ≥ 1 and other AIS ≤ 1). The area under the receiver operating characteristic curve (AUROC) was used to evaluate model performance. A total of 21,077 cases (2009–2015) were in the training set. 6057 cases from 2016 to 2017 were used for testing, with 472 (7.8%) severe TBI (GCS 3–8), 223 (3.7%) moderate TBI (GCS 9–12), and 5913 (88.5%) mild TBI (GCS 13–15). In the TBI with systemic injury group, ML-TRISS had similar AUROC (0.963) to TRISS (0.965) in predicting mortality. ML-RTS had AUROC (0.823) and RTS had AUROC 0.928. In the isolated TBI group, ML-TRISS had AUROC 0.977, and TRISS had AUROC 0.983. ML-RTS had AUROC 0.790 and RTS had AUROC 0.957. Estimation of ISS and GCS from machine learning based modeling of vital sign features can be utilized to provide accurate assessments of the RTS and TRISS in a population of TBI patients. Automation of these scores could be utilized to enhance triage and resource allocation during the ultra-early phase of resuscitation.

Development of Clinical Radiomics-Based Models to Predict Survival Outcome in Pancreatic Ductal Adenocarcinoma: A Multicenter Retrospective Study.

This multicenter retrospective study aims to identify reliable clinical and radiomic features to build machine learning models that predict progression-free survival (PFS) and overall survival (OS) in pancreatic ductal adenocarcinoma (PDAC) patients. Between 2010 and 2020 pre-treatment contrast-enhanced CT scans of 287 pathology-confirmed PDAC patients from two sites of the Hopital Universitaire de Bruxelles (HUB) and from 47 hospitals within the HUB network were retrospectively analysed. Demographic, clinical, and survival data were also collected. Gross tumour volume (GTV) and non-tumoral pancreas (RPV) were semi-manually segmented and radiomics features were extracted. Patients from two HUB sites comprised the training dataset, while those from the remaining 47 hospitals of the HUB network constituted the testing dataset. A three-step method was used for feature selection. Based on the GradientBoostingSurvivalAnalysis classifier, different machine learning models were trained and tested to predict OS and PFS. Model performances were assessed using the C-index and Kaplan-Meier curves. SHAP analysis was applied to allow for post hoc interpretability. A total of 107 radiomics features were extracted from each of the GTV and RPV. Fourteen subgroups of features were selected: clinical, GTV, RPV, clinical & GTV, clinical & GTV & RPV, GTV-volume and RPV-volume both for OS and PFS. Subsequently, 14 Gradient Boosting Survival Analysis models were trained and tested. In the testing dataset, the clinical & GTV model demonstrated the highest performance for OS (C-index: 0.72) among all other models, while for PFS, the clinical model exhibited a superior performance (C-index: 0.70). An integrated approach, combining clinical and radiomics features, excels in predicting OS, whereas clinical features demonstrate strong performance in PFS prediction.

Building Machine Learning Models for Fraud Detection in Customs Declarations in Senegal

To improve the customs declaration control system in Senegal, we propose fraud risk prediction models built with machine learning methods such as Neural Networks (MLP), Support Vector Machine (SVM), Random Forest (RF) and eXtreme Gradient Boosting (XGBoost). These models were built from historical customs declaration data and then tested on a part of the data reserved for this purpose to evaluate their prediction performance according to the metrics of accuracy, precision, recall, and F1-Score. The RF model proved to be the more performant model and is followed, in order, by the XGBoost model, and the MLP and SVM models.

Towards Reliable Learning in the Wild: Generalization and Adaptation

The real-world deployment of machine learning algorithms often poses challenges due to shifts in data distributions and tasks. These shifts can lead to a degradation in model performance, as the model may not have encountered such changes during training. Additionally, they can make it difficult for the model to generalize to new scenarios and can result in poor performance in real-world applications. In this talk, I will present our research on building machine learning models that are highly generalizable and easily adaptable to different shifts. Specifically, I will first discuss our approach to improving out-of-distribution robustness and mitigating spurious correlations by training environment-invariant models through selective augmentation and post-hoc rectification. Second, I will present our techniques for continuous and rapid adaptation of models to new tasks and environments. This includes methods to facilitate compositional generalization and adaptation by extracting relationships from historical observations and to enhance reliable adaptation even in the face of imperfect observations. Additionally, I will showcase our successful practices for addressing shifts in real-world applications, such as in the healthcare, e-commerce, and transportation industries. The talk will also touch upon the remaining challenges and outline future research directions in this area.

Predicting Rate Constants of Alkane Cracking Reactions Using Machine Learning.

Calculating the thermal rate constants of elementary combustion reactions is of great importance in theoretical chemistry. Machine learning has become a powerful, data-driven method for predicting rate constants nowadays. Recently, the molecular similarity combined with the topological indices were proposed to represent the hydrogen abstraction reactions of alkane [J. Chem. Inf. Model. 2023, 63, 5097-5106], which are, however, not applicable to alkane cracking reactions, another important class of combustion reactions, due to the cleavage of the C-C bond. In this work, a new feature selection scheme is proposed to describe both bimolecular and unimolecular cracking reactions. Molecular descriptors are elaborately selected individually for both reactants and products from those generated by the open-source software RDKit. Machine learning models combined with these molecular descriptors are proven to have the ability to accurately predict rate constants of both the hydrogen abstraction reactions of alkanes by CH3 and the alkane cracking reactions. The average deviation of the XGB-FNN model for prediction is around 60% for the hydrogen abstraction reactions of alkanes and 100% for the alkane cracking reactions. It is expected that the descriptors proposed in this work can be applied to build machine learning models for other reactions.

PvFed: Personalized Vertical Federated learning for Client‐Specific Tasks

Federated Learning (FL) is a distributed machine learning paradigm that enables multiple data holders to collaborate on building machine learning models while preserving the privacy of their data. FL can be categorized as horizontal or vertical, depending on the distribution characteristics of the data. Specifically, horizontal FL uses data partitioned in the sample space, whereas vertical FL uses data partitioned in the feature space. Traditional vertical FL methods aim to facilitate collaboration among clients to infer a single global target. However, these methods may be impractical because each client often has a unique target to be inferred. In this paper, we propose a novel vertical FL method, called personalized vertical federated learning (pvFed), which addresses this limitation by allowing each client to perform inferences specific to their individual task. To the best of our knowledge, no existing method currently addresses this limitation. The objective of pvFed is to construct a global model that generates a representation vector to support client inference. The global model, constructed using distillation and dimensionality reduction, takes a sample ID common to all clients as input and outputs a sample‐specific representation vector. Clients utilize the intermediate representation of their own model and the representation vectors output by the global model for inference. Because these vectors are not dependent on client‐specific tasks, clients can repurpose them for any additional tasks. Our experiments, conducted on two distinct data types—image and tabular data sets, under a vertical partitioning where each client had its own specific task, demonstrated the efficacy of vectors generated by the global model in pvFed for client inference. © 2024 Institute of Electrical Engineer of Japan and Wiley Periodicals LLC.

On building machine learning models for medical dataset with correlated features

Abstract This work builds machine learning models for the dataset generated using a numerical model developed on an idealized human artery. The model has been constructed accounting for varying blood characteristics as it flows through arteries with variable vascular properties, and it is applied to simulate blood flow in the femoral and its continued artery. For this purpose, we designed a pipeline model consisting of three components to include the major segments of the femoral artery: CFA, the common femoral artery and SFA, the superficial artery, and its continued one, the popliteal artery (PA). A notable point of this study is that the features and target variables of the former component pipe form the set of features of the latter, thus resulting in multicollinearity among the features in the third component pipe. Thus, we worked on understanding the effect of these correlated features on the target variables using regularized linear regression models, ensemble, and boosting algorithms. This study highlighted the blood velocity in CFA as the primary influential factor for wall shear stress in both CFA and SFA. Additionally, it established the blood rheology in PA as a significant factor for the same in it. Nevertheless, because the study relies on idealized conditions, these discoveries necessitate thorough clinical validation.

The application of machine learning based on computed tomography images in the identification of renal tumors in children.

The clinical manifestations of Wilms tumor and non-Wilms tumor in children are similar, and the only way to confirm the diagnosis is by postoperative pathology. Computed tomography (CT) is one of the main methods for preoperative diagnosis of the two, but it is also difficult to distinguish because it is easily affected by the subjective influence and the experience of the radiologists. The CT images of 82 children with renal tumors admitted to the Department of Pediatric Urology, Shandong Provincial Hospital from January 2011 to March 2022 were retrospectively analyzed. First, we drew the two-dimensional (2D) region of interest (ROI) of the largest cross-section on the corticomedullary phase (CMP) and nephrogenic phase (NP) images, and extracted seven types of 107 features in the ROI. Then, the texture features with similarity greater than 95% and repetition less than 90% were screened out, and the remaining texture features were further screened by analysis of variance (ANOVA) and recursive feature elimination (RFE). Finally, 15 texture feature were used to build the machine learning (ML) models. We used the synthetic minority oversampling technique (SMOTE) and 10-fold cross-validation to build ML models and verified them in the training, testing, and internal validation sets. The area under the receiver-operating characteristic curve (AUC) and calibration curve were used to evaluate the diagnostic performance. We collected 77 CMP and 81 NP images, which were randomly divided into the training set and the testing set according to the ratio of 7:3. In the internal validation of CMP, the Mean-PCC-ANOVA-5-AE pipeline model achieved the highest AUC 0.792 [95% confidence interval (CI): 0.653-0.930], and its accuracy (ACC), sensitivity (SEN), and specificity (SPE) were 0.833, 0.539 and 0.927, respectively. Correspondingly, in NP, the Mean-PCC-ANOVA-2-LR pipeline model achieved the highest AUC 0.655 (95% CI: 0.485-0.82) in the internal validation. The ACC, SEN, and SPE were 0.696, 0.539, and 0.744, respectively. The ML models based on CT images have good diagnostic efficiency in differentiating Wilms tumors from non-Wilms tumors in children.

Predicting baccalaureate student result to prevent failure: a hybrid model approach

<div class="OutlineElement Ltr SCXW634036 BCX2" style="direction: ltr;"><p>The Moroccan Ministry of National Education has seen substantial modifications over the previous ten years, which have contributed to improving the quality of education. However, there is a discrepancy in the percentage of academic achievement between the regional directorates and educational institutions. Machine learning techniques have become a powerful tool for proactively predicting student admission. The goal of our paper is to build machine learning models using various algorithms to predict the final baccalaureate school year outcomes. We compare regression and classification to find the reasons behind students' failure and to choose an appropriate model for predicting the results. This helps decision-makers make appropriate interventions.</p></div>

Machine learning-based prediction of lithium-ion battery life cycle for capacity degradation modelling

With a particular focus on capacity degradation modeling, this paper offers a ground-breaking examination of the use of machine learning approaches for the precise prediction of lithium-ion battery life cycles. Because lithium-ion batteries are essential to many technological applications, it's critical to comprehend and anticipate their life cycles in order to maximize performance and guarantee sustainable energy solutions. The study starts with an extensive examination of the literature, assessing current approaches critically and setting the stage for the introduction of models based on machine learning. The process entails the methodical collecting of data across a range of operational settings, environmental variables, and charging-discharging cycles. Thorough preprocessing guarantees the dataset's consistency and quality for further machine learning model training. Predictive models are created using a variety of machine learning algorithms, including regression models, support vector machines, and deep neural networks. In order to improve prediction accuracy, the paper focuses on the reasoning behind model selection, parameter tuning, and the incorporation of ensemble approaches. In order to uncover important elements influencing the life cycles of lithium-ion batteries and provide important insights into degradation mechanisms, feature selection approaches are used. Using cross-validation techniques and real-world lithium-ion battery datasets, the built machine learning models go through rigorous evaluation and validation processes to determine their robustness, capacity for generalization, and performance metrics. Comparing machine learning-based predictions with conventional models, the results are presented and discussed, offering insights into the interpretability of the models and the identification of important affecting elements. In order to promote proactive maintenance and optimize battery usage, predictive models are integrated into real-time monitoring systems. The consequences for battery management systems are examined. The paper continues by discussing the challenges that come with using machine learning to estimate the life cycle of batteries and outlining possible directions for further research and development, such as scalability, interpretability, and the incorporation of emerging technologies. This research contributes to the ongoing efforts to increase the reliability and sustainability of lithium-ion battery technologies by highlighting the potential impact of machine learning on energy storage system optimization.

Identification of New Modulators and Inhibitors of Palmitoyl-Protein Thioesterase 1 for CLN1 Batten Disease and Cancer.

Palmitoyl-protein thioesterase 1 (PPT1) is an understudied enzyme that is gaining attention due to its role in the depalmitoylation of several proteins involved in neurodegenerative diseases and cancer. PPT1 is overexpressed in several cancers, specifically cholangiocarcinoma and esophageal cancers. Inhibitors of PPT1 lead to cell death and have been shown to enhance the killing of tumor cells alongside known chemotherapeutics. PPT1 is hence a viable target for anticancer drug development. Furthermore, mutations in PPT1 cause a lysosomal storage disorder called infantile neuronal ceroid lipofuscinosis (CLN1 disease). Molecules that can inhibit, stabilize, or modulate the activity of this target are needed to address these diseases. We used PPT1 enzymatic assays to identify molecules that were subsequently tested by using differential scanning fluorimetry and microscale thermophoresis. Selected compounds were also tested in neuroblastoma cell lines. The resulting PPT1 screening data was used for building machine learning models to help select additional compounds for testing. We discovered two of the most potent PPT1 inhibitors reported to date, orlistat (IC50 178.8 nM) and palmostatin B (IC50 11.8 nM). When tested in HepG2 cells, it was found that these molecules had decreased activity, indicating that they were likely not penetrating the cells. The combination of in vitro enzymatic and biophysical assays enabled the identification of several molecules that can bind or inhibit PPT1 and may aid in the discovery of modulators or chaperones. The molecules identified could be used as a starting point for further optimization as treatments for other potential therapeutic applications outside CLN1 disease, such as cancer and neurological diseases.

Reduced representative methylome profiling of cell-free DNA for breast cancer detection.

Whole-genome methylation sequencing of cfDNA is not cost-effective for tumor detection. Here, we introduce reduced representative methylome profiling (RRMP), which employs restriction enzyme for depletion of AT-rich sequence to achieve enrichment and deep sequencing of CG-rich sequences. We first verified the ability of RRMP to enrich CG-rich sequences using tumor cell genomic DNA and analyzed differential methylation regions between tumor cells and normal whole blood cells. We then analyzed cfDNA from 29 breast cancer patients and 27 non-breast cancer individuals to detect breast cancer by building machine learning models. RRMP captured 81.9% CpG islands and 75.2% gene promoters when sequenced to 10 billion base pairs, with an enrichment efficiency being comparable to RRBS. RRMP allowed us to assess DNA methylation changes between tumor cells and whole blood cells. Applying our approach to cfDNA from 29 breast cancer patients and 27 non-breast cancer individuals, we developed machine learning models that could discriminate between breast cancer and non-breast cancer controls (AUC = 0.85), suggesting possibilities for truly non-invasive cancer detection. We developed a new method to achieve reduced representative methylome profiling of cell-free DNA for tumor detection.