Interpretable Machine Learning Models Research Articles

Projecting Large Fires in the Western US With an Interpretable and Accurate Hybrid Machine Learning Method

AbstractMore frequent and widespread large fires are occurring in the western United States (US), yet reliable methods for predicting these fires, particularly with extended lead times and a high spatial resolution, remain challenging. In this study, we proposed an interpretable and accurate hybrid machine learning (ML) model, that explicitly represented the controls of fuel flammability, fuel availability, and human suppression effects on fires. The model demonstrated notable accuracy with a F1‐score of 0.846 ± 0.012, surpassing process‐driven fire danger indices and four commonly used ML models by up to 40% and 9%, respectively. More importantly, the ML model showed remarkably higher interpretability relative to other ML models. Specifically, by demystifying the “black box” of each ML model using the explainable AI techniques, we identified substantial structural differences across ML fire models, even among those with similar accuracy. The relationships between fires and their drivers, identified by our model, were aligned closer with established fire physical principles. The ML structural discrepancy led to diverse fire predictions and our model predictions exhibited greater consistency with actual fire occurrence. With the highly interpretable and accurate model, we revealed the strong compound effects from multiple climate variables related to evaporative demand, energy release component, temperature, and wind speed, on the dynamics of large fires and megafires in the western US. Our findings highlight the importance of assessing the structural integrity of models in addition to their accuracy. They also underscore the critical need to address the rise in compound climate extremes linked to large wildfires.

Interpretable machine learning models for predicting probabilistic axial buckling strength of steel circular hollow section members considering discreteness of geometries and material

Steel circular hollow section (CHS) members are widely utilized as axial force-resisting structural members in civil engineering structures. The buckling strength under axial loads is one of the critical parameters to determine the performance of the steel CHS members, which is significantly affected by the discreteness introduced by geometries, material, and initial imperfections. However, the reduction factor employed in the modern design codes (i.e. Chinese codes and EC3) only accounts for the reduction caused by all kinds of discreteness and does not reflect the impacts of every single discreteness and imperfection. To fill the gap, this paper proposed an interpretable machine-learning method to provide the probabilistic axial buckling strength of steel CHS members prediction result in a distribution form with the consideration of detailed discreteness. The model to predict the nominal axial buckling strength of steel CHS members was first developed utilizing ten machine learning algorithms after sufficient numerical simulations, where the numerical model was verified using test results. The artificial neural network (ANN) was selected for developing the prediction model due to its highly reliable performance in testing. The developed ANN models were further interpreted utilizing Shapley Additive exPlanations (SHAP) to determine the interrelationship of different parameters. Then, the probabilistic axial bucking strength prediction model was established based on the developed ANN models, where the Latin hypercube sampling method was applied to address the discreteness of geometries, material, and initial imperfections. The generated probabilistic axial bucking strength prediction model’s effectiveness was verified by the evidence that the machine learning prediction results can highly match the numerical results' probability density function and the result from codes while significantly reducing the computation time. Finally, the design parameters’ impact on the axial buckling strength’s discreteness was evaluated using the global sensitivity analysis (GSA) method. The result shows that the discreteness of design parameters substantially influences the distribution of the axial buckling strength of the steel CHS members and the proposed prediction model can provide an accurate probabilistic distribution prediction.

Urban greenery distribution and its link to social vulnerability

Urban greenery plays a pivotal role in urban environments, impacting the environmental well-being and people’s comfort. Several studies have demonstrated a strong link between urban greenery and socioeconomic status but still lack an analysis of greenery on uneven distribution in social vulnerability. This study assesses how multi-level greenery rates distribute and associate the social vulnerability of people in 429 census tracts in the Seattle metropolitan area. It integrates multi-source urban informatics data, including remote sensing data and street view imagery, to identify various vegetation types. Then, it uses the interpretable machine learning model to explore the relationship between street-level green space distribution and community vulnerability. The results show a serious problem of uneven distribution of green spaces in urban centers since urban areas are built up and fragmented the landscape. Areas with low urban greening in the Seattle area have higher rates of poverty, unemployment, racial segregation, and housing overcrowding. Besides, greening features like street green views, which are more related to human perception, have a great association with social vulnerability. These findings contribute to the urban green spaces to better promote community equity and vulnerability.

Graft Failure and Contralateral ACL Injuries After Primary ACL Reconstruction: An Analysis of Risk Factors Using Interpretable Machine Learning.

Anterior cruciate ligament (ACL) reconstruction (ACLR) can be successful in restoring knee stability. However, secondary ACL injury, either through graft failure or contralateral injury, is a known complication and can significantly impact the ability of a patient to return successfully to previous activities. To develop and internally validate an interpretable machine learning model to quantify the risk of graft failure and contralateral ACL injury in a longitudinal cohort treated with ACLR. Case-control study; Level of evidence, 3. An established geographic database of >600,000 patients was used to identify patients with a diagnosis of ACL rupture between 1990 and 2016 with a minimum 2-year follow-up. Medical records were reviewed for relevant patient information and 4 candidate machine learning algorithms were evaluated for prediction of graft failure and contralateral ACL injury in patients after ACLR as identified either on magnetic resonance imaging or via arthroscopy. Performance of the algorithms was assessed through discrimination, calibration, and decision curve analysis. Model interpretability was enhanced utilizing global variable importance plots and partial dependence curves. A total of 1497 patients met inclusion criteria. Among them, 140 (9.4%) had graft failure and 128 (8.6%) had a contralateral ACL injury after index surgery at a median follow-up of 140.7 months (interquartile range, 77.2-219.2 months). The best performing models achieved an area under the receiver operating characteristics curve of 0.70 for prediction of graft failure and 0.67 for prediction of contralateral ACL injury, outperforming a logistic regression fitted on the identical feature set. Notable predictors for increased risk of graft failure included younger age at injury, body mass index (BMI) <30, return to sports <13 months, initial time to surgery >75 days, utilization of allograft, femoral/tibial fixation with suspension/expansion devices, concomitant collateral ligament injury, and active or former smoking history. Predictors of contralateral ACL injury included greater preoperative pain, younger age at initial injury, BMI <30, active smoking history, initial time to surgery >75 days, history of contralateral knee arthroscopies, and involvement in contact sports. Less than 18% of all patients who undergo ACLR should be expected to sustain either a graft failure or contralateral ACL injury. Machine learning models outperformed logistic regression and identified greater preoperative pain, younger age, BMI <30, earlier return to higher activity, and time to surgical intervention >75 days as common risk factors for both graft failure as well as contralateral ACL injury after ACLR. Surgeon-modifiable risk factors for graft failure included allograft and femoral/tibial fixation with a suspension/expansion combination.

Interpreting machine learning models based on SHAP values in predicting suspended sediment concentration

Machine learning (ML) has become a powerful tool for predicting suspended sediment concentration (SSC). Nonetheless, the ability to interpret the physical process is considered the main issue in applying most of ML approaches. In this regard, the current study presents a novel framework involving four standalone ML models (extra trees (ET), random forest (RF), categorical boosting (CatBoost), and extreme gradient boosting (XGBoost)) and their combination with genetic programming (GP). Three metrics (coefficient of correlation (r), root mean square error (RMSE), and Nash–Sutcliffe model-fit efficiency (NSE)) and a more advanced interpretation system SHapley Additive exPlanations (SHAP) are used to assess the performance of these models applied to hydro-climatic datasets for prediction of SSC. The calibration process was based on data from 2016 to 2020, and the validation was done for 2021 data. Further description and application of the framework are provided based on a case study of the Bouregreg watershed. The results revealed that all implemented models are efficient in SSC prediction with NSE, RMSE, and r varying from 0.53 to 0.86, 1.20 to 2.55 g/L, and 0.83 to 0.91 g/L respectively. Box plot diagrams confirm the enhanced performance of these combined models, and the best-performing ones for the four hydrological stations being the combined RF+GP model at the Aguibat Ziar station, the combined XGBoost+GP model at the Ain Loudah station, the CatBoost model at the Ras Fathia station, and the RF model at the Sidi Med Cherif station. The interpretability results showed that flow (Q) and seasonality (S) are the features most impacting SSC. These outcomes indicate that the applied models can extract accurate and detailed information from the interactions between the hydroclimatic factors and the generation of sediment by erosion (output). ML approaches illustrated the good reliability and transparency of the models developed for predicting SSC in a semi-arid setting, offered new perspectives for reducing ML models' "black box" character, and provided a useful source of information for assessing the consequences of SSC on water quality. The SHAP system and exploring other interpretable techniques are recommended to provide further information in future research. In addition, incorporating additional input data could enhance SSC predictions and deepen understanding of sediment transport dynamics.

Enhancing severe hypoglycemia prediction in type 2 diabetes mellitus through multi-view co-training machine learning model for imbalanced dataset

Patients with type 2 diabetes mellitus (T2DM) who have severe hypoglycemia (SH) poses a considerable risk of long-term death, especially among the elderly, demanding urgent medical attention. Accurate prediction of SH remains challenging due to its multifaced nature, contributed from factors such as medications, lifestyle choices, and metabolic measurements. In this study, we propose a systematic approach to improve the robustness and accuracy of SH predictions using machine learning models, guided by clinical feature selection. Our focus is on developing long-term SH prediction models using both semi-supervised learning and supervised learning algorithms. Using the action to control cardiovascular risk in diabetes trial, which includes electronic health records for over 10,000 individuals, we focus on studying adults with T2DM. Our results indicate that the application of a multi-view co-training method, incorporating the random forest algorithm, improves the specificity of SH prediction, while the same setup with Naive Bayes replacing random forest demonstrates better sensitivity. Our framework also provides interpretability of machine learning models by identifying key predictors for hypoglycemia, including fasting plasma glucose, hemoglobin A1c, general diabetes education, and NPH or L insulins. The integration of data routinely available in electronic health records significantly enhances our model’s capability to predict SH events, showcasing its potential to transform clinical practice by facilitating early interventions and optimizing patient management. By enhancing prediction accuracy and identifying crucial predictive features, our study contributes to advancing the understanding and management of hypoglycemia in this population.

Counterfactual Explanations for Medical Image Classification and Regression using Diffusion Autoencoder

Counterfactual explanations (CEs) aim to enhance the interpretability of machine learning models by illustrating how alterations in input features would affect the resulting predictions. Common CE approaches require an additional model and are typically constrained to binary counterfactuals. In contrast, we propose a novel method that operates directly on the latent space of a generative model, specifically a Diffusion Autoencoder (DAE). This approach offers inherent interpretability by enabling the generation of CEs and the continuous visualization of the model’s internal representation across decision boundaries. Our method leverages the DAE’s ability to encode images into a semantically rich latent space in an unsupervised manner, eliminating the need for labeled data or separate feature extraction models. We show that these latent representations are helpful for medical condition classification and the ordinal regression of severity pathologies, such as vertebral compression fractures (VCF) and diabetic retinopathy (DR). Beyond binary CEs, our method supports the visualization of ordinal CEs using a linear model, providing deeper insights into the model’s decision-making process and enhancing interpretability. Experiments across various medical imaging datasets demonstrate the method’s advantages in interpretability and versatility. The linear manifold of the DAE’s latent space allows for meaningful interpolation and manipulation, making it a powerful tool for exploring medical image properties. Our code is available at &lt;a href='https://github.com/matanat/dae_counterfactual'&gt;https://github.com/matanat/dae_counterfactual&lt;/a&gt;

Polynomial-SHAP analysis of liver disease markers for capturing of complex feature interactions in machine learning models

Liver disease diagnosis is pivotal for effective patient management, and machine learning techniques have shown promise in this domain. In this study, we investigate the impact of Polynomial-SHapley Additive exPlanations analysis on enhancing the performance and interpretability of machine learning models for liver disease classification. Our results demonstrate significant improvements in accuracy, precision, recall, F1_score, and Matthews correlation coefficient across various algorithms when polynomial- SHapley Additive exPlanations analysis is applied. Specifically, the Light Gradient Boosting Machine model achieves exceptional performance with 100 % accuracy in both scenarios. Furthermore, by comparing the results obtained with and without the approach, we observe substantial differences in the performance, highlighting the importance of incorporating Polynomial-SHapley Additive exPlanations analysis for improved model performance. The Polynomial features and SHapley Additive exPlanations values also enhance the interpretability of machine learning models by capturing complex feature interactions, enabling users to gain deeper insights into the underlying mechanisms driving the diagnosis. Moreover, data rebalancing using Synthetic Minority Over-sampling Technique and parameter tuning were employed to optimize the performance of the models. These findings underscore the significance of employing this analytical approach in machine-learning-based diagnostic systems for liver diseases, offering superior performance and enhanced interpretability for informed decision-making in clinical practice.

Interpretable machine learning models for the prediction of all-cause mortality and time to death in hemodialysis patients.

The elevated mortality and hospitalization rates among hemodialysis (HD) patients underscore the necessity for the development of accurate predictive tools. This study developed two models for predicting all-cause mortality and time to death-one using a comprehensive database and another simpler model based on demographic and clinical data without laboratory tests. A retrospective cohort study was conducted from January 2017 to June 2023. Two models were created: Model A with 85 variables and Model B with 22 variables. We assessed the models using random forest (RF), support vector machine, and logistic regression, comparing their performance via the AU-ROC. The RF regression model was used to predict time to death. To identify the most relevant factors for prediction, the Shapley value method was used. Among 359 HD patients, the RF model provided the most reliable prediction. The optimized Model A showed an AU-ROC of 0.86 ± 0.07, a sensitivity of 0.86, and a specificity of 0.75 for predicting all-cause mortality. It also had an R2 of 0.59 for predicting time to death. The optimized Model B had an AU-ROC of 0.80 ± 0.06, a sensitivity of 0.81, and a specificity of 0.70 for predicting all-cause mortality. In addition, it had an R2 of 0.81 for predicting time to death. Two new interpretable clinical tools have been proposed to predict all-cause mortality and time to death in HD patients using machine learning models. The minimal and readily accessible data on which Model B is based makes it a valuable tool for integrating into clinical decision-making processes.

Understanding machine learning predictions of wastewater treatment plant sludge with explainable artificial intelligence.

This study investigates the use of machine learning (ML) models for wastewater treatment plant (WWTP) sludge predictions and explainable artificial intelligence (XAI) techniques for understanding the impact of variables behind the prediction. Three ML models, random forest (RF), gradient boosting machine (GBM), and gradient boosting tree (GBT), were evaluated for their performance using statistical indicators. Input variable combinations were selected through different feature selection (FS) methods. XAI techniques were employed to enhance the interpretability and transparency of ML models. The results suggest that prediction accuracy depends on the choice of model and the number of variables. XAI techniques were found to be effective in interpreting the decisions made by each ML model. This study provides an example of using ML models in sludge production prediction and interpreting models applying XAI to understand the factors influencing it. Understandable interpretation of ML model prediction can facilitate targeted interventions for process optimization and improve the efficiency and sustainability of wastewater treatment processes. PRACTITIONER POINTS: Explainable artificial intelligence can play a crucial role in promoting trust between machine learning models and their real-world applications. Widely practiced machine learning models were used to predict sludge production of a United States wastewater treatment plant. Feature selection methods can reduce the required number of input variables without compromising model accuracy. Explainable artificial intelligence techniques can explain driving variables behind machine learning prediction.

Predicting bacterial fitness in Mycobacterium tuberculosis with transcriptional regulatory network-informed interpretable machine learning.

Mycobacterium tuberculosis (Mtb) is the causative agent of tuberculosis disease, the greatest source of global mortality by a bacterial pathogen. Mtb adapts and responds to diverse stresses such as antibiotics by inducing transcriptional stress-response regulatory programs. Understanding how and when these mycobacterial regulatory programs are activated could enable novel treatment strategies for potentiating the efficacy of new and existing drugs. Here we sought to define and analyze Mtb regulatory programs that modulate bacterial fitness. We assembled a large Mtb RNA expression compendium and applied these to infer a comprehensive Mtb transcriptional regulatory network and compute condition-specific transcription factor activity profiles. We utilized transcriptomic and functional genomics data to train an interpretable machine learning model that can predict Mtb fitness from transcription factor activity profiles. We demonstrated that this transcription factor activity-based model can successfully predict Mtb growth arrest and growth resumption under hypoxia and reaeration using only RNA-seq expression data as a starting point. These integrative network modeling and machine learning analyses thus enable the prediction of mycobacterial fitness under different environmental and genetic contexts. We envision these models can potentially inform the future design of prognostic assays and therapeutic intervention that can cripple Mtb growth and survival to cure tuberculosis disease.

Advanced High-Throughput Rational Design of Porphyrin-Sensitized Solar Cells Using Interpretable Machine Learning.

Accurately predicting the power conversion efficiency (PCE) in dye-sensitized solar cells (DSSCs) represents a crucial challenge, one that is pivotal for the high throughput rational design and screening of promising dye sensitizers. This study presents precise, predictive, and interpretable machine learning (ML) models specifically designed for Zn-porphyrin-sensitized solar cells. The model leverages theoretically computable, effective, and reusable molecular descriptors (MDs) to address this challenge. The models achieve excellent performance on a "blind test" of 17 newly designed cells, with a mean absolute error (MAE) of 1.02%. Notably, 10 dyes are predicted within a 1% error margin. These results validate the ML models and their importance in exploring uncharted chemical spaces of Zn-porphyrins. SHAP analysis identifies crucial MDs that align well with experimental observations, providing valuable chemical guidelines for the rational design of dyes in DSSCs. These predictive ML models enable efficient in silico screening, significantly reducing analysis time for photovoltaic cells. Promising Zn-porphyrin-based dyes with exceptional PCE are identified, facilitating high-throughput virtual screening. The prediction tool is publicly accessible at https://ai-meta.chem.ncu.edu.tw/dsc-meta.

Personalized Prediction of Long-Term Renal Function Prognosis Following Nephrectomy Using Interpretable Machine Learning Algorithms: Case-Control Study.

Acute kidney injury (AKI) is a common adverse outcome following nephrectomy. The progression from AKI to acute kidney disease (AKD) and subsequently to chronic kidney disease (CKD) remains a concern; yet, the predictive mechanisms for these transitions are not fully understood. Interpretable machine learning (ML) models offer insights into how clinical features influence long-term renal function outcomes after nephrectomy, providing a more precise framework for identifying patients at risk and supporting improved clinical decision-making processes. This study aimed to (1) evaluate postnephrectomy rates of AKI, AKD, and CKD, analyzing long-term renal outcomes along different trajectories; (2) interpret AKD and CKD models using Shapley Additive Explanations values and Local Interpretable Model-Agnostic Explanations algorithm; and (3) develop a web-based tool for estimating AKD or CKD risk after nephrectomy. We conducted a retrospective cohort study involving patients who underwent nephrectomy between July 2012 and June 2019. Patient data were randomly split into training, validation, and test sets, maintaining a ratio of 76.5:8.5:15. Eight ML algorithms were used to construct predictive models for postoperative AKD and CKD. The performance of the best-performing models was assessed using various metrics. We used various Shapley Additive Explanations plots and Local Interpretable Model-Agnostic Explanations bar plots to interpret the model and generated directed acyclic graphs to explore the potential causal relationships between features. Additionally, we developed a web-based prediction tool using the top 10 features for AKD prediction and the top 5 features for CKD prediction. The study cohort comprised 1559 patients. Incidence rates for AKI, AKD, and CKD were 21.7% (n=330), 15.3% (n=238), and 10.6% (n=165), respectively. Among the evaluated ML models, the Light Gradient-Boosting Machine (LightGBM) model demonstrated superior performance, with an area under the receiver operating characteristic curve of 0.97 for AKD prediction and 0.96 for CKD prediction. Performance metrics and plots highlighted the model's competence in discrimination, calibration, and clinical applicability. Operative duration, hemoglobin, blood loss, urine protein, and hematocrit were identified as the top 5 features associated with predicted AKD. Baseline estimated glomerular filtration rate, pathology, trajectories of renal function, age, and total bilirubin were the top 5 features associated with predicted CKD. Additionally, we developed a web application using the LightGBM model to estimate AKD and CKD risks. An interpretable ML model effectively elucidated its decision-making process in identifying patients at risk of AKD and CKD following nephrectomy by enumerating critical features. The web-based calculator, found on the LightGBM model, can assist in formulating more personalized and evidence-based clinical strategies.

Leveraging interpretable machine learning in intensive care

AbstractIn healthcare, especially within intensive care units (ICU), informed decision-making by medical professionals is crucial due to the complexity of medical data. Healthcare analytics seeks to support these decisions by generating accurate predictions through advanced machine learning (ML) models, such as boosted decision trees and random forests. While these models frequently exhibit accurate predictions across various medical tasks, they often lack interpretability. To address this challenge, researchers have developed interpretable ML models that balance accuracy and interpretability. In this study, we evaluate the performance gap between interpretable and black-box models in two healthcare prediction tasks, mortality and length-of-stay prediction in ICU settings. We focus specifically on the family of generalized additive models (GAMs) as powerful interpretable ML models. Our assessment uses the publicly available Medical Information Mart for Intensive Care dataset, and we analyze the models based on (i) predictive performance, (ii) the influence of compact feature sets (i.e., only few features) on predictive performance, and (iii) interpretability and consistency with medical knowledge. Our results show that interpretable models achieve competitive performance, with a minor decrease of 0.2–0.9 percentage points in area under the receiver operating characteristic relative to state-of-the-art black-box models, while preserving complete interpretability. This remains true even for parsimonious models that use only 2.2 % of patient features. Our study highlights the potential of interpretable models to improve decision-making in ICUs by providing medical professionals with easily understandable and verifiable predictions.

Assessing the nonlinear impact of green space exposure on psychological stress perception using machine learning and street view images.

Urban green space (GS) exposure is recognized as a nature-based strategy for addressing urban challenges. However, the stress relieving effects and mechanisms of GS exposure are yet to be fully explored. The development of machine learning and street view images offers a method for large-scale measurement and precise empirical analysis. This study focuses on the central area of Shanghai, examining the complex effects of GS exposure on psychological stress perception. By constructing a multidimensional psychological stress perception scale and integrating machine learning algorithms with extensive street view images data, we successfully developed a framework for measuring urban stress perception. Using the scores from the psychological stress perception scale provided by volunteers as labeled data, we predicted the psychological stress perception in Shanghai's central urban area through the Support Vector Machine (SVM) algorithm. Additionally, this study employed the interpretable machine learning model eXtreme Gradient Boosting (XGBoost) algorithm to reveal the nonlinear relationship between GS exposure and residents' psychological stress. Results indicate that the GS exposure in central Shanghai is generally low, with significant spatial heterogeneity. GS exposure has a positive impact on reducing residents' psychological stress. However, this effect has a threshold; when GS exposure exceeds 0.35, its impact on stress perception gradually diminishes. We recommend combining the threshold of stress perception with GS exposure to identify urban spaces, thereby guiding precise strategies for enhancing GS. This research not only demonstrates the complex mitigating effect of GS exposure on psychological stress perception but also emphasizes the importance of considering the "dose-effect" of it in urban planning and construction. Based on open-source data, the framework and methods developed in this study have the potential to be applied in different urban environments, thus providing more comprehensive support for future urban planning.

Tissue-aware interpretation of genetic variants advances the etiology of rare diseases.

Pathogenic variants underlying Mendelian diseases often disrupt the normal physiology of a few tissues and organs. However, variant effect prediction tools that aim to identify pathogenic variants are typically oblivious to tissue contexts. Here we report a machine-learning framework, denoted "Tissue Risk Assessment of Causality by Expression for variants" (TRACEvar, https://netbio.bgu.ac.il/TRACEvar/ ), that offers two advancements. First, TRACEvar predicts pathogenic variants that disrupt the normal physiology of specific tissues. This was achieved by creating 14 tissue-specific models that were trained on over 14,000 variants and combined 84 attributes of genetic variants with 495 attributes derived from tissue omics. TRACEvar outperformed 10 well-established and tissue-oblivious variant effect prediction tools. Second, the resulting models are interpretable, thereby illuminating variants' mode of action. Application of TRACEvar to variants of 52 rare-disease patients highlighted pathogenicity mechanisms and relevant disease processes. Lastly, the interpretation of alltissue models revealed that top-ranking determinants of pathogenicity included attributes of disease-affected tissues, particularly cellular process activities. Collectively, these results show thattissue contexts and interpretable machine-learning models can greatly enhance the etiology of rare diseases.

Machine learning approaches to identify the link between heavy metal exposure and ischemic stroke using the US NHANES data from 2003 to 2018.

There is limited understanding of the link between exposure to heavy metals and ischemic stroke (IS). This research aimed to develop efficient and interpretable machine learning (ML) models to associate the relationship between exposure to heavy metals and IS. The data of this research were obtained from the National Health and Nutrition Examination Survey (US NHANES, 2003-2018) database. Seven ML models were used to identify IS caused by exposure to heavy metals. To assess the strength of the models, we employed 10-fold cross-validation, the area under the curve (AUC), F1 scores, Brier scores, Matthews correlation coefficient (MCC), precision-recall (PR) curves, and decision curve analysis (DCA) curves. Following these tests, the best-performing model was selected. Finally, the DALEX package was used for feature explanation and decision-making visualization. A total of 15,575 participants were involved in this study. The best-performing ML models, which included logistic regression (LR) (AUC: 0.796) and XGBoost (AUC: 0.789), were selected. The DALEX package revealed that age, total mercury in blood, poverty-to-income ratio (PIR), and cadmium were the most significant contributors to IS in the logistic regression and XGBoost models. The logistic regression and XGBoost models showed high efficiency, accuracy, and robustness in identifying associations between heavy metal exposure and IS in NHANES 2003-2018 participants.

Integrated germline and somatic features reveal divergent immune pathways driving response to immune checkpoint blockade.

Immune Checkpoint Blockade (ICB) has revolutionized cancer treatment, however the mechanisms determining patient response remain poorly understood. Here, we used machine learning to predict ICB response from germline and somatic biomarkers and interpreted the learned model to uncover putative mechanisms driving superior outcomes. Patients with higher infiltration of T follicular helper cells had responses even in the presence of defects in the class-I Major Histocompatibility Complex (MHC-I). Further investigation uncovered different ICB responses in tumors when responses were reliant on MHC-I versus MHC-II neoantigens. Despite similar response rates, MHC-II reliant responses were associated with significantly longer durable clinical benefit (Discovery: Median OS=63.6 vs. 34.5 months P=0.0074; Validation: Median OS=37.5 vs. 33.1 months, P=0.040). Characteristics of the tumor immune microenvironment reflected MHC neoantigen reliance, and analysis of immune checkpoints revealed LAG3 as a potential target in MHC-II but not MHC-I reliant responses. This study highlights the value of interpretable machine learning models in elucidating the biological basis of therapy responses.

Interpretable Machine Learning Models for Predicting Critical Outcomes in Patients with Suspected Urinary Tract Infection with Positive Urine Culture.

(1) Background: Urinary tract infection (UTI) is a leading cause of emergency department visits and hospital admissions. Despite many studies identifying UTI-related risk factors for bacteremia or sepsis, a significant gap remains in developing predictive models for in-hospital mortality or the necessity for emergent intensive care unit admission in the emergency department. This study aimed to construct interpretable machine learning models capable of identifying patients at high risk for critical outcomes. (2) Methods: This was a retrospective study of adult patients with urinary tract infection (UTI), extracted from the Medical Information Mart for Intensive Care IV Emergency Department (MIMIC-IV-ED) database. The critical outcome is defined as either in-hospital mortality or transfer to an intensive care unit within 12 h. ED visits were randomly partitioned into a 70%/30% split for training and validation. The extreme gradient boosting (XGBoost), random forest (RF), and support vector machine (SVM) algorithms were constructed using variables selected from the stepwise logistic regression model. The XGBoost model was then compared to the traditional model and clinical decision rules (CDRs) on the validation data using the area under the curve (AUC). (3) Results: There were 3622 visits among 3235 unique patients diagnosed with UTI. Of the 2535 patients in the training group, 836 (33%) experienced critical outcomes, and of the 1087 patients in the validation group, 358 (32.9%) did. The AUCs for different machine learning models were as follows: XGBoost, 0.833; RF, 0.814; and SVM, 0.799. The XGBoost model performed better than others. (4) Conclusions: Machine learning models outperformed existing traditional CDRs for predicting critical outcomes of ED patients with UTI. Future research should prospectively evaluate the effectiveness of this approach and integrate it into clinical practice.

Interpretable machine learning models for ESG stock prices under transition and physical climate risk

Interpretable machine learning models for ESG stock prices under transition and physical climate risk