Machine Learning Feature Selection Research Articles

Optimal selection of the most informative nodes in Opinion Dynamics on Networks

Finding the optimal subset to observe in a network system is a fundamental problem in science and engineering, with a wide range of applications like monitoring spatial phenomena, control of epidemic spread, feature selection in machine learning, or active surveying in social studies. The goal of this paper is to address the subset selection problem on an Opinion Dynamics model where the variable of interest Y is the average opinion of the community. We consider the opinion vector X to be updated according to a Friedkin-Johnsen opinion dynamics model where every agent i is equipped with an original unknown belief ui, which is assumed to be normally distributed, and a parameter λi describing its openness to interactions. The objective function of the optimization problem is the variance reduction from the observation of the steady-state opinions of a subset K ⊆ V of agents. We show how this functional can be rewritten in terms of the Bonacich centrality and the cycle centrality of the agents in social network when the subset selection is of cardinality 1, providing particular graph-theoretic interpretations related to the network itself. In addition, first exploratory simulations highlight a behaviour which deviates from the one of known centrality measures depending on the choice of model parameters. Finally, we show that the submodularity of the functional is not guaranteed in our case and thus results taken from known literature are non-enforceable. This paves the way for further analysis.

Combining machine learning and effective feature selection for real-time stock trading in variable time-frames

The unpredictability and volatility of the stock market render it challenging to make a substantial profit using any generalised scheme. Many previous studies tried different techniques to build a machine learning model, which can make a significant profit in the US stock market by performing live trading. However, very few studies have focused on the importance of finding the best features for a particular period for trading. Our top approach used the performance to narrow down the features from a total of 148 to about 30. Furthermore, the top 25 features were dynamically selected before each time training our machine learning model. It uses ensemble learning with four classifiers: Gaussian naive Bayes, decision tree, logistic regression with L1 regularisation and stochastic gradient descent, to decide whether to go long or short on a particular stock. Our best model performed daily trade between July 2011 and January 2019, generating 54.35% profit. Finally, our work showcased that mixtures of weighted classifiers perform better than any individual predictor about making trading decisions in the stock market.

Certain Investigation on the Mango Leaves Diseases Detection Classification using Machine Learning Techniques and Feature Selection

Certain Investigation on the Mango Leaves Diseases Detection Classification using Machine Learning Techniques and Feature Selection

An Efficient Approach to Predict Eye Diseases from Symptoms Using Machine Learning and Ranker-Based Feature Selection Methods.

The eye is generally considered to be the most important sensory organ of humans. Diseases and other degenerative conditions of the eye are therefore of great concern as they affect the function of this vital organ. With proper early diagnosis by experts and with optimal use of medicines and surgical techniques, these diseases or conditions can in many cases be either cured or greatly mitigated. Experts that perform the diagnosis are in high demand and their services are expensive, hence the appropriate identification of the cause of vision problems is either postponed or not done at all such that corrective measures are either not done or done too late. An efficient model to predict eye diseases using machine learning (ML) and ranker-based feature selection (r-FS) methods is therefore proposed which will aid in obtaining a correct diagnosis. The aim of this model is to automatically predict one or more of five common eye diseases namely, Cataracts (CT), Acute Angle-Closure Glaucoma (AACG), Primary Congenital Glaucoma (PCG), Exophthalmos or Bulging Eyes (BE) and Ocular Hypertension (OH). We have used efficient data collection methods, data annotations by professional ophthalmologists, applied five different feature selection methods, two types of data splitting techniques (train-test and stratified k-fold cross validation), and applied nine ML methods for the overall prediction approach. While applying ML methods, we have chosen suitable classic ML methods, such as Decision Tree (DT), Random Forest (RF), Naive Bayes (NB), AdaBoost (AB), Logistic Regression (LR), k-Nearest Neighbour (k-NN), Bagging (Bg), Boosting (BS) and Support Vector Machine (SVM). We have performed a symptomatic analysis of the prominent symptoms of each of the five eye diseases. The results of the analysis and comparison between methods are shown separately. While comparing the methods, we have adopted traditional performance indices, such as accuracy, precision, sensitivity, F1-Score, etc. Finally, SVM outperformed other models obtaining the highest accuracy of 99.11% for 10-fold cross-validation and LR obtained 98.58% for the split ratio of 80:20.

Predictive factors for allergy at 4–6 years of age based on machine learning: A pilot study

BackgroundIn Europe, allergic diseases are the most common chronic childhood illnesses and the result of a complex interplay between genetics and environmental factors. A new approach for analyzing this complex data is to employ machine learning (ML) algorithms. Therefore, the aim of this pilot study was to find predictors for the presence of parental-reported allergy at 4–6 years of age by using feature selection in ML. MethodsA recursive ensemble feature selection (REFS) was used, with a 20% step reduction and with eight different classifiers in the ensemble, and resampling given the class unbalance. Thereafter, the Receiver Operating Characteristic Curves for five different classifiers, not included in the original ensemble feature selection technique, were calculated. ResultsIn total, 130 children (14 with and 116 without parental-reported allergy) and 248 features were included in the ML analyses. The REFS algorithm showed a result of 20 features and particularly, the Multi-layer Perceptron Classifier had an area under the curve (AUC) of 0.86 (SD 0.08). The features predictive for allergy were: tobacco exposure during pregnancy, atopic parents, gestational age, days of: diarrhea, cough, rash, and fever during first year of life, ever being exposed to antibiotics, Resistin, IL-27, MMP9, CXCL8, CCL13, Vimentin, IL-4, CCL22, GAL1, IL-6, LIGHT, and GMCSF. ConclusionsThis ML model shows that a combination of environmental exposures and cytokines can predict later allergy with an AUC of 0.86 despite the small sample size. In the future, our ML model still needs to be externally validated.

Diagnostic gene signatures and aberrant pathway activation based on m6A methylation regulators in rheumatoid arthritis.

Rheumatoid arthritis (RA) is a chronic autoimmune disease (AD) characterized by persistent synovial inflammation, bone erosion and progressive joint destruction. This research aimed to elucidate the potential roles and molecular mechanisms of N6-methyladenosine (m6A) methylation regulators in RA. An array of tissues from 233 RA and 126 control samples was profiled and integrated for mRNA expression analysis. Following quality control and normalization, the cohort was split into training and validation sets. Five distinct machine learning feature selection methods were applied to the training set and validated in validation sets. Among the six models, the LASSO_λ-1se model not only performed better in the validation sets but also exhibited more stringent performance. Two m6A methylation regulators were identified as significant biomarkers by consensus feature selection from all four methods. IGF2BP3 and YTHDC2, which are differentially expressed in patients with RA and controls, were used to predict RA diagnosis with high accuracy. In addition, IGF2BP3 showed higher importance, which can regulate the G2/M transition to promote RA-FLS proliferation and affect M1 macrophage polarization. This consensus of multiple machine learning approaches identified two m6A methylation regulators that could distinguish patients with RA from controls. These m6A methylation regulators and their target genes may provide insight into RA pathogenesis and reveal novel disease regulators and putative drug targets.

Predicting Conversions from Normal to MCI and from MCI to AD with Clinical Variables in the National Alzheimer’s Coordinating Center Uniform Data Set Version 3: Application of Big Data Analytics and Transition Probability Calculators

AbstractBackgroundClinical trials are increasingly focused on pre‐manifest and early Alzheimer’s disease (AD). Accurately predicting those who transit from normal to MCI or from MCI to dementia/AD (true positives) versus non‐progressors (false positives) is challenging. Biomarker positivity and comorbidity information are often not sufficient to identify those who are destined to have symptom progressions. Accurate identification of progressors is important to avoid un‐necessary treatment as well as to improve trial efficiency.MethodUsing only clinical variables and big data analytics, we aimed to predict those who convert from normal to MCI and from MCI to AD, estimating probabilities of these conversions with a small set of selected clinical variables. This work updates our previous work using National Alzheimer’s Coordinating Center (NACC) Uniform Data Set Version 2. (1) We also generated a user‐friendly conversion probability calculator to estimate individual subject conversion. We used data from the NACC Uniform Data Set Version 3 and machine learning feature selection methods to identify clinical variables that predicted MCI and dementia conversion. We built a pipeline with machine learning techniques to test model performance.ResultUsing NACC data (frozen 3/2020), 577 subjects were used to predict conversions from normal to MCI within 4 years, and 538 and 882 subjects were used to predict conversion from MCI to AD within 2 years and 3 years, respectively. An exploratory analysis was also conducted for amnestic MCI (aMCI) conversion. For each prediction model, we selected 20 clinical variables optimizing predictions using Receiver Operating Characteristics Area‐under‐Curve (AUC). Our model achieved 82.1% prediction accuracy, 75.9% specificity, 85.1% sensitivity, and 80.5% AUC for MCI to AD conversion within 3 years. Similar results were obtained for normal to MCI conversion. The variables which ranked highest in predicting transitions include clinician’s judgement of subject’s meaningful cognitive impairment (conversion to MCI) and CDR community affairs item (conversion to AD in 3 years).ConclusionAlthough NACC UDS is extensive, our algorithm is based on a large sample size of potential trial participants who visited or are referred to Alzheimer’s Disease Centers. The probability calculator can be used to effectively pre‐screen and enrich trial participants.

Pneumonitis Prediction Modeling of a Prospective 4DCT-Ventilation Functional Avoidance Clinical Trial

<h3>Purpose/Objective(s)</h3> Functional avoidance radiotherapy proposes to use functional imaging to reduce pulmonary toxicity by designing radiotherapy treatment plans that reduce doses to functional regions of the lung. A novel form of lung functional imaging has been proposed that uses 4DCT imaging to calculate 4DCT-based lung ventilation (4DCT-ventilation) maps. A phase II, multi-center, prospective study was completed that showed that 4DCT-ventilation functional avoidance radiotherapy results in decreased ≥ grade 2 pneumonitis when compared to historical controls. Studies are needed that apply machine learning methods to evaluate the large prospective trial dataset to elucidate which factors are most critical in predicting toxicity. The purpose of this work is to use machine learning to predict ≥ grade 2 radiation pneumonitis for patients treated with 4DCT-ventilation functional avoidance radiotherapy. <h3>Materials/Methods</h3> Lung cancer patients receiving curative chemo-radiotherapy (doses of 45-75 Gy) were accrued from 2 institutions. Patient 4DCTs along with image processing techniques were used to generate 4DCT-ventilation images. The 4DCT-ventilation images were used to generate functional avoidance plans that reduced doses to functional portions of the lung while delivering the prescribed tumor dose and respecting tolerances of organs-at-risk. 49 factors were initially used in a machine learning feature selection algorithm, and 36 factors with the highest participation were selected to train 26 classifiers to predict ≥ grade 2 pneumonitis. The assessed factors included patients (age, gender, performance status, etc.), clinical (immunotherapy, surgery, chemotherapy), dose metrics (lung, esophagus, heart, and PTV), and metrics that combine 4DCT-ventilation based function and dose (e.g., V20 to lung portions with ≥50% function). The classifiers were trained using a 5-fold cross-validation scheme. The performance of the classifiers was evaluated using accuracy and receiver operating characteristic area under the curve (AUC). <h3>Results</h3> Of the 67 accrued patients, 10 (14.9%) developed ≥ grade 2 radiation pneumonitis. The best performing classifier was a decision tree classifier with an accuracy of 0.92 and an AUC of 0.75. Dose-function and heart dosimetric indices were the metrics most predictive of ≥ grade 2 pneumonitis. <h3>Conclusion</h3> This is the first study to comprehensively evaluate factors predictive of pneumonitis for patients treated on a prospective 4DCT-ventilation functional avoidance clinical trial. Machine learning methods revealed that lung dose-function metrics and heart doses were significant in predicting toxicity. The results validate the clinical significance in reducing doses to functional portions of the lung and provide seminal clinical guidance for functional avoidance thoracic radiotherapy.

In-hospital risk stratification algorithm of Asian elderly patients

Limited research has been conducted in Asian elderly patients (aged 65 years and above) for in-hospital mortality prediction after an ST-segment elevation myocardial infarction (STEMI) using Deep Learning (DL) and Machine Learning (ML). We used DL and ML to predict in-hospital mortality in Asian elderly STEMI patients and compared it to a conventional risk score for myocardial infraction outcomes. Malaysia's National Cardiovascular Disease Registry comprises an ethnically diverse Asian elderly population (3991 patients). 50 variables helped in establishing the in-hospital death prediction model. The TIMI score was used to predict mortality using DL and feature selection methods from ML algorithms. The main performance metric was the area under the receiver operating characteristic curve (AUC). The DL and ML model constructed using ML feature selection outperforms the conventional risk scoring score, TIMI (AUC 0.75). DL built from ML features (AUC ranging from 0.93 to 0.95) outscored DL built from all features (AUC 0.93). The TIMI score underestimates mortality in the elderly. TIMI predicts 18.4% higher mortality than the DL algorithm (44.7%). All ML feature selection algorithms identify age, fasting blood glucose, heart rate, Killip class, oral hypoglycemic agent, systolic blood pressure, and total cholesterol as common predictors of mortality in the elderly. In a multi-ethnic population, DL outperformed the TIMI risk score in classifying elderly STEMI patients. ML improves death prediction by identifying separate characteristics in older Asian populations. Continuous testing and validation will improve future risk classification, management, and results.

Differential expression of MDGA1 in major depressive disorder

The identification of gene expression-based biomarkers for major depressive disorder (MDD) continues to be an important challenge. In order to identify candidate biomarkers and mechanisms, we apply statistical and machine learning feature selection to an RNA-Seq gene expression dataset of 78 unmedicated individuals with MDD and 79 healthy controls. We identify 49 genes by LASSO penalized logistic regression and 45 genes at the false discovery rate threshold 0.188. The MDGA1 gene has the lowest P-value (4.9e-5) and is expressed in the developing brain, involved in axon guidance, and associated with related mood disorders in previous studies of bipolar disorder (BD) and schizophrenia (SCZ). The expression of MDGA1 is associated with age and sex, but its association with MDD remains significant when adjusted for covariates. MDGA1 is in a co-expression cluster with another top gene, ATXN7L2 (ataxin 7 like 2), which was associated with MDD in a recent GWAS. The LASSO classification model of MDD includes MDGA1, and the model has a cross-validation accuracy of 79%. Another noteworthy top gene, IRF2BPL, is in a close co-expression cluster with MDGA1 and may be related to microglial inflammatory states in MDD. Future exploration of MDGA1 and its gene interactions may provide insights into mechanisms and heterogeneity of MDD.

Enhanced Heart Disease Prediction Based on Machine Learning and χ2 Statistical Optimal Feature Selection Model

Automatic heart disease prediction is a major global health concern. Effective cardiac treatment requires an accurate heart disease prognosis. Therefore, this paper proposes a new heart disease classification model based on the support vector machine (SVM) algorithm for improved heart disease detection. To increase prediction accuracy, the χ2 statistical optimum feature selection technique was used. The suggested model’s performance was then validated by comparing it to traditional models using several performance measures. The proposed model increased accuracy from 85.29% to 89.7%. Additionally, the componential load was reduced by half. This result indicates that our system outperformed other state-of-the-art methods in predicting heart disease.

Solar Radiation Forecasting Using Machine Learning and Ensemble Feature Selection

Accurate solar radiation forecasting is essential to operate power systems safely under high shares of photovoltaic generation. This paper compares the performance of several machine learning algorithms for solar radiation forecasting using endogenous and exogenous inputs and proposes an ensemble feature selection method to choose not only the most related input parameters but also their past observations values. The machine learning algorithms used are: Support Vector Regression (SVR), Extreme Gradient Boosting (XGBT), Categorical Boosting (CatBoost) and Voting-Average (VOA), which integrates SVR, XGBT and CatBoost. The proposed ensemble feature selection is based on Pearson coefficient, random forest, mutual information and relief. Prediction accuracy is evaluated based on several metrics using a real database from Salvador, Brazil. Different prediction time-horizons are considered: 1 h, 2 h and 3 h ahead. Numerical results demonstrate that the proposed ensemble feature selection approach improves forecasting accuracy and that VOA performs better than the other algorithms in all prediction time horizons.

PSVIII-B-13 Machine Learning-Based co-Expression Network Analysis Unravels Fertility-Related Genes in Beef Cattle

Abstract Reproductive efficiency is a critical component of a sustainable cow-calf system. While several studies have identified factors underlying fertility, the genetic mechanisms contributing to this complex trait are still unclear. Our goal was to identify a set of predictive biomarker signatures from transcriptomics data associated with female cattle fertility. We implemented a multi-tiered approach using machine learning (ML) feature selection, gene co-expression network, and functional analysis. To this end, we retrieved public data from the Gene Expression Omnibus database (GEO GSE171577). The RNA-Seq data was generated from uterine luminal epithelial cells of recipient cows sampled on day four before embryo transfer. The data (n = 18 non-pregnant – NP and n = 25 pregnant – P) were analyzed using a standard pipeline based on FastQC, MultiQC, STAR, and DESeq2. Genes with expression values &amp;gt; 0.5 counts per million in 50% of the samples were filtered out. Feature and model selection was implemented through BioDiscML. Further, the PCIT algorithm was used to create gene co-expression networks from 15,039 genes kept after quality control. Our ML approach identified nine genes as predictors of pregnancy status. The genes included: SERPINE3, MRTFA, MEF2B, NAA16, ARHGEF7, PDCD1, FNDC1, ENSBTAG00000054585, and ENSBTAG00000019474. The networks from P and NP cows resulted in five and four thousand significantly co-expressed gene pairs, respectively. We then kept 1,837 pairs with a |r| &amp;gt; 0.7 and were co-expressed with the gene predictors from the ML analysis. Biological processes, such as vasculature development, oxidative phosphorylation, and focal adhesion were over-represented by genes from the P network. We identified immune system development, negative regulation of the biological process, and protein modification over-represented processes in the NP gene network. We have demonstrated the potential of combining different methods to identify fertility-related biomarkers and have provided insights into the complex genomic basis underlying pregnancy establishment in cattle.

Identification of diagnostic mRNA biomarkers in whole blood for ankylosing spondylitis using WGCNA and machine learning feature selection.

Ankylosing spondylitis (AS) is a common inflammatory spondyloarthritis affecting the spine and sacroiliac joint that finally results in sclerosis of the axial skeleton. Aside from human leukocyte antigen B27, transcriptomic biomarkers in blood for AS diagnosis still remain unknown. Hence, this study aimed to investigate credible AS-specific mRNA biomarkers from the whole blood of AS patients by analyzing an mRNA expression profile (GSE73754) downloaded Gene Expression Omnibus, which includes AS and healthy control blood samples. Weighted gene co-expression network analysis was performed and revealed three mRNA modules associated with AS. By performing gene set enrichment analysis, the functional annotations of these modules revealed immune biological processes that occur in AS. Several feature mRNAs were identified by analyzing the hubs of the protein-protein interaction network, which was based on the intersection between differentially expressed mRNAs and mRNA modules. A machine learning-based feature selection method, SVM-RFE, was used to further screen out 13 key feature mRNAs. After verifying by qPCR, IL17RA, Sqstm1, Picalm, Eif4e, Srrt, Lrrfip1, Synj1 and Cxcr6 were found to be significant for AS diagnosis. Among them, Cxcr6, IL17RA and Lrrfip1 were correlated with severity of AS symptoms. In conclusion, our findings provide a framework for identifying the key mRNAs in whole blood of AS that is conducive for the development of novel diagnostic markers for AS.

Machine Learning Models and Data-Balancing Techniques for Credit Scoring: What Is the Best Combination?

Forecasting the creditworthiness of customers is a central issue of banking activity. This task requires the analysis of large datasets with many variables, for which machine learning algorithms and feature selection techniques are a crucial tool. Moreover, the percentages of “good” and “bad” customers are typically imbalanced such that over- and undersampling techniques should be employed. In the literature, most investigations tackle these three issues individually. Since there is little evidence about their joint performance, in this paper, we try to fill this gap. We use five machine learning classifiers, and each of them is combined with different feature selection techniques and various data-balancing approaches. According to the empirical analysis of a retail credit bank dataset, we find that the best combination is given by random forests, random forest recursive feature elimination and random oversampling.

Prediction of Occupant Behavior toward Natural Ventilation in Japanese Dwellings: Machine Learning Models and Feature Selection

Occupant behavior based on natural ventilation has a significant impact on building energy consumption. It is important for the quantification of occupant-behavior models to select observed variables, i.e., features that affect the state of window opening and closing, and to consider machine learning models that are effective in predicting this state. In this study, thermal comfort was investigated, and machine learning data were analyzed for 30 houses in Gifu, Japan. Among the selected machine learning models, the logistic regression and deep neural network models produced consistently excellent results. The accuracy of the prediction of open and closed windows differed among the models, and the factors influencing the window-opening behaviors of the occupants differed from those influencing their window-closing behavior. In the selection of features, the analysis using thermal indices representative of the room and cooling features showed excellent results, indicating that cooling features, which have conflicting relationships with natural ventilation, are useful for improving the accuracy of occupant-behavior prediction. The present study indicates that building designers should incorporate occupant behavior based on natural ventilation into their designs.

Differentiating between liver diseases by applying multiclass machine learning approaches to transcriptomics of liver tissue or blood-based samples

Background & AimsLiver disease carries significant healthcare burden and frequently requires a combination of blood tests, imaging, and invasive liver biopsy to diagnose. Distinguishing between inflammatory liver diseases, which may have similar clinical presentations, is particularly challenging. In this study, we implemented a machine learning pipeline for the identification of diagnostic gene expression biomarkers across several alcohol-associated and non-alcohol-associated liver diseases, using either liver tissue or blood-based samples.MethodsWe collected peripheral blood mononuclear cells (PBMCs) and liver tissue samples from participants with alcohol-associated hepatitis (AH), alcohol-associated cirrhosis (AC), non-alcohol-associated fatty liver disease, chronic HCV infection, and healthy controls. We performed RNA sequencing (RNA-seq) on 137 PBMC samples and 67 liver tissue samples. Using gene expression data, we implemented a machine learning feature selection and classification pipeline to identify diagnostic biomarkers which distinguish between the liver disease groups. The liver tissue results were validated using a public independent RNA-seq dataset. The biomarkers were computationally validated for biological relevance using pathway analysis tools.ResultsUtilizing liver tissue RNA-seq data, we distinguished between AH, AC, and healthy conditions with overall accuracies of 90% in our dataset, and 82% in the independent dataset, with 33 genes. Distinguishing 4 liver conditions and healthy controls yielded 91% overall accuracy in our liver tissue dataset with 39 genes, and 75% overall accuracy in our PBMC dataset with 75 genes.ConclusionsOur machine learning pipeline was effective at identifying a small set of diagnostic gene biomarkers and classifying several liver diseases using RNA-seq data from liver tissue and PBMCs. The methodologies implemented and genes identified in this study may facilitate future efforts toward a liquid biopsy diagnostic for liver diseases.Lay summaryDistinguishing between inflammatory liver diseases without multiple tests can be challenging due to their clinically similar characteristics. To lay the groundwork for the development of a non-invasive blood-based diagnostic across a range of liver diseases, we compared samples from participants with alcohol-associated hepatitis, alcohol-associated cirrhosis, chronic hepatitis C infection, and non-alcohol-associated fatty liver disease. We used a machine learning computational approach to demonstrate that gene expression data generated from either liver tissue or blood samples can be used to discover a small set of gene biomarkers for effective diagnosis of these liver diseases.

A novel biomarker selection method combining graph neural network and gene relationships applied to microarray data

BackgroundThe discovery of critical biomarkers is significant for clinical diagnosis, drug research and development. Researchers usually obtain biomarkers from microarray data, which comes from the dimensional curse. Feature selection in machine learning is usually used to solve this problem. However, most methods do not fully consider feature dependence, especially the real pathway relationship of genes.ResultsExperimental results show that the proposed method is superior to classical algorithms and advanced methods in feature number and accuracy, and the selected features have more significance.MethodThis paper proposes a feature selection method based on a graph neural network. The proposed method uses the actual dependencies between features and the Pearson correlation coefficient to construct graph-structured data. The information dissemination and aggregation operations based on graph neural network are applied to fuse node information on graph structured data. The redundant features are clustered by the spectral clustering method. Then, the feature ranking aggregation model using eight feature evaluation methods acts on each clustering sub-cluster for different feature selection.ConclusionThe proposed method can effectively remove redundant features. The algorithm’s output has high stability and classification accuracy, which can potentially select potential biomarkers.

Causal network inference in a dam system and its implications on feature selection for machine learning forecasting

A fundamental goal across many research fields is to explain possible mechanisms behind a phenomenon and infer the correct causal relationships between variables. In this work, we employed various causal inference methods to derive the causal network of a dam system from time series data. Here we explored the lagged effects of water levels in two dams, climate and weather variables, and domestic and agricultural water demands on each other. Among the methods considered, we demonstrated that convergent cross mapping (CCM), a method for inferring causal relationships in complex systems using time series data, is the most consistent with an actual dam system: (1) causal links were consistent with the direction of the physical flow of water, (2) the effects of climate and weather variables were successfully captured, (3) the time lags shed light on the dynamics of the dam system and possibly reflected planning schedules which are not explicit in the data. Our results captured both intuitive and counter-intuitive causal links, some of which were validated by domain experts. Using the resulting causal links to pre-select the input variables in machine learning-based forecasting models significantly reduces the prediction errors compared to using randomly selected features. Specifically, the best reduction in MAE is 4.2–4.4 meters, which corresponds to an improvement of 2.8–3.0 times lower than using random selection of features. CCM was also able to filter the top 20 significant predictors, where further addition of other variables yielded negligible improvement in the MAE. This is the first work that demonstrates successful inference of time-lagged causal network of endogenous and exogenous variables in a dam system.

EARLY DIFFERENTIATION BETWEEN SEPSIS AND STERILE INFLAMMATION VIA URINARY GENE SIGNATURES OF METABOLIC DYSREGULATION.

ABSTRACTObjective: The aim of this study was to characterize early urinary gene expression differences between patients with sepsis and patients with sterile inflammation and summarize in terms of a reproducible sepsis probability score. Design: This was a prospective observational cohort study. Setting: The study was conducted in a quaternary care academic hospital. Patients: One hundred eighty-six sepsis patients and 78 systemic inflammatory response syndrome (SIRS) patients enrolled between January 2015 and February 2018. Interventions: Whole-genome transcriptomic analysis of RNA was extracted from urine obtained from sepsis patients within 12 hours of sepsis onset and from patients with surgery-acquired SIRS within 4 hours after major inpatient surgery. Measurements and Main Results: We identified 422 of 23,956 genes (1.7%) that were differentially expressed between sepsis and SIRS patients. Differentially expressed probes were provided to a collection of machine learning feature selection models to identify focused probe sets that differentiate between sepsis and SIRS. These probe sets were combined to find an optimal probe set (UrSepsisModel) and calculate a urinary sepsis score (UrSepsisScore), which is the geometric mean of downregulated genes subtracted from the geometric mean of upregulated genes. This approach summarizes the expression values of all decisive genes as a single sepsis score. The UrSepsisModel and UrSepsisScore achieved area under the receiver operating characteristic curves 0.91 (95% confidence interval, 0.86–0.96) and 0.80 (95% confidence interval, 0.70–0.88) on the validation cohort, respectively. Functional analyses of probes associated with sepsis demonstrated metabolic dysregulation manifest as reduced oxidative phosphorylation, decreased amino acid metabolism, and decreased oxidation of lipids and fatty acids. Conclusions: Whole-genome transcriptomic profiling of urinary cells revealed focused probe panels that can function as an early diagnostic tool for differentiating sepsis from sterile SIRS. Functional analysis of differentially expressed genes demonstrated a distinct metabolic dysregulation signature in sepsis.