Interpretable Machine Learning Research Articles

Rapid Identification of Geographical Origin of Sliced Polygonati Rhizoma by Auto-focus Laser-induced Breakdown Spectroscopy Combined with Interpretable Machine Learning

Rapid Identification of Geographical Origin of Sliced Polygonati Rhizoma by Auto-focus Laser-induced Breakdown Spectroscopy Combined with Interpretable Machine Learning

Predicting the compressive strength of high-performance concrete using an interpretable machine learning model.

Reliable predictions of concrete strength can reduce construction time and labor costs, providing strong support for building construction quality inspection. To enhance the accuracy of concrete strength prediction, this paper introduces an interpretable framework for machine learning (ML) models to predict the compressive strength of high-performance concrete (HPC). By leveraging information from a concrete dataset, an additional six features were added as inputs in the training process of the random forest (RF), AdaBoost, XGBoost and LightGBM models, and the optimal hyperparameters of the models were determined using 5-fold cross-validation and random search methods. Four interpretable ML models for predicting the compressive strength of HPC, including the RF, AdaBoost, XGBoost and LightGBM models, which combine feature derivation and random search, were constructed. In addition, the SHapley Additive exPlanations (SHAP) method was applied to analyze the effects of the input features on the prediction results of the LightGBM model, which combines feature derivation and random search. The results showed that input features such as age, water/cement ratio, slag, and water were the key influences for predicting the compressive strength of HPC. Input features such as the superplastic/cement ratio, slag/cement ratio, and ash/cement ratio had nonsignificant impacts on the predicted compressive strength.

Improved prognostication of overall survival after radiotherapy in lung cancer patients by an interpretable machine learning model integrating lung and tumor radiomics and clinical parameters.

Accurate prognostication of overall survival (OS) for non-small cell lung cancer (NSCLC) patients receiving definitive radiotherapy (RT) is crucial for developing personalized treatment strategies. This study aims to construct an interpretable prognostic model that combines radiomic features extracted from normal lung and from primary tumor with clinical parameters. Our model aimed to clarify the complex, nonlinear interactions between these variables and enhance prognostic accuracy. We included 661 stage III NSCLC patients from three multi-national datasets: a training set (N = 349), test-set-1 (N = 229), and test-set-2 (N = 83), all undergoing definitive RT. A total of 104 distinct radiomic features were separately extracted from the regions of interest in the lung and the tumor. We developed four predictive models using eXtreme gradient boosting and selected the top 10 features based on the Shapley additive explanations (SHAP) values. These models were the tumor radiomic model (Model-T), lung radiomic model (Model-L), a combined radiomic model (Model-LT), and an integrated model incorporating clinical parameters (Model-LTC). Model performance was evaluated through Harrell's concordance index, Kaplan-Meier survival curves, time-dependent area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis. Interpretability was assessed using the SHAP framework. Model-LTC exhibited superior performance, with notable predictive accuracy (C-index: training set, 0.87; test-set-2, 0.76) and time-dependent AUC above 0.75. Complex nonlinear relationships and interactions were evident among the model's variables. The integration of radiomic and clinical factors within an interpretable framework significantly improved OS prediction. The SHAP analysis provided insightful interpretability, enhancing the model's clinical applicability and potential for aiding personalized treatment decisions.

On Leveraging Machine Learning in Sport Science in the Hypothetico-deductive Framework.

Supervised machine learning (ML) offers an exciting suite of algorithms that could benefit research in sport science. In principle, supervised ML approaches were designed for pure prediction, as opposed to explanation, leading to a rise in powerful, but opaque, algorithms. Recently, two subdomains of ML-explainable ML, which allows us to "peek into the black box," and interpretable ML, which encourages using algorithms that are inherently interpretable-have grown in popularity. The increased transparency of these powerful ML algorithms may provide considerable support for the hypothetico-deductive framework, in which hypotheses are generated from prior beliefs and theory, and are assessed against data collected specifically to test that hypothesis. However, this paper shows why ML algorithms are fundamentally different from statistical methods, even when using explainable or interpretable approaches. Translating potential insights from supervised ML algorithms, while in many cases seemingly straightforward, can have unanticipated challenges. While supervised ML cannot be used to replace statistical methods, we propose ways in which the sport sciences community can take advantage of supervised ML in the hypothetico-deductive framework. In this manuscript we argue that supervised machine learning can and should augment our exploratory investigations in sport science, but that leveraging potential insights from supervised ML algorithms should be undertaken with caution. We justify our position through a careful examination of supervised machine learning, and provide a useful analogy to help elucidate our findings. Three case studies are provided to demonstrate how supervised machine learning can be integrated into exploratory analysis. Supervised machine learning should be integrated into the scientific workflow with requisite caution. The approaches described in this paper provide ways to safely leverage the strengths of machine learning-like the flexibility ML algorithms can provide for fitting complex patterns-while avoiding potential pitfalls-at best, like wasted effort and money, and at worst, like misguided clinical recommendations-that may arise when trying to integrate findings from ML algorithms into domain knowledge. KEY POINTS: Some supervised machine learning algorithms and statistical models are used to solve the same problem, y = f(x) + ε, but differ fundamentally in motivation and approach. The hypothetico-deductive framework-in which hypotheses are generated from prior beliefs and theory, and are assessed against data collected specifically to test that hypothesis-is one of the core frameworks comprising the scientific method. In the hypothetico-deductive framework, supervised machine learning can be used in an exploratory capacity. However, it cannot replace the use of statistical methods, even as explainable and interpretable machine learning methods become increasingly popular. Improper use of supervised machine learning in the hypothetico-deductive framework is tantamount to p-value hacking in statistical methods.

Predictive models for secondary epilepsy in patients with acute ischemic stroke within one year.

Post-stroke epilepsy (PSE) is a critical complication that worsens both prognosis and quality of life in patients with ischemic stroke. An interpretable machine learning model was developed to predict PSE using medical records from four hospitals in Chongqing. Medical records, imaging reports, and laboratory test results from 21,459 ischemic stroke patients were collected and analyzed. Univariable and multivariable statistical analyses identified key predictive factors. The dataset was split into a 70% training set and a 30% testing set. To address the class imbalance, the Synthetic Minority Oversampling Technique combined with Edited Nearest Neighbors was employed. Nine widely used machine learning algorithms were evaluated using relevant prediction metrics, with SHAP (SHapley Additive exPlanations) used to interpret the model and assess the contributions of different features. Regression analyses revealed that complications such as hydrocephalus, cerebral hernia, and deep vein thrombosis, as well as specific brain regions (frontal, parietal, and temporal lobes), significantly contributed to PSE. Factors such as age, gender, NIH Stroke Scale (NIHSS) scores, and laboratory results like WBC count and D-dimer levels were associated with increased PSE risk. Tree-based methods like Random Forest, XGBoost, and LightGBM showed strong predictive performance, achieving an AUC of 0.99. The model accurately predicts PSE risk, with tree-based models demonstrating superior performance. NIHSS score, WBC count, and D-dimer were identified as the most crucial predictors. The research is funded by Central University basic research young teachers and students research ability promotion sub-projec t(2023CDJYGRH-ZD06), and by Emergency Medicine Chongqing Key Laboratory Talent Innovation and development joint fund project (2024RCCX10).

Predicting stroke severity of patients using interpretable machine learning algorithms.

Stroke is a significant global health concern, ranking as the second leading cause of death and placing a substantial financial burden on healthcare systems, particularly in low- and middle-income countries. Timely evaluation of stroke severity is crucial for predicting clinical outcomes, with standard assessment tools being the Rapid Arterial Occlusion Evaluation (RACE) and the National Institutes of Health Stroke Scale (NIHSS). This study aims to utilize Machine Learning (ML) algorithms to predict stroke severity using these two distinct scales. We conducted this study using two datasets collected from hospitals in Urmia, Iran, corresponding to stroke severity assessments based on RACE and NIHSS. Seven ML algorithms were applied, including K-Nearest Neighbor (KNN), Decision Tree (DT), Random Forest (RF), Adaptive Boosting (AdaBoost), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), and Artificial Neural Network (ANN). Hyperparameter tuning was performed using grid search to optimize model performance, and SHapley Additive Explanations (SHAP) were used to interpret the contribution of individual features. Among the models, the RF achieved the highest performance, with accuracies of 92.68% for the RACE dataset and 91.19% for the NIHSS dataset. The Area Under the Curve (AUC) was 92.02% and 97.86% for the RACE and NIHSS datasets, respectively. The SHAP analysis identified triglyceride levels, length of hospital stay, and age as critical predictors of stroke severity. This study is the first to apply ML models to the RACE and NIHSS scales for predicting stroke severity. The use of SHAP enhances the interpretability of the models, increasing clinicians' trust in these ML algorithms. The best-performing ML model can be a valuable tool for assisting medical professionals in predicting stroke severity in clinical settings.

Machine Learning in Genomics: Applications in Whole Genome Sequencing, Whole Exome Sequencing, Single-Cell Genomics, and Spatial Transcriptomics

The application of machine learning (ML) to genomics has transformed the process of analyzing and interpreting large-scale, complex datasets, leading to important breakthroughs in our knowledge of biological systems. This review provides a comprehensive overview of ML applications in key genomic areas: Whole Genome Sequencing (WGS), Whole Exome Sequencing (WES), single-cell genomics, and spatial transcriptomics. In WGS and WES, ML techniques are employed for variant calling, genome-wide association studies, rare variant analysis, and the prediction of pathogenicity. In single-cell genomics, ML facilitates clustering, trajectory inference, and cell type identification, while in spatial transcriptomics, it aids in deciphering spatial patterns of gene expression and tissue heterogeneity. This review further explores the application of ML in related omics fields, including proteomics, transcriptomics, metagenomics, epigenomics, and microbiome research. These applications encompass protein structure prediction, functional annotation, microbial community profiling, and the analysis of epigenetic modifications. We address the challenges caused by high dimensionality, variability in the data, and the requirement for interpretable machine learning models when dealing with genomic data. Emerging technologies like explainable AI and federated learning are highlighted for their potential to address these challenges. Additionally, the review addresses ethical considerations, data privacy issues, and the necessity for standardized protocols in ML applications. This comprehensive examination underscores the transformative impact of ML in genomics and highlights its potential to drive future innovations in personalized medicine and biological research.

Detecting interactions with random forests: a comment on Gries’ words of caution and suggestions for improvement

Abstract Tree-based methods are being both successfully applied and critically discussed in corpus linguistics. In this article, we would like to contribute a few aspects to this discussion from a methodological point of view. These aspects include the interpretation of interaction effects in single trees and random forests, as well as more general aspects like stability and overfitting. In particular, we have conducted a simulation study to investigate an approach suggested by Gries for computing the importance of interactions in random forests more systematically than the previous literature. The evidence of this simulation study shows that, even when interaction predictors are explicitly added, the permutation variable importance is not suited for distinguishing between main effects and interaction effects or between interaction effects of different orders. We also discuss the use of partial dependence (PD) and individual conditional expectation (ICE) plots for illustrating the functional form and potential interaction effects, and other means of interpretable machine learning.

Exploring the cytotoxic and antioxidant properties of lanthanide-doped ZnO nanoparticles: a study with machine learning interpretation

BackgroundLanthanide-based nanomaterials offer a promising alternative for cancer therapy because of their selectivity and effectiveness, which can be modified and predicted by leveraging the improved accuracy and enhanced decision-making of machine learning (ML) modeling.MethodsIn this study, erbium (Er3+) and ytterbium (Yb3+) were used to dope zinc oxide (ZnO) nanoparticles (NPs). Various characterization techniques and biological assays were employed to investigate the physicochemical and optical properties of the (Er, Yb)-doped ZnO NPs, revealing the influence of the lanthanide elements.ResultsThe (Er, Yb)-doped ZnO NPs exhibited laminar-type morphologies, negative surface charges, and optical bandgaps that vary with the presence of Er3+ and Yb3+. The incorporation of lanthanide ions reduced the cytotoxicity activity of ZnO against HEPG-2, CACO-2, and U87 cell lines. Conversely, doping with Er3+ and Yb3+ enhanced the antioxidant activity of the ZnO against DPPH, ABTS, and H2O2 radicals. The extra tree (ET) and random forest (RF) models predicted the relevance of the characterization results vis-à-vis the cytotoxic properties of the synthesized NPs.ConclusionThis study demonstrates, for the first time, the synthesis of ZnO NPs doped with Er and Yb via a solution polymerization route. According to characterization results, it was unveiled that the effect of optical bandgap variations influenced the cytotoxic performance of the developed lanthanide-doped ZnO NPs, being the undoped ZnO NPs the most cytotoxic ones. The presence alone or in combination of Er and Yb enhanced their scavenging capacity. ML models such as ET and RF efficiently demonstrated that the concentration and cell line type are key parameters that influence the cytotoxicity of (Er, Yb)-doped ZnO NPs achieving high accuracy rates of 98.96% and 98.67%, respectively. This study expands the knowledge of lanthanides as dopants of nanomaterials for biological and medical applications and supports their potential in cancer therapy by integrating robust ML approaches.Graphical abstract

Development and external validation of an interpretable machine learning model for the prediction of intubation in the intensive care unit

Given the limited capacity to accurately determine the necessity for intubation in intensive care unit settings, this study aimed to develop and externally validate an interpretable machine learning model capable of predicting the need for intubation among ICU patients. Seven widely used machine learning (ML) algorithms were employed to construct the prediction models. Adult patients from the Medical Information Mart for Intensive Care IV database who stayed in the ICU for longer than 24 h were included in the development and internal validation. The model was subsequently externally validated using the eICU-CRD database. In addition, the SHapley Additive exPlanations method was employed to interpret the influence of individual parameters on the predictions made by the model. A total of 11,988 patients were included in the final cohort for this study. The CatBoost model demonstrated the best performance (AUC: 0.881). In the external validation set, the efficacy of our model was also confirmed (AUC: 0.750), which suggests robust generalization capabilities. The Glasgow Coma Scale (GCS), body mass index (BMI), arterial partial pressure of oxygen (PaO2), respiratory rate (RR) and length of stay (LOS) before ICU were the top 5 features of the CatBoost model with the greatest impact. We developed an externally validated CatBoost model that accurately predicts the need for intubation in ICU patients within 24 to 96 h of admission, facilitating clinical decision-making and has the potential to improve patient outcomes. The prediction model utilizes readily obtainable monitoring parameters and integrates the SHAP method to enhance interpretability, providing clinicians with clear insights into the factors influencing predictions.

Advancing source apportionment of soil potentially toxic elements using a hybrid model: a case study in urban parks, Beijing, China.

Identifying the source-specific health risks of potentially toxic elements (PTE) in urban park soils is essential for human health protection. However, previous studies have mostly focused on the deterministic source-specific health risks, ignoring the health risk assessment from a probabilistic perspective. To fill this gap, we developed a hybrid model that incorporated machine learning (ML) interpretability into positive matrix factorization (PMF) and probability health risk assessment (PHRA) based on the Monte Carlo simulation. The results indicated that concentrations of soil PTEs except for Mn and Sb were significantly higher than their corresponding background values. Random forest (RF) was regarded as the best ML model to identify key drivers for As, Cd, Cr, Cu, Ni, Pb, and Zn, with R2 > 0.60, but was less effective for other soil PTEs (R2 < 0.49). Specifically, the contributions of the four potential pollutionsources were mixed sources, traffic emission, fuel combustion, and building materials, with contribution rate of 24.88%, 30.56%, 28.99%, and 15.56%, respectively. Fuel combustion contributed the most to non-carcinogenic for children (39.45%), male (43.84%), and female (43.76%), and the non-carcinogenic risk could be considered negligiblefor human. However, building materials was the major contributor to carcinogenic risk for children (36.1%), male (44.9%), and female (43.2%). The integration of the RF model with PMF and PHRA improved the accuracy of the results by identifying and quantifying the specific sources of each soil PTE using the relative importance analysis from the RF model. The results of this study assisted in providing efficient strategies for risk management and control of soil PTEs in Beijing parks.

Predictions of the Optical Properties of Brown Carbon Aerosol by Machine Learning with Typical Chromophores.

The linkages between BrC optical properties and chemical composition remain inadequately understood, with quantified chromophores explaining less than 25% of ambient aerosol light absorption. This study characterized 38 typical chromophores in aerosols collected in Xi'an, with light absorption contributions to BrC ranging from 1.6 ± 0.3 to 5.8 ± 2.6% at 365 nm. Based on these quantified chromophores, an interpretable machine learning model and the Shapley Additive Explanation (SHAP) method were employed to explore the relationships between BrC optical properties and chemical composition. The model attained high accuracy with Pearson correlation coefficients (r) exceeding 0.93 for the absorption coefficient (Absλ) and surpassing 0.57 for mass absorption efficiency (MAEλ) of BrC. It explains more than 80% of the variance in Abs and over 50% in MAE, significantly improving the understanding of BrC light absorption. Polycyclic aromatic hydrocarbons (PAHs) and oxygenated PAHs (OPAHs) with four and five rings exhibit significant positive effects on Absλ, suggesting that similar unidentified chromophores may also notably impact BrC optical characteristics. The model based on chromophore mass concentrations further simplifies studying BrC optical characteristics. This study advances understanding of the relationship between BrC composition and optical properties and guides the investigation of unrecognized chromophores.

Prediction of holey graphyne-supported single atom catalyst for nitrogen reduction reaction by interpretable machine learning and first-principles calculations

The electrocatalytic nitrogen reduction reaction (NRR) with efficient and selective electrocatalysts has emerged as a promising alternative for sustainable ammonia synthesis. In this study, we designed a novel class of single-atom catalysts (SACs), denoted as TM@HGY, by embedding 3d, 4d, and 5d transition metals into synthesized holey graphyne (HGY), and investigated their potential for NRR using a combined approach of density functional theory (DFT) calculations and machine learning. Through stability assessments and a three-step screening strategy, Sc@HGY, V@HGY, and Mo@HGY were identified as promising NRR electrocatalysts. Notably, V@HGY exhibited an exceptionally low limiting potential of −0.16 V, which is superior to all the known NRR SACs supported by graphyne-family members. Machine learning (ML) analysis revealed that the Mendeleev number (Nm), group (G), and d-orbital radius (Rd) of the absorbed metal atom are the primary contributors to the structural stability and catalytic activity of these SACs, and clear strategies for optimizing catalyst design were further suggested based on their intrinsic relationships. This work reveals the significant potential of TM@HGY in NRR, providing powerful guidance for designing high-performance SACs in the field of sustainable ammonia synthesis.

Genome-wide association studies of ischemic stroke based on interpretable machine learning

Despite the identification of several dozen genetic loci associated with ischemic stroke (IS), the genetic bases of this disease remain largely unexplored. In this research we present the results of genome-wide association studies (GWAS) based on classical statistical testing and machine learning algorithms (logistic regression, gradient boosting on decision trees, and tabular deep learning model TabNet). To build a consensus on the results obtained by different techniques, the Pareto-Optimal solution was proposed and applied. These methods were applied to real genotypic data of sick and healthy individuals of European ancestry obtained from the Database of Genotypes and Phenotypes (5,581 individuals, 883,749 single nucleotide polymorphisms). Finally, 131 genes were identified as candidates for association with the onset of IS. UBQLN1, TRPS1, and MUSK were previously described as associated with the course of IS in model animals. ACOT11 taking part in metabolism of fatty acids was shown for the first time to be associated with IS. The identified genes were compared with genes from the Illuminating Druggable Genome project. The product of GPR26 representing the G-coupled protein receptor can be considered as a therapeutic target for stroke prevention. The approaches presented in this research can be used to reprocess GWAS datasets from other diseases.

Predicting temporomandibular disorders in adults using interpretable machine learning methods: a model development and validation study

IntroductionTemporomandibular disorders (TMD) have a high prevalence and complex etiology. The purpose of this study was to apply a machine learning (ML) approach to identify risk factors for the occurrence of TMD in adults and to develop and validate an interpretable predictive model for the risk of TMD in adults.MethodsA total of 949 adults who underwent oral examinations were enrolled in our study. 5 different ML algorithms were used for model development and comparison, and feature selection was performed by feature importance ranking and feature decreasing methods. Several evaluation indexes, including the area under the receiver-operating-characteristic curve (AUC), were used to compare the predictive performance. The precision-recall curve (PR), calibration curve, and decision curve analysis (DCA) further assessed the accuracy and clinical utility of the model.ResultsThe performance of the random forest (RF) model was the best among the 5 ML models. An interpretable RF model was developed with 7 features (gender, malocclusion, unilateral chewing, chewing hard substances, grinding teeth, clenching teeth, and anxiety). The AUCs of the final model on the training set, internal validation set, and external test set were 0.892, 0.854, and 0.857, respectively. Calibration and DCA curves showed high accuracy and clinical applicability of the model.DiscussionAn efficient and interpretable TMD risk prediction model for adults was successfully developed using the ML method. The model not only has good predictive performance, but also enhances the clinical application value of the model through the SHAP method. This model can provide clinicians with a practical and efficient TMD risk assessment tool that can help them better predict and assess TMD risk in adults, supporting more efficient disease management and targeted medical interventions.

Interpretable causal machine learning optimization tool for improving efficiency of internal carbon source-biological denitrification

Interpretable causal machine learning (ICML) was used to predict the performance of denitrification and clarify the relationships between influencing factors and denitrification. Multiple models were examined, and XG-Boost model provided the best prediction (R2 = 0.8743). Based on the ICML framework, hydraulic retention time (HRT), mixture chemical oxygen demand/total nitrogen (COD/TN = C/N), mixture COD concentration, and pretreatment technology were identified as important features affecting the denitrification performance. Further, tapping point and partial dependence analyses provided the range of key factors that precisely regulate denitrification. In the application analysis, HRT (6–10.5 h), mixture C/N (6–12), and mixture COD concentration (300–600 mg L−1) were the appropriate operating ranges, achieving TN removal of approximately 73 %–77 %. The effluent TN and COD concentrations met the discharge standards for wastewater in China (class 1A) and EU. These findings provide support for regulating excess sludge as internal carbon source to promote denitrification.

Urban Form and Seasonal PM2.5 Dynamics: Enhancing Air Quality Prediction using Interpretable Machine Learning and IoT Sensor Data

Urban Form and Seasonal PM2.5 Dynamics: Enhancing Air Quality Prediction using Interpretable Machine Learning and IoT Sensor Data

A Comprehensive Evaluation of Machine Learning Algorithms for Digital Soil Organic Carbon Mapping on a National Scale

The aim of this study was to narrow the research gap of ambiguity in which machine learning algorithms should be selected for evaluation in digital soil organic carbon (SOC) mapping. This was performed by providing a comprehensive assessment of prediction accuracy for 15 frequently used machine learning algorithms in digital SOC mapping based on studies indexed in the Web of Science Core Collection (WoSCC), providing a basis for algorithm selection in future studies. Two study areas, including mainland France and the Czech Republic, were used in the study based on 2514 and 400 soil samples from the LUCAS 2018 dataset. Random Forest was first ranked for France (mainland) and then ranked for the Czech Republic regarding prediction accuracy; the coefficients of determination were 0.411 and 0.249, respectively, which was in accordance with its dominant appearance in previous studies indexed in the WoSCC. Additionally, the K-Nearest Neighbors and Gradient Boosting Machine regression algorithms indicated, relative to their frequency in studies indexed in the WoSCC, that they are underrated and should be more frequently considered in future digital SOC studies. Future studies should consider study areas not strictly related to human-made administrative borders, as well as more interpretable machine learning and ensemble machine learning approaches.

Predicting the Residual Compressive Strength of Concrete Exposed to Elevated Temperatures Using Interpretable Machine Learning

Predicting the Residual Compressive Strength of Concrete Exposed to Elevated Temperatures Using Interpretable Machine Learning

Aerodynamic robustness optimization of aeroengine fan performance based on an interpretable dynamic machine learning method

Aerodynamic robustness optimization of aeroengine fan performance based on an interpretable dynamic machine learning method