Random Forest Machine Learning Model Research Articles

Predicting CRISPR-Cas12a guide efficiency for targeting using machine learning.

Genome editing through the development of CRISPR (Clustered Regularly Interspaced Short Palindromic Repeat)-Cas technology has revolutionized many fields in biology. Beyond Cas9 nucleases, Cas12a (formerly Cpf1) has emerged as a promising alternative to Cas9 for editing AT-rich genomes. Despite the promises, guide RNA efficiency prediction through computational tools search still lacks accuracy. Through a computational meta-analysis, here we report that Cas12a target and off-target cleavage behavior are a factor of nucleotide bias combined with nucleotide mismatches relative to the protospacer adjacent motif (PAM) site. These features helped to train a Random Forest machine learning model to improve the accuracy by at least 15% over existing algorithms to predict guide RNA efficiency for the Cas12a enzyme. Despite the progresses, our report underscores the need for more representative datasets and further benchmarking to reliably and accurately predict guide RNA efficiency and off-target effects for Cas12a enzymes.

Energy efficient fault detection and classification using hyperparameter-tuned machine learning classifiers with sensors

This paper presents a novel approach in which hyperparameter-tuned Machine Learning (ML) classifiers with Optuna is used for fault detection and classification over a power transmission line. In this paper, popular ML models Random Forest (RF), Decision Tree (DT), XGBoost (XGB), and Light Gradient Boosting Machine (LGBM) are used for fault detection and classification. This study uses a two-layer approach for fault detection and classification. An electrical utility and data simulation provided the PMU measurement and recorded data from a simulated grid. The faults had varied impedances and included various fault classes at distinct line locations. The optimal feature from 3-phase current and voltage signals is extracted using Pearson correlation, recursive feature elimination, and univariate feature (t-test) methods. The synthetic minority class oversampling technique (SMOTE) was used to address the issue of imbalanced data. Hyperparameters of the evaluated LGBM classifier is trained with the Optuna. The performance of the proposed classifier is measured in terms of the accuracy, precision, Recall and F1-score metrics. The proposed method outperformed the conventional ML methods.

Verifying origin claims on dairy products using stable isotope ratio analysis and random forest classification.

Scientifically underpinning geographic origin claims will improve consumer trust in food labels. Stable isotope ratio analysis (SIRA) is an analytical technique that supports origin verification of food products based on naturally occurring differences in isotopic compositions. SIRA of five relevant elements (C, H, N, O, S) was conducted on casein isolated from butter (n=60), cheese (n=96), and whole milk powder (WMP) (n=41). Samples were divided into four geographic regions based on their commercial origin: Ireland (n=79), Europe (n=67), Australasia (n=29) and USA (n=22). A random forest machine learning model built using δ13C, δ2H, δ15N, δ18O and δ34S values of all products (n=197) accurately (88% model accuracy rate) predicted the region of origin with class accuracy of 95% for Irish, 84% for European, 71% for Australasia, and 94% for US products.

The advantage of intergenic regions as genomic features for machine-learning-based host attribution of Salmonella Typhimurium from the USA.

Salmonella enterica is a taxonomically diverse pathogen with over 2600 serovars associated with a wide variety of animal hosts including humans, other mammals, birds and reptiles. Some serovars are host-specific or host-restricted and cause disease in distinct host species, while others, such as serovar S. Typhimurium (STm), are generalists and have the potential to colonize a wide variety of species. However, even within generalist serovars such as STm it is becoming clear that pathovariants exist that differ in tropism and virulence. Identifying the genetic factors underlying host specificity is complex, but the availability of thousands of genome sequences and advances in machine learning have made it possible to build specific host prediction models to aid outbreak control and predict the human pathogenic potential of isolates from animals and other reservoirs. We have advanced this area by building host-association prediction models trained on a wide range of genomic features and compared them with predictions based on nearest-neighbour phylogeny. SNPs, protein variants (PVs), antimicrobial resistance (AMR) profiles and intergenic regions (IGRs) were extracted from 3883 high-quality STm assemblies collected from humans, swine, bovine and poultry in the USA, and used to construct Random Forest (RF) machine learning models. An additional 244 recent STm assemblies from farm animals were used as a test set for further validation. The models based on PVs and IGRs had the best performance in terms of predicting the host of origin of isolates and outperformed nearest-neighbour phylogenetic host prediction as well as models based on SNPs or AMR data. However, the models did not yield reliable predictions when tested with isolates that were phylogenetically distinct from the training set. The IGR and PV models were often able to differentiate human isolates in clusters where the majority of isolates were from a single animal source. Notably, IGRs were the feature with the best performance across multiple models which may be due to IGRs acting as both a representation of their flanking genes, equivalent to PVs, while also capturing genomic regulatory variation, such as altered promoter regions. The IGR and PV models predict that ~45 % of the human infections with STm in the USA originate from bovine, ~40 % from poultry and ~14.5 % from swine, although sequences of isolates from other sources were not used for training. In summary, the research demonstrates a significant gain in accuracy for models with IGRs and PVs as features compared to SNP-based and core genome phylogeny predictions when applied within the existing population structure. This article contains data hosted by Microreact.

B-036 Multi-Biomarker Approach to Latent Tuberculosis Diagnosis Using Microring Resonator Sensors

Abstract Background Tuberculosis (TB) is an infectious disease that induces a complex response from the host immune system. Most carriers of the bacteria experience the asymptomatic, non-infectious phase, latent tuberculosis infection (LTBI), but have the potential to reactivate to active, contagious TB infection. Current diagnostics for TB are not sufficient for LTBI and are focused on using one biomarker, interferon-gamma (IFN-g). To build better diagnostics for the complex disease, multi-biomarker approaches to profile the immune system, coupled with machine learning algorithms, are necessary to diagnose LTBI and determine patients' risks of reactivation to active TB. We aim to use a multiplexed silicon photonic microring resonator sensing platform to profile thirteen biomarkers in patients with LTBI infection and use machine learning data analysis to identify biomarkers relevant for diagnosing LTBI and predicting reactivation risk. Methods We employed microring resonators coupled with a sandwich-style detection format to simultaneously detect thirteen cytokines and chemokines in plasma in under 40 minutes. Chip-integrated silicon photonic microrings are small in size and easily multiplexed with up to sixteen different capture agents. Biomolecular binding events are measured by monitoring the resonant wavelength within the sensing cavity, which shifts as binding events occur. Quantitation of each biomarker is possible with calibration curves that correlate standard protein concentrations to wavelength shifts. The multiplexed assay has been optimized for each individual biomarker, tested for cross-reactivity, and calibrated in the plasma matrix of interest. The LODs vary, depending on biomarker and matrix dilution, but range from 4.9 pg/mL to 691 pg/mL. The patient plasma samples are left over from the TB screening IFN-g release assay, QuantiFERON-TB Gold Plus (QFT), putting this method within the current clinical workflow. We use precision normalization approaches from the four QFT stimulations to account for immunologic differences among the subjects. The biomarker concentrations are evaluated using a random forest machine learning model to identify which biomarkers are important for identifying patients with LTBI and predicting their reactivation risk. Results We used this sensor method to profile 42 patients, 24 LTBI negative controls and 18 LTBI positive controls, with 13 being considered at high risk of reactivation. Using a precision normalization approach, a combination of nine normalized conditions using five of the thirteen biomarkers (CCL4, CCL8, IP-10, IL-2, IL-17) discriminated between LTBI positive and negative subjects with a ROC AUC of 0.90. Additionally, eight normalized conditions using four of the same biomarkers discriminated between high and low-risk subjects with an AUC of 0.83. Currently, we have expanded the sample cohort to include an additional 72 subjects, 25 LTBI positive and 47 LTBI negative. Preliminary analysis shows CCL8, IL-2, and IP-10 alone are able to distinguish between LTBI status (P values <0.01) using raw target concentrations prior to precision normalization. Conclusion Overall, we show the importance of multi-biomarker approaches to develop diagnostic and prognostic tools for a complex disease, such as TB. Further work to use the initial machine learning algorithm with the expanded cohort to validate biomarker importance is underway.

Estimation of historical daily airborne pollen concentrations across Switzerland using a spatio temporal random forest model

High concentrations of airborne pollen trigger seasonal allergies and possibly more severe adverse respiratory and cardiovascular health events. Predicting pollen concentration accurately is valuable for epidemiological studies, in order to study the effects of pollen exposure. We aimed to develop a spatiotemporal machine learning model predicting daily pollen concentrations at a spatial resolution of 1 × 1 km across Switzerland between 2000 and 2019. Daily pollen concentrations for five common, highly allergenic pollen types (hazel, alder, birch ash, and grasses) were available from fourteen measurement sites across Switzerland. We considered several predictors such as elevation, species distribution, wind speed, wind direction, temperature, precipitation, relative humidity, satellite-observed Normalized Difference Vegetation Index, and land-use (CORINE, Landsat satellite) to explain variation in pollen concentration. We employed feature engineering techniques to encode categorical variables and fill in missing values. We applied a random forest machine learning model with 5-fold cross-validation. The 5th–99th percentiles for concentrations of hazel, alder, birch, ash, and grass pollen at the pollen monitoring stations were 0–298, 0–306, 0–1153, 0–800, and 0–290 pollen grains/m3, respectively. The results of a predictive model for these concentrations yielded overall R2 values of 0.87, 0.84, 0.89, 0.88, and 0.91, and temporal root mean squared errors (RMSEs) of 16.07, 16.72, 69.04, 41.50, and 22.45 pollen grains/m3. An analysis of predictor variable importance indicates that the average national daily pollen concentration is the most important predictor of pollen concentrations for all pollen types. Furthermore, meteorological variables including temperature, total precipitation, humidity, boundary layer height, wind speed, and wind direction, as well as date and satellite features, are important factors in pollen concentration prediction. These spatiotemporal pollen models will serve to estimate individual residential pollen exposure for epidemiological studies. Resulting estimates will enable us to study respiratory and cardiovascular mortality and hospital admissions in Switzerland.

Inundation–Desiccation State Prediction for Salt Pans in the Western Pannonian Basin Using Remote Sensing, Groundwater, and Meteorological Data

Salt pans are unique wetland ecosystems. In the Austrian Seewinkel region, salt pans are in an increasingly vulnerable state due to groundwater drainage and heightened climatic pressures. It is crucial to model how seasonal and long-term hydrological and climatological variations affect the salt pan dynamics in Seewinkel, yet a comprehensive understanding of the driving processes is lacking. The goal of this study is to develop random forest machine learning models driven by hydrological and meteorological data that allow us to predict in early spring (March) of each year the inundation state in the subsequent summer and fall. We utilize Earth observation data from Landsat 5 (L5), 8 (L8), and 9 (L9) to derive the time series of the inundation state for 34 salt pans for the period 1984–2022. Furthermore, we demonstrate that the groundwater level observed in March is the strongest predictor of the salt pan inundation state in summer and fall. Utilizing local groundwater data yields a Matthews correlation coefficient of 0.59. Models using globally available meteorological data, either instead of or in addition to groundwater data, provide comparable results. This allows the global transfer of the approach to comparable ecosystems where no in situ data are available.

MIM-ML: A Novel Quantum Chemical Fragment-Based Random Forest Model for Accurate Prediction of NMR Chemical Shifts of Nucleic Acids.

We developed a random forest machine learning (ML) model for the prediction of 1H and 13C NMR chemical shifts of nucleic acids. Our ML model is trained entirely on reproducing computed chemical shifts obtained previously on 10 nucleic acids using a Molecules-in-Molecules (MIM) fragment-based density functional theory (DFT) protocol including microsolvation effects. Our ML model includes structural descriptors as well as electronic descriptors from an inexpensive low-level semiempirical calculation (GFN2-xTB) and trained on a relatively small number of DFT chemical shifts (2080 1H chemical shifts and 1780 13C chemical shifts on the 10 nucleic acids). The ML model is then used to make chemical shift predictions on 8 new nucleic acids ranging in size from 600 to 900 atoms and compared directly to experimental data. Though no experimental data was used in the training, the performance of our model is excellent (mean absolute deviation of 0.34 ppm for 1H chemical shifts and 2.52 ppm for 13C chemical shifts for the test set), despite having some nonstandard structures. A simple analysis suggests that both structural and electronic descriptors are critical for achieving reliable predictions. This is the first attempt to combine ML from fragment-based DFT calculations to predict experimental chemical shifts accurately, making the MIM-ML model a valuable tool for NMR predictions of nucleic acids.

Seasonal drivers of dissolved oxygen across a tidal creek–marsh interface revealed by machine learning

AbstractDissolved oxygen (DO) is a key biogeochemical control in coastal systems, and its concentration and drivers vary markedly through time and space. This makes it difficult to accurately represent coastal DO and associated biogeochemical processes in models, limiting our ability to predict how these systems will respond to global change. We obtained high‐frequency (5‐min) in situ measurements of DO collected at three locations across the interface of a tidal creek and coastal marsh in the Pacific Northwest, USA. Random Forest machine learning models quantified the importance of three categories of environmental drivers (Aquatic, Climatic, and Terrestrial) of DO variability across the creek–marsh interface. We selected two 4‐month datasets representing Summer and Winter seasonal periods to test two hypotheses on the dominant drivers of DO at the coastal interface. We found that the Terrestrial driver—characterized by long periods of anaerobic conditions and episodic pulses in DO after floods—was most important during the Winter, whereas the Aquatic driver—characterized by variability over tidal, diel, and lunar cycles—was most important during the Summer. We explored how future climate change scenarios could alter the drivers of DO variability using a cumulative sums driver–response framework. Our results suggest that under climate change, Aquatic and Climatic drivers may increase in importance during the Summer, potentially linked to changing metabolic regimes and sea level, with Terrestrial driver importance potentially increasing during the Winter. Our approach highlights useful methods for understanding the spatiotemporal complexity of oxygen across coastal interfaces and quantifying the relative importance of distinct environmental drivers.

Five machine learning-based radiomics models for preoperative prediction of histological grade in hepatocellular carcinoma.

To compare the efficacy of radiomics models via five machine learning algorithms in predicting the histological grade of hepatocellular carcinoma (HCC) before surgery and to develop the most stable model to classify high-risk HCC patients. Contrast-enhanced computed tomography (CECT) images of 175 HCC patients before surgery were analysed, and radiomics features were extracted from CECT images (including arterial and portal phases). Five machine learning models, including Bayes, random forest (RF), k-nearest neighbors (KNN), logistic regression (LR), and support vector machine (SVM), were applied to establish the model. The stability of the five models was weighed by the relative standard deviation (RSD), and the lowest RSD value was chosen as the most stable model to predict the histological grade of HCC. The area under the curve (AUC) and Delong tests were devoted to assessing the predictive efficacy of the models. High-grade HCC accounted for 28.57% (50/175) of the 175 patients. The RSD value of AUC via the RF machine learning model was the lowest (2.3%), followed by Bayes (3.2%), KNN (6.4%), SVM (8.7%) and LR (31.3%). In addition, the RF model (AUC = 0.995) was better than the other four models in the training set (p < 0.05), as well as obtained good predictive performance in the test set (AUC = 0.837). Among the five machine learning models, the RF-based radiomics model was the most stable and performed excellently in identifying high histological grade of HCC.

A Personalized Risk Model for Azacitidine Outcome in Myelodysplastic Syndrome and Other Myeloid Neoplasms Identified by Machine Learning Model Utilizing Real-World Data.

Azacitidine is an approved therapy for higher-risk myelodysplastic syndrome (MDS). However, only 30-40% patients respond to azacitidine, and the responses may take up to six cycles to become evident. Delayed responses and the myelosuppressive effects of azacitidine make it challenging to predict which patients will benefit. This is further compounded by a lack of uniform prognostic tools to identify patients at risk of early treatment failure. Hence, we performed a retrospective analysis of 273 consecutive azacytidine-treated patients. The median overall survival was 16.25 months with only 9% alive at 5 years. By using pre-treatment variables incorporated into a random forest machine learning model, we successfully identified those patients unlikely to benefit from azacytidine upfront (7.99 vs. 22.8 months, p < 0.0001). This model also identified those who required significantly more hospitalizations and transfusion support. Notably, it accurately predicted survival outcomes, outperforming the existing prognostic scoring system. By integrating somatic mutations, we further refined the model and identified three distinct risk groups with significant differences in survival (5.6 vs. 10.5 vs. 43.5 months, p < 0.0001). These real-world findings emphasize the urgent need for personalized prediction tools tailored to hypomethylating agents, reducing unnecessary complications and resource utilization in MDS treatment.

The intestinal microbiota predicts COVID-19 severity and fatality regardless of hospital feeding method.

SARS-CoV-2-positive patients exhibit gut and oral microbiome dysbiosis, which is associated with various aspects of COVID-19 disease (1-4). Here, we aim to identify gut and oral microbiome markers that predict COVID-19 severity in hospitalized patients, specifically severely ill patients compared to moderately ill ones. Moreover, we investigate whether hospital feeding (solid versus enteral), an important cofounder, influences the microbial composition of hospitalized COVID-19 patients. We used random forest classification machine learning models with interpretable secondary analyses. The gut, but not the oral microbiota, was a robust predictor of both COVID-19-related fatality and severity of hospitalized patients, with a higher predictive value than most clinical variables. In addition, perturbations of the gut microbiota due to enteral feeding did not associate with species that were predictive of COVID-19 severity. IMPORTANCE SARS-CoV-2 infection leads to wide-ranging, systemic symptoms with sometimes unpredictable morbidity and mortality. It is increasingly clear that the human microbiome plays an important role in how individuals respond to viral infections. Our study adds to important literature about the associations of gut microbiota and severe COVID-19 illness during the early phase of the pandemic before the availability of vaccines. Increased understanding of the interplay between microbiota and SARS-CoV-2 may lead to innovations in diagnostics, therapies, and clinical predictions.

Key predictors for climate policy support and political mobilization: The role of beliefs and preferences

Public support and political mobilization are two crucial factors for the adoption of ambitious climate policies in line with the international greenhouse gas reduction targets of the Paris Agreement. Despite their compound importance, they are mainly studied separately. Using a random forest machine-learning model, this article investigates the relative predictive power of key established explanations for public support and mobilization for climate policies. Predictive models may shape future research priorities and contribute to theoretical advancement by showing which predictors are the most and least important. The analysis is based on a pre-election conjoint survey experiment on the Swiss CO2 Act in 2021. Results indicate that beliefs (such as the perceived effectiveness of policies) and policy design preferences (such as for subsidies or tax-related policies) are the most important predictors while other established explanations, such as socio-demographics, issue salience (the relative importance of issues) or political variables (such as the party affiliation) have relatively weak predictive power. Thus, beliefs are an essential factor to consider in addition to explanations that emphasize issue salience and preferences driven by voters’ cost-benefit considerations.

Integrating machine learning and environmental variables to constrain uncertainty in crop yield change projections under climate change

Robust crop yield projections under future climates are fundamental prerequisites for reliable policy formation. Both process-based crop models and statistical models are commonly used for this purpose. Process-based models tend to simplify processes, minimize the effects of extreme events, and ignore biotic pressures, while statistical models cannot deterministically capture intricate biological and physiological processes underpinning crop growth. We attempted to integrate and overcome shortcomings in both modelling frameworks by integrating the dynamic linear model (DLM) and random forest machine learning model (RF) with nine global gridded crop models (GGCM), respectively, in order to improve projections and reduce uncertainties of maize (Zea mays L.) and soybean (Glycine max [L.] Merrill) yield projections. Our results demonstrated substantial improvements in model performance accuracy by using RF in concert with GGCM across China’s maize and soybean belt. This improvement surpasses that achieved using DLM. For maize, the GGCM+RF models increased the r values from 0.15 to 0.61–0.64–0.77 and decreased nRMSE from approximately 0.20 to 0.50–0.13–0.17 compared with using GGCM alone. For soybean, the models increased r from 0.37 to 0.70–0.54–0.70 and decreased nRMSE from 0.17 to 0.35–0.17–0.20 compared with using GGCM alone. The main factors influencing maize yield changes included chilling days (CD), crop pests and diseases (CPDs), and drought, while for soybean the primary influencing factors included CPD, tropical days (based on exceeding a maximum temperature), and drought. Our approach decreased uncertainties by 33–78% for maize and by 56–68% for soybean. The main source of uncertainty for GGCM was the crop model. For GGCM+RF, the main source of uncertainty for the 2040–2069 period was the global climate model, while the main source of uncertainty for the 2070–2099 period was the climate scenario. Our results provide a novel, robust, and pragmatic framework to constrain uncertainties in order to accurately assess the impact of future climate change on crop yields. These results could be used to interpret future ensemble studies by accounting for uncertainty in crop and climate models, as well as to assess future emissions scenarios.

Predicting transcriptional responses to heat and drought stress from genomic features using a machine learning approach in rice.

Plants have evolved various mechanisms to adapt to adverse environmental stresses, such as the modulation of gene expression. Expression of stress-responsive genes is controlled by specific regulators, including transcription factors (TFs), that bind to sequence-specific binding sites, representing key components of cis-regulatory elements and regulatory networks. Our understanding of the underlying regulatory code remains, however, incomplete. Recent studies have shown that, by training machine learning (ML) algorithms on genomic sequence features, it is possible to predict which genes will transcriptionally respond to a specific stress. By identifying the most important features for gene expression prediction, these trained ML models allow, in theory, to further elucidate the regulatory code underlying the transcriptional response to abiotic stress. Here, we trained random forest ML models to predict gene expression in rice (Oryza sativa) in response to heat or drought stress. Apart from thoroughly assessing model performance and robustness across various input training data, the importance of promoter and gene body sequence features to train ML models was evaluated. The use of enriched promoter oligomers, complementing known TF binding sites, allowed us to gain novel insights in DNA motifs contributing to the stress regulatory code. By comparing genomic feature importance scores for drought and heat stress over time, general and stress-specific genomic features contributing to the performance of the learned models and their temporal variation were identified. This study provides a solid foundation to build and interpret ML models accurately predicting transcriptional responses and enables novel insights in biological sequence features that are important for abiotic stress responses.

Classification of structural building damage grades from multi-temporal photogrammetric point clouds using a machine learning model trained on virtual laser scanning data

Automatic damage assessment by analysing UAV-derived 3D point clouds provides fast information on the damage situation after an earthquake. However, the assessment of different damage grades is challenging given the variety in damage characteristics and limited transferability of methods to other geographic regions or data sources. We present a novel change-based approach to automatically assess multi-class building damage from real-world point clouds using a machine learning model trained on virtual laser scanning (VLS) data. Therein, we (1) identify object-specific point cloud-based change features, (2) extract changed building parts using k-means clustering, (3) train a random forest machine learning model with VLS data based on object-specific change features, and (4) use the classifier to assess building damage in real-world photogrammetric point clouds. We evaluate the classifier with respect to its capacity to classify three damage grades (heavy, extreme, destruction) in pre-event and post-event point clouds of an earthquake in L’Aquila (Italy). Using object-specific change features derived from bi-temporal point clouds, our approach is transferable with respect to multi-source input point clouds used for model training (VLS) and application (real-world photogrammetry). We further achieve geographic transferability by using simulated training data which characterises damage grades across different geographic regions. The model yields high multi-target classification accuracies (overall accuracy: 92.0%–95.1%). Classification performance improves only slightly when using real-world region-specific training data (< 3% higher overall accuracies). We consider our approach especially relevant for applications where timely information on the damage situation is required and sufficient real-world training data is not available.

Forecasting the progression of human civilization on the Kardashev Scale through 2060 with a machine learning approach

Energy has been propelling the development of human civilization for millennia. Humanity presently stands at Type 0.7276 on the Kardashev Scale, which was proposed to quantify the relationship between energy consumption and the development of civilizations. However, current predictions of human civilization remain underdeveloped and energy consumption models are oversimplified. In order to improve the precision of the prediction, we use machine learning models random forest and autoregressive integrated moving average to simulate and predict energy consumption on a global scale and the position of humanity on the Kardashev Scale through 2060. The result suggests that global energy consumption is expected to reach ~ 887 EJ in 2060, and humanity will become a Type 0.7449 civilization. Additionally, the potential energy segmentation changes before 2060 and the influence of the advent of nuclear fusion are discussed. We conclude that if energy strategies and technologies remain in the present course, it may take human civilization millennia to become a Type 1 civilization. The machine learning tool we develop significantly improves the previous projection of the Kardashev Scale, which is critical in the context of civilization development.

Hydrogeochemical and sediment parameters improve predication accuracy of arsenic-prone groundwater in random forest machine-learning models

The relative importance of groundwater geochemicals and sediment characteristics in predicting groundwater arsenic distributions was rarely documented. To figure this out, we established a random forest machine-learning model to predict groundwater arsenic distributions in the Hetao Basin, China, by using 22 variables of climate, topographic features, soil properties, sediment characteristics, groundwater geochemicals, and hydraulic gradients of 492 groundwater samples. The established model precisely captured the patchy distributions of groundwater arsenic concentrations in the basin with an AUC value of 0.84. Results suggest that Fe(II) was the most prominent variable in predicting groundwater arsenic concentrations, which supported that the enrichment of arsenic in groundwater was caused by the reductive dissolution of Fe(III) oxides. The high relative importance of SO42− indicated that sulfate reduction was also conducive to groundwater arsenic enrichment in inland basins. Nevertheless, parameters of climate variables, sediment characteristics, and soil properties showed secondly important roles in predicting groundwater arsenic concentrations. The other two models, which excluded parameters of groundwater geochemicals and/or sediment characteristics, showed much worse predictions than the model considering all variables. This highlights the importance of variables of groundwater geochemicals and sediment characteristics in improving the precision and accuracy of predicting results. Future studies should probe a method constructing the random forest predicting model with high precision based on the limited number of groundwater samples and sediment samples.

Assessing treatment switch among patients with multiple sclerosis: A machine learning approach

BackgroundPatients with multiple sclerosis (MS) frequently switch their Disease-Modifying Agents (DMA) for effectiveness and safety concerns. This study aimed to develop and compare the random forest (RF) machine learning (ML) model with the logistic regression (LR) model for predicting DMA switching among MS patients. MethodsThis retrospective longitudinal study used the TriNetX data from a federated electronic medical records (EMR) network. Between September 2010 and May 2017, adults (aged ≥18) MS patients with ≥1 DMA prescription were identified, and the earliest DMA date was assigned as the index date. Patients prescribed any DMAs different from their index DMAs were considered as treatment switch. . The RF and LR models were built with 72 baseline characteristics and trained with 70% of the randomly split data after up-sampling. Area Under the Curves (AUC), accuracy, recall, G-measure, and F-1 score were used to evaluate the model performance. ResultsIn this study, 7258 MS patients with ≥1 DMA were identified. Within two years, 16% of MS patients switched to a different DMA. The RF model obtained significantly better discrimination than the LR model (AUC = 0.65 vs. 0.63, p < 0.0001); however, the RF model had a similar predictive performance to the LR model with respect to F- and G-measures (RF: 72% and 73% vs. LR: 72% and 73%, respectively). The most influential features identified from the RF model were age, type of index medication, and year of index. ConclusionsCompared to the LR model, RF performed better in predicting DMA switch in MS patients based on AUC measures; however, judged by F- and G-measures, the RF model performed similarly to LR. Further research is needed to understand the role of ML techniques in predicting treatment outcomes for the decision-making process to achieve optimal treatment goals.

Enrollment, retention, and inclusivity of Marine Stewardship Council (MSC) eco-labelling certifications

In the past twenty years, seafood sustainability certifications have been heralded as a complement to traditional government-based fisheries management. However, certifying organizations such as the Marine Stewardship Council (MSC) have often struggled to reach fisheries in the Global South, which are less likely to already have strong top-down management practices. To further understand how MSC’s recruitment may be inhibited in different regions and nations, we developed a random forest machine-learning model which identified MSC-enrolled and unenrolled fisheries with 79% accuracy based on 36 socioeconomic characteristics. Fisheries from countries with high Human Development Indices (HDI) and those targeting high biomass regions were much more likely to enroll, reflecting a lack of socioeconomic and geographic diversity among program fisheries. Almost a third of all issued MSC certifications have also been voluntarily withdrawn, and using a similar predictive model, we found withdrawals were most likely in fisheries in overharvested regions, regardless of HDI, and those that targeted forage fish species. Results suggest MSC has struggled to recruit fisheries outside of particular regions, such as Alaska and the far North Atlantic, and to retain fisheries in need of harvest controls or other significant changes in management practices. Considering the organization’s Theory of Change and existing case studies, we outline a potential program-wide mismatch between where MSC certifications are most likely to occur, and where they are most likely to successfully generate improvements.