Build Prediction Models Research Articles

An introduction to predictive distribution modelling for conservation to encourage novel perspectives

The rapid pace and potentially irreversible consequences of global change create an urgent need to predict the spatial responses of biota for conservation to better inform the prioritization and management of terrestrial habitats and prevent future extinctions. Here, we provide an accessible entry point to the field to guide near-future work building predictive species distribution models (SDMs) by synthesizing a technical framework for the proactive conservation of avian biodiversity. Our framework offers a useful approach to navigate the challenges surrounding the large spatio-temporal resolution of datasets and datasets that favor hypothesis testing at broad spatio-temporal scales and coarse resolutions, which can affect our ability to assess the validity of current predicted distributions. We explain how to improve the accuracy of predictive models by determining the extent to which: 1) dispersal limitation impacts the rate of range shifts, 2) taxa are rare at their range limits, and 3) land use and climate change interact. Finally, we offer approaches to filling knowledge gaps by creatively leveraging existing methods and data sources.

Identification of Ovarian High-Grade Serous Carcinoma with Mitochondrial Gene Variation

Women diagnosed with advanced-stage ovarian cancer have a much worse survival rate than women diagnosed with early-stage ovarian cancer, but the early detection of this disease remains a clinical challenge. Some recent reports indicate that genetic variations could be useful for the early detection of several malignancies. In this pilot observational retrospective study, we aimed to assess whether mitochondrial DNA (mtDNA) variations could discriminate the most frequent type of ovarian cancer, high-grade serous carcinoma (HGSC), from normal tissue. We identified mtDNA variations from 20 whole-exome sequenced (WES) HGSC samples and 14 controls (normal tubes) using the best practices of genome sequencing. We built prediction models of cancer with these variants, with good performance measured by the area under the curve (AUC) of 0.88 (CI: 0.74–1.00). The variants included in the best model were correlated with gene expression to assess the potentially affected processes. These analyses were validated with the Cancer Genome Atlas (TCGA) dataset, (including over 420 samples), with a fair performance in AUC terms (0.63–0.71). In summary, we identified a set of mtDNA variations that can discriminate HGSC with good performance. Specifically, variations in the MT-CYB gene increased the risk for HGSC by over 30%, and MT-CYB expression was significantly decreased in HGSC patients. Robust models of ovarian cancer detection with mtDNA variations could be applied to liquid biopsy technology, like those which have been applied to other cancers, with a special focus on the early detection of this lethal disease.

Diagnostics and Prognostics of Boilers in Power Plant Based on Data-Driven and Machine Learning

This paper reports diagnostics and prognostics study of boiler in power plant using actual boiler operating data. This study aims to early detect anomalies that occur in the boiler and to predict the remaining useful life (RUL) after anomalies are detected. The proposed method utilizes machine learning techniques through support vector machine (SVM) and random forest algorithm (RFA) for anomaly detection and similarity-based method of dynamic time warping (DTW) for RUL prediction. The developed method is validated by testing the prediction models using real operating data acquired from three boilers in power plant. The results show that some anomalies are successfully detected by prediction model even though there are anomalies that give low accuracies in predictions. RUL prediction also provides fair results given the limitations of the real data used in building prediction models. Overall, the results of this study have potential to be applied in real system as an auxiliary tool in the boiler condition monitoring to support boiler maintenance programs.

Patient Sickness Prediction Using Machine Learning

The ability to predict a patient's sickness is essential to modern healthcare since it allows for timely treatments that can significantly improve patient outcomes. This work aims to predict the risk of specific illnesses using machine learning (ML) techniques based on large patient data. This project intends to build prediction models utilizing test results, lifestyle factors, and electronic health records (EHRs) to assist healthcare professionals in identifying at-risk patients and facilitating preventative healthcare treatments. The results of the study will enhance clinical judgment and patient care by applying efficient predictive analytics in the medical field. Keywords: At-Risk Patients, Disease Prediction, Electronic Health Records (EHRs), Machine Learning (ML)

Enhancing Identification of High-Risk cN0 Lung Adenocarcinoma Patients Using MRI-Based Radiomic Features.

This study aimed to develop a magnetic resonance imaging (MRI)-based radiomics model to predict high-risk pathologic features for lung adenocarcinoma: micropapillary and solid pattern (MPsol), spread through air space, and poorly differentiated patterns. As a prospective study, we screened clinical N0 lung cancer patients who were surgical candidates and had undergone both 18F-fluorodeoxyglucose (FDG) positron emission tomography-computed tomography (PET/CT) and chest CT from August 2018 to January 2020. We recruited patients meeting our proposed imaging criteria indicating high-risk, that is, poorer prognosis of lung adenocarcinoma, using CT and FDG PET/CT. If possible, these patients underwent an MRI examination from which we extracted 77 radiomics features from T1-contrast-enhanced and T2-weighted images. Additionally, patient demographics, maximum standardized uptake value on FDG PET/CT, and the mean apparent diffusion coefficient value on diffusion-weighted image, were considered together to build prediction models for high-risk pathologic features. Among 616 patients, 72 patients met the imaging criteria for high-risk lung cancer and underwent lung MRI. The magnetic resonance (MR)-eligible group showed a higher prevalence of nodal upstaging (29.2% vs. 4.2%, p < 0.001), vascular invasion (6.5% vs. 2.1%, p=0.011), high-grade pathologic features (p < 0.001), worse 4-year disease-free survival (p < 0.001) compared with non-MR-eligible group. The prediction power for MR-based radiomics model predicting high-risk pathologic features was good, with mean area under the receiver operating curve (AUC) value measuring 0.751-0.886 in test sets. Adding clinical variables increased the predictive performance for MPsol and the poorly differentiated pattern using the 2021 grading system (AUC, 0.860 and 0.907, respectively). Our imaging criteria can effectively screen high-risk lung cancer patients and predict high-risk pathologic features by our MR-based prediction model using radiomics.

High-Throughput Phenotyping for Agronomic Traits in Cassava Using Aerial Imaging.

Large-scale phenotyping using unmanned aerial vehicles (UAVs) has been considered an important tool for plant selection. This study aimed to estimate the correlations between agronomic data and vegetation indices (VIs) obtained at different flight heights and to select prediction models to evaluate the potential use of aerial imaging in cassava breeding programs. Various VIs were obtained and analyzed using mixed models to derive the best linear unbiased predictors, heritability parameters, and correlations with various agronomic traits. The VIs were also used to build prediction models for agronomic traits. Aerial imaging showed high potential for estimating plant height, regardless of flight height (r = 0.99), although lower-altitude flights (20 m) resulted in less biased estimates of this trait. Multispectral sensors showed higher correlations compared to RGB, especially for vigor, shoot yield, and fresh root yield (-0.40 ≤ r ≤ 0.50). The heritability of VIs at different flight heights ranged from moderate to high (0.51 ≤ HCullis2 ≤ 0.94), regardless of the sensor used. The best prediction models were observed for the traits of plant vigor and dry matter content, using the Generalized Linear Model with Stepwise Feature Selection (GLMSS) and the K-Nearest Neighbor (KNN) model. The predictive ability for dry matter content increased with flight height for the GLMSS model (R2 = 0.26 at 20 m and R2 = 0.44 at 60 m), while plant vigor ranged from R2 = 0.50 at 20 m to R2 = 0.47 at 40 m in the KNN model. Our results indicate the practical potential of implementing high-throughput phenotyping via aerial imaging for rapid and efficient selection in breeding programs.

Occupant-Aware Energy Consumption Prediction in Smart Buildings Using a LSTM Model and Time Series Data

Accurate energy consumption prediction in commercial buildings is a challenging research task. Energy prediction plays a crucial role in energy efficiency, management, planning, sustainability, risk management, diagnosis, and demand response. Although many studies have been conducted on building energy predictions, the impact of occupancy on energy prediction models for office-type commercial buildings remains insufficiently explored, despite its potential to improve energy efficiency by 20%. This study investigates energy prediction using a Long Short-Term Memory (LSTM) model that incorporates time-series power consumption data and considers occupancy. A real-world dataset containing the per-minute electricity consumption of various appliances in an office building in Houston, TX, USA, is utilized. The proposed machine learning models forecast future energy consumption based on hourly, 3-hourly, daily, and quarterly predictions for individual appliances and total energy usage. The model’s performance is evaluated using the following three metrics: Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). The results demonstrate the superiority of the proposed system.

Application of machine learning methods for predicting esophageal variceal bleeding in patients with cirrhosis.

To develop and compare machine learning models based on CT morphology features, serum biomarkers, and basic physical conditions to predict esophageal variceal bleeding. Two hundred twenty-four cirrhotic patients with esophageal variceal bleeding and non-bleeding were included in the retrospective study. Clinical and serum biomarkers were used in our study. In addition, the open-access segmentation model was used to generate segmentation masks of the liver and spleen. Four machine learning models based on selected features are used for building prediction models, and the diagnostic performances of models were measured using the receiver operator characteristic analysis. Two hundred twenty-four cirrhosis patients with esophageal varices, including 112 patients with bleeding (mean age 52.8 ± 11.5 years, range 18-80 years) and 112 patients with non-bleeding (mean age 57.3 ± 10.5 years, range 34-85 years). The two groups showed significant differences in standardized spleen volume, fibrinogen, alanine aminotransferase, aspartate aminotransferase, D-dimer, platelet, and age. The ratio of the training set to the test set was 8:2in our research, and the 5-fold cross-validation was used in the research. The AUCs of linear regression, random forest, support vector machine, and adaptive boosting were, respectively, 0.742, 0.854, 0.719, and 0.821 in the training set. For the test set, the AUCs of models were, respectively, 0.763, 0.818, 0.648, and 0.804. Our study used CT morphological measurements, serum biomarkers, and age to build machine learning models, and the random forest and adaptive boosting had potential added value in predictive model construction. Question Esophageal variceal bleeding is an intractable complication of liver cirrhosis. Early prediction and prevention of esophageal variceal bleeding is important for patients with liver cirrhosis. Findings It was feasible and clinically meaningful to construct machine learning models based on CT morphology features, serum biomarkers, and physical conditions to predict variceal bleeding. Clinical relevance Our study may provide a promising tool with which clinicians can conduct therapeutic decisions on fewer invasive procedures for the prediction of esophageal variceal bleeding.

Time series (ARIMA) as a tool to predict the temperature-humidity index in the dairy region of the northern desert of Mexico.

The environment in which an animal is situated can have a profound impact on its health, welfare, and productivity. This phenomenon is particularly evident in the case of dairy cattle, then, in order to quantify the impact of ambient temperature (°C) and the relative humidity (%) on dairy cattle, the temperature-humidity index (THI) is employed as a metric. This indicator enables the practical estimation of the stress imposed on cattle by ambient temperature and humidity. A seasonal autoregressive integrated moving average (SARIMA) (4,1,0)(0,1,0)365 model was estimated using daily data from the maximum daily THI of 4 years (2016-2019) of the Comarca Lagunera, an arid region of central-northern Mexico. The resulting model indicated that the THI of any given day in the area can be estimated based on the THI values of the previous four days. Furthermore, the data demonstrate an annual increase in the number of days the THI indicates a risk of heat stress. It is essential to continue building predictive models to develop effective strategies to mitigate the adverse effects of heat stress in dairy cattle (and other species) in the region.

Low-carbohydrate diet score and chronic obstructive pulmonary disease: a machine learning analysis of NHANES data.

Recent research has identified the Low-Carbohydrate Diet (LCD) score as a novel biomarker, with studies showing that LCDs can reduce carbon dioxide retention, potentially improving lung function. While the link between the LCD score and chronic obstructive pulmonary disease (COPD) has been explored, its relevance in the US population remains uncertain. This study aims to explore the association between the LCD score and the likelihood of COPD prevalence in this population. Data from 16,030 participants in the National Health and Nutrition Examination Survey (NHANES) collected between 2007 and 2023 were analyzed to examine the relationship between LCD score and COPD. Propensity score matching (PSM) was employed to reduce baseline bias. Weighted multivariable logistic regression models were applied, and restricted cubic spline (RCS) regression was used to explore possible nonlinear relationships. Subgroup analyses were performed to evaluate the robustness of the results. Additionally, we employed eight machine learning methods-Boost Tree, Decision Tree, Logistic Regression, MLP, Naive Bayes, KNN, Random Forest, and SVM RBF-to build predictive models and evaluate their performance. Based on the best-performing model, we further examined variable importance and model accuracy. Upon controlling for variables, the LCD score demonstrated a strong correlation with the odds of COPD prevalence. In compared to the lowest quartile, the adjusted odds ratios (ORs) for the high quartile were 0.77 (95% CI: 0.63, 0.95), 0.74 (95% CI: 0.59, 0.93), and 0.61 (95% CI: 0.48, 0.78). RCS analysis demonstrated a linear inverse relationship between the LCD score and the odds of COPD prevalence. Furthermore, the random forest model exhibited robust predictive efficacy, with an area under the curve (AUC) of 71.6%. Our study of American adults indicates that adherence to the LCD may be linked to lower odds of COPD prevalence. These findings underscore the important role of the LCD score as a tool for enhancing COPD prevention efforts within the general population. Nonetheless, additional prospective cohort studies are required to assess and validate these results.

Student Learning Satisfaction and Academic Performance in Philippine Science High School Chemistry: A Prediction Model Building Study for Online Learning

Learning satisfaction is a key indicator of enriching student learning whether it is face-to-face or online mode of delivery. Learning has shifted more to remote online learning due to the Covid-19 pandemic. This study aimed to identify the determinants of Student Learning Satisfaction (SLS) in online learning and develop a model that represented the relationship of the determinants to SLS. The study explored both internal and external determinants hypothesised to influence SLS among 169 students in grades 9-12 taking Chemistry at the Philippine Science High School Cordillera Administrative Region Campus (PSHS CARC). The internal factors included gender, academic self-concept (ASC), academic motivation (AM), interest to learn (IL), and mental well-being (MW), while the external factors comprised teacher-related indicators (Teacher), assessment-related indicators (Assessment), learning guide-related indicators (LG) and Knowledge Hub-related indicators (KHub). The study utilised Structural Equation Modelling (SEM) with SmartPLS software to develop the SLS model and Academic Performance (AP) model. Both online learning platforms used by PSHS CARC, the LG, and KHub, were found to influence SLS. The external determinants (LG, KHub, Teacher, Assessment) dominated the SLS models while internal determinants (ASC, AM, IL, MW) dominated the AP model for the PSHS CARC Chemistry students. The external determinants influenced the student's learning satisfaction while the internal determinants influenced the AP of the Chemistry students. Assessment indirectly affected SLS. Another finding was that gender, LG, and ASC directly influenced AP. The results validate that online learning tools (LG and KHub) were crucial determinants in ensuring learning satisfaction and academic success in the online learning of PSHS CARC Chemistry students. The models (SLS and AP) can be utilised to evaluate and enhance the quality of online learning and the development of effective online learning strategies and policies especially regarding the use of online tools like LG and KHub.

Reporting a regular medical doctor index: A new measure of patient-physician affiliation for health administrative data.

Having a regular medical doctor is associated with better process of care and health outcomes. The goal of this study was to harness the richness in health administrative data to create a measure which accurately predicted whether patients self-identified as having a regular medical doctor. The Canadian Community Health Survey (2007-2012) was linked with health administrative data (HAD) (2002-2012) from Quebec, Canada's second largest province. The Canadian Community Health Survey includes respondents' answer to whether they have a regular medical doctor, but health administrative data does not. We therefore used LASSO and Random Forests to build prediction models that predict whether a patient reports having a regular medical doctor using their data only available in the HAD. Our results show that predicting patient responses to 'do you have a regular medical doctor?' using an average of single-year Usual Provider Continuity over 3 years results in an area under the receiver operator characteristic curve of 0.782 (0.778-0.787). This was almost a 14% improvement in predictive accuracy compared to the frequently used single-year Usual Provider Continuity (0.688 (0.683-0.694)). We have called this new measure the Reporting a Regular Medical Doctor (RRMD) index. The RRMD index is easy to implement in HAD, is an elegant solution to the difficulties associated with low-users having unstable UPC scores, and brings a patient-oriented perspective to previous efforts to capture patient-physician affiliations in HAD. We recommend that researchers seeking to measure whether patients have a regular medical doctor using HAD consider using the RRMD index.

CT-based radiomics and cluster analysis for the prediction of local progression in stage I NSCLC patients treated with microwave ablation.

To predict local progression after microwave ablation (MWA) in patients with stage I non-small cell lung cancer (NSCLC), we developed a CT-based radiomics model. Postoperative CT images were used. The intraclass correlation coefficients, two-sample t-test, least absolute shrinkage and selection operator (LASSO) regression, and Pearson correlation analysis were applied to select radiomics features and establish radiomics score. The Radiomics score was used to classify patients into new radiomics labels. The k-means cluster algorithm was employed to cluster patients into new cluster labels based on radiomics features. Logistic regression was used to build prediction models. The optimal model incorporating clinical risk factors, radiomics labels, and cluster labels achieved the best discrimination. This study proposes a radiomics model that accurately predicts local progression in patients with stage I NSCLC treated with MWA. This prediction tool may be helpful in determining MWA efficacy and individualized risk classification and treatment.

A065: Analysis of the Prediction of Adolescent Depressive Diseases and the Mediating Effect of Exercise Factors Based on Basic Census Data

Objective: By analyzing census databases, we aimed to understand the correlation between depression and other psychological and physiological diseases. Further, we expanded our research to the correlation with physiology, lifestyle habits, and other aspects. At the same time, we attempted to use artificial intelligence methods to predict the risk of depression, obtain relevant models, and derive depression. Methods: Data from census databases were used in this research. Traditional data analysis methods were used for basic quantitative analysis, and highly correlated data with depression was selected for prediction model building. Results: (1) Analysis showed a significant correlation between depression and other diseases such as anxiety and hyperactivity disorder (=0.535, 0.231, 0.284; p &lt; 0.01). Besides, physiological and life-related issues may become the cause of depression. The prediction model established through big data can better predict and detect depressive tendencies. (2) There was a positive correlation between adolescent living standards and depressive psychology (r &gt; 0.85, r &gt; 0.87). Exercise levels and dietary levels were mediating variables between adolescent living standards and depressive psychology, with the mediating effect accounting for 33.6% of the total effect. Conclusion: Physiological trauma, personality traits, and lifestyle habits of close relationships among adolescents may become factors that trigger psychological disorders in adolescents. Teenagers' exercise and dietary levels play a mediating role between their living standards and depressive psychology.

Radiomic features of PET/CT imaging of large B cell lymphoma lesions predicts CAR T cell therapy efficacy.

Relapsed and refractory Diffuse large B cell lymphoma (DLBCL) can be successfully treated with axicabtagene ciloleucel (axi-cel), a CD19-directed autologous chimeric antigen receptor T cell (CAR-T) therapy. Diagnostic image-based features could help identify the patients who would clinically respond to this advanced immunotherapy. The aim of this study was to establish a radiomic image feature-based signature derived from positron emission tomography/computed tomography (PET/CT), including metabolic tumor burden, which can predict a durable response to CAR-T therapy in refractory/relapsed DLBCL. We conducted a retrospective review of 155 patients with relapsed/refractory DLBCL treated with axi-cel CAR-T therapy. The patients' disease involvement was evaluated based on nodal or extranodal sites. A sub-cohort of these patients with at least one nodal lesion (n=124) was assessed, while an overlapping sub-cohort (n=94) had at least one extranodal lesion. The lesion regions were characterized using 306 quantitative imaging metrics for PET images and CT images independently. Principal component (PC) analysis was performed to reduce the dimensionality in feature-based functional categories: size (n=38), shape (n=9), and texture (n=259). The selected features were used to build prediction models for survival at 1 year and tested for prognosis to overall/progression-free survival (OS/PFS) using a Kaplan-Meier (KM) plot. The Shape-based PC features of the largest extranodal lesion on PET were predictive of 1-year survival (AUC 0.68 [0.43,0.94]) and prognostic of OS/PFS (p<0.018). Metabolic tumor volume (MTV) was an independent predictor with an area under the curve (AUC) of 0.74 [0.58, 0.87]. Combining these features improved the predictor performance (AUC of 0.78 [0.7, 0.87]). Additionally, the Shape-based PC features were unrelated to total MTV (Spearman's ρ of 0.359, p≤ 0.001). Our study found that shape-based radiomic features on PET imaging were predictive of treatment outcome (1-year survival) and prognostic of overall survival. We also found non-size-based radiomic predictors that had comparable performance to MTV and provided complementary information to improve the predictability of treatment outcomes.

Abstract A057: Cancer-induced epigenomic changes in circulating immune cells enable breast, colorectal and lung cancer detection

Abstract Background: Epigenomic changes via the measurement of 5-hydroxymethylation cytosine (5hmC) in DNA have been shown to enable detection in a variety of cancers including pancreatic and ovarian cancer in plasma-derived cell-free DNA (cfDNA). Here, we investigated whether blood-based 5hmC profiling of genomic DNA (gDNA) derived from cells of the buffy coat (BC) possess signals that enable the detection of breast, colorectal and lung cancers. Method: Blood samples collected in Streck tubes from cancer patients or non-cancer controls were processed to obtain buffy coat and plasma samples. gDNA and cfDNA was isolated from buffy coat and plasma, respectively. gDNA or cfDNA was then subjected to 5hmC enrichment by click chemistry followed by preparation of sequencing libraries. Sequencing was performed to obtain genome-wide 5hmC profiles of buffy coat gDNA and cfDNA samples. Results: Differential hydroxymethylation analysis of BC gDNA revealed increased 5hmC over genes associated with myeloid lineage in breast, colorectal and lung cancers relative to non-cancer controls. Flow cytometry analysis confirmed increases in myeloid cells in cancer samples relative to non-cancer controls. 5hmC patterns in BC gDNA showed a larger number of cancer-associated changes compared to cfDNA, in particular this was found in early-stage disease. 5hmC features from BC gDNA or plasma-derived cfDNA were used to build prediction models that yielded cancer classification with overall sensitivity of 51.3% and 52.6% at 98% specificity, for BC or cfDNA, respectively. For early-stage cancer, the sensitivity at 98% specificity was 46.5% for the BC and 28.2% for cfDNA model. A final prediction model incorporating features from both tissue sources achieved superior classification (65.8% sensitivity) at 98% specificity, compared with using features from either tissue source alone, particularly for early-stage cancer which achieved 53.5% sensitivity of cancer detection. Conclusion: Our results demonstrate that cancer impacts the 5hmC profiles of buffy coat cells which can be leveraged for cancer detection using blood. Citation Format: Gulfem D Guler, Giuliana P Mognol, Yuhong Ning, Carolina Fraire, David Haan, Michael Kesling, Roger Malta, Melissa Peters, Anna Bergamaschi, Tierney Phillips, Shimul Chowdhury, Wayne Volkmuth, Samuel Levy. Cancer-induced epigenomic changes in circulating immune cells enable breast, colorectal and lung cancer detection [abstract]. In: Proceedings of the AACR Special Conference: Liquid Biopsy: From Discovery to Clinical Implementation; 2024 Nov 13-16; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2024;30(21_Suppl):Abstract nr A057.

Synergizing Machine Learning, Conceptual Density Functional Theory, and Biochemistry: No-Code Explainable Predictive Models for Mutagenicity in Aromatic Amines.

This study synergizes machine learning (ML) with conceptual density functional theory (CDFT) to develop OECD-compliant predictive models for the mutagenic activity of aromatic amines (AAs) with a fully No-Code methodology using a comprehensive data set of 251 AAs, Leave-One-Out-Cross-Validation (LOOCV), and three distinct data splits. Our research employs the GFN2-xTB method, known for its robustness and speed, to compute descriptors for procarcinogens and their activated metabolites in vacuum and aqueous phases. We evaluate the effectiveness of different theoretical definitions of electrophilicity within CDFT, namely, PSL, GCV, and CDP schemes, and the newly introduced Log QP descriptor to approximate Log P information. SPAARC, RandomTree, and JCHAID* ML methods were used to build explainable predictive models with highly robust internal validation (Avg. Correct Classifications = 76% and Avg. Kappa = 0.29) and external validation (Avg. Correct Classifications = 79% and Avg. Kappa = 0.33) metrics, and the results were compared to those of a two hidden layer Multilayer Perceptron. The results indicate that the second CDP definition for the electrophilicity in both vacuum and aqueous phases and also the newly presented Log QP descriptors are the most important ones for predicting the mutagenic activity of AA (namely ω+VacCDP2+, ω+AqCDP2+, and LogQP1+Vac, respectively). The results indicate that metabolic activation, aqueous solvent properties, and the CDP electrophilicity schemes and Log QP should be considered when building predictive models for the mutagenic activity of AA. This study offers a replicable, No-Code approach to QSAR research, making high-level ML and CDFT applications accessible to a broader audience. Future work will expand these methods to other compound families, enhancing predictive capabilities in the study of mutagenic activities and other biological phenomena.

Machine learning driven design of high-performance Al alloys

Aluminum (Al) alloys with both high strength and thermal conductivity (TC) are promising structural materials for wide application across different industries. Yet, design of such alloys is challenging, since strength and TC often share a trade-off. In this paper, we build prediction models for TC and ultimate tensile strength (UTS) of Al alloys using eXtreme gradient boosting (XGBoost) and support vector machine (SVM) algorithms, respectively. The models take physical descriptors from the alloy composition into account. Lasso and Gini Impurity algorithms were adopted for feature engineering. Guided by the models, an Al-2.64Si-0.43Mg-0.10Zn-0.03Cu alloy with TC over 190 W·m-1·K-1 and UTS over 220 MPa was designed. The alloy was fabricated and tested by experiment, and its UTS and TC are close to the model prediction. Microstructure characterization suggests that the fragmented and spherical Si phase, along with a few non-spherical Si phases, may be a key reason for the improved properties.

Repairs and Breaks Prediction for Deep Neural Networks

With the increasing prevalence of software incorporating deep neural networks (DNNs), quality assurance for these software systems has become a crucial concern. To this end, various methods have been proposed to repair the misbehavior of DNNs by modifying their weights. However, these repair methods may not meet the developer's needs for a given dataset and model. In this study, we build prediction models for repair outcomes (i.e., repairs and breaks) to help determine whether the repair method is likely to work. By using our prediction models, developers and operators of DNNs can decide whether or not to apply a repair method, and if so, which method to use. Our prediction models utilize four metrics as explanatory metrics that represent the confidence or ambiguity in the DNN predictions. We experimented with four repair methods and 10 datasets. The experimental results demonstrate that our prediction models successfully select a repair method that meets developers’ needs in 16 out of 24 cases, resulting in an average time saving of 16.29% compared to the naive method. Based on these results, our prediction models can reduce costs for developers and operators when deciding whether to employ repair methods for real-world applications of DNNs.

Use of Patient-Reported Outcomes in Risk Prediction Model Development to Support Cancer Care Delivery: A Scoping Review.

The integration of patient-reported outcomes (PROs) into electronic health records (EHRs) has enabled systematic collection of symptom data to manage post-treatment symptoms. The use and integration of PRO data into routine care are associated with overall treatment success, adherence, and satisfaction. Clinical trials have demonstrated the prognostic value of PROs including physical function and global health status in predicting survival. It is unknown to what extent routinely collected PRO data are used in the development of risk prediction models (RPMs) in oncology care. The objective of the scoping review is to assess how PROs are used to train risk RPMs to predict patient outcomes in oncology care. Using the scoping review methodology outlined in the Joanna Briggs Institute Manual for Evidence Synthesis, we searched four databases (MEDLINE, CINAHL, Embase, and Web of Science) to locate peer-reviewed oncology articles that used PROs as predictors to train models. Study characteristics including settings, clinical outcomes, and model training, testing, validation, and performance data were extracted for analyses. Of the 1,254 studies identified, 18 met inclusion criteria. Most studies performed retrospective analyses of prospectively collected PRO data to build prediction models. Post-treatment survival was the most common outcome predicted. Discriminative performance of models trained using PROs was better than models trained without PROs. Most studies did not report model calibration. Systematic collection of PROs in routine practice provides an opportunity to use patient-reported data to develop RPMs. Model performance improves when PROs are used in combination with other comprehensive data sources.