Monte Carlo Cross Validation Method Research Articles

Management zone classification for variable-rate soil residual herbicide applications

AbstractThe use of soil residual herbicides, along with other practices that diversify weed management strategies, have been recommended to improve weed management and deter the progression of herbicide resistance. Although soil characteristics influence recommended application rates for these herbicides, the common practice is to apply a uniform dose of soil residual herbicides across fields with variable soil characteristics. Mapping fields for soil characteristics that dictate the optimal dose of soil residual herbicides could improve the efficiency and effectiveness of these herbicides, as well as improve environmental stewardship. The objectives of this research were to develop and quantify the accuracy of management zone classifications for variable-rate residual herbicide applications using multiple soil data sources and soil sampling intensities. The maps were created from soil data that included (i) Soil Survey Geographic database (SSURGO), (ii) soil samples (SS), (iii) soil samples regressed onto soil electrical conductivity (EC) measurements (SSEC), (iv) soil samples with organic matter (OM) data from SmartFirmer® (SF) sensors (SSSF), and (v) soil samples regressed onto EC measurements plus OM data from SmartFirmer® sensor (SSECSF). A modified Monte Carlo cross validation method was used on ten commercial Indiana fields to generate 36,000 maps across all sources of spatial soil data, sampling density, and three representative herbicides (pyroxasulfone, s-metolachlor, and metribuzin). Maps developed from SSEC data were most frequently ranked with the highest management zone classification accuracy compared to maps developed from SS data. However, SS and SSEC maps concurrently had the highest management zone classification accuracy of 34% among maps developed across all fields, herbicides, and sampling intensities. One soil sample per hectare was the most reliable sampling intensity to generate herbicide application management zones compared to one soil sample for every 2 or 4 hectares. In conclusion, soil sampling with ECa data should be used for defining the management zones for variable-rate (VR) residual herbicide applications.

Comparison of the diagnostic efficacy of mathematical models in distinguishing ultrasound imaging of breast nodules

This study compared the diagnostic efficiency of benign and malignant breast nodules using ultrasonographic characteristics coupled with several machine-learning models, including logistic regression (Logistics), partial least squares discriminant analysis (PLS-DA), linear support vector machine (Linear SVM), linear discriminant analysis (LDA), K-nearest neighbor (KNN), artificial neural network (ANN) and random forest (RF). The clinical information and ultrasonographic characteristics of 926 female patients undergoing breast nodule surgery were collected and their relationships were analyzed using Pearson's correlation. The stepwise regression method was used for variable selection and the Monte Carlo cross-validation method was used to randomly divide these nodule cases into training and prediction sets. Our results showed that six independent variables could be used for building models, including age, background echotexture, shape, calcification, resistance index, and axillary lymph node. In the prediction set, Linear SVM had the highest diagnosis rate of benign nodules (0.881), and Logistics, ANN and LDA had the highest diagnosis rate of malignant nodules (0.910~0.912). The area under the ROC curve (AUC) of Linear SVM was the highest (0.890), followed by ANN (0.883), LDA (0.880), Logistics (0.878), RF (0.874), PLS-DA (0.866), and KNN (0.855), all of which were better than that of individual variances. On the whole, the diagnostic efficacy of Linear SVM was better than other methods.

Rapid detection of maize seed germination rate based on Gaussian process regression with selection kernel function

Near-infrared spectroscopy (NIR) detection technology enables nondestructive analysis of samples. To achieve fast detection of maize seed germination rates using NIR spectroscopy, 245 samples of maize seeds from three different varieties were purchased from the market and subjected to artificial aging to obtain maize seeds with varying levels of vigor. NIR spectroscopy was collected for each sample, and germination experiments were performed. The Monte Carlo cross-validation method was employed to eliminate 12 outliers, and the baseline correction method along with multiplicative scatter correction (MSC) was applied to the NIR. Partial least squares (PLS), support vector machine (SVM) models, and Gaussian process regression (GPR) models based on different kernel functions were established for detecting maize germination rates. Through comparison of the results of different models, it was found that the GPR model based on the Matern32 kernel function performed best. The R² values for the training and testing sets of this model were 0.9727 and 0.9599, respectively. Experimental results showed that the Matern32-GPR model exhibited greater stability and superior capability in detecting maize seed germination rates. This method can provide a reference for other applications based on NIR detection.

Pseudo-random number generator based on linear congruence and delayed Fibonacci method

Pseudo-random number generation techniques are an essential tool to correctly test machine learning processes. The methodologies are many, but also the possibilities to combine them in a new way are plenty. Thus, there is a chance to create mechanisms potentially useful in new and better generators. In this paper, we present a new pseudo-random number generator based on a hybrid of two existing generators - a linear congruential method and a delayed Fibonacci technique. We demonstrate the implementation of the generator by checking its correctness and properties using chi-square, Kolmogorov and TestU01.1.2.3 tests and we apply the Monte Carlo Cross Validation method in classification context to test the performance of the generator in practice.

Prediction of response to cardiac resynchronization therapy using a multi-feature learning method.

We hypothesized that a multiparametric evaluation, based on the combination of electrocardiographic and echocardiographic parameters, could enhance the appraisal of the likelihood of reverse remodeling and prognosis of favorable clinical evolution to improve the response of cardiac resynchronization therapy (CRT). Three hundred and twenty-three heart failure patients were retrospectively included in this multicenter study. 221 patients (68%) were responders, defined by a decrease in left ventricle end-systolic volume ≥15% at the 6-month follow-up. In addition, strain data coming from echocardiography were analyzed with custom-made signal processing methods. Integrals of regional longitudinal strain signals from the beginning of the cardiac cycle to strain peak and to the instant of aortic valve closure were analyzed. QRS duration, septal flash and different other features manually extracted were also included in the analysis. The random forest (RF) method was applied to analyze the relative feature importance, to select the most significant features and to build an ensemble classifier with the objective of predicting response to CRT. The set of most significant features was composed of Septal Flash, E, E/A, E/EA, QRS, left ventricular end-diastolic volume and eight features extracted from strain curves. A Monte Carlo cross-validation method with 100 runs was applied, using, in each run, different random sets of 80% of patients for training and 20% for testing. Results show a mean area under the curve (AUC) of 0.809 with a standard deviation of 0.05. A multiparametric approach using a combination of echo-based parameters of left ventricular dyssynchrony and QRS duration helped to improve the prediction of the response to cardiac resynchronization therapy.

Machine Learning Predicts Prolonged Acute Hypoxemic Respiratory Failure in Pediatric Severe Influenza.

We aimed to identify which clinical variables best predict prolonged acute hypoxemic respiratory failure in influenza-infected critically ill children. Acute hypoxemic respiratory failure was defined using hypoxemia cutoffs from international consensus definitions of acute respiratory distress syndrome in patients with ventilatory support. Prolonged acute hypoxemic respiratory failure was defined by acute hypoxemic respiratory failure criteria still present at PICU day 7. In this prospective multicenter study across 34 PICUs from November 2009 to April 2018, we included children (< 18 yr) without comorbid risk factors for severe disease. We used a Monte Carlo cross validation method with N 2 random train-test splits at a 70-30% proportion per model. Using clinical data at admission (day 1) and closest to 8 am on PICU day 2, we calculated the area under the receiver operating characteristic curve using random forests machine learning algorithms and logistic regression. We included 258 children (median age = 6.5 yr) and 11 (4.2%) died. By day 2, 65% (n = 165) had acute hypoxemic respiratory failure dropping to 26% (n = 67) with prolonged acute hypoxemic respiratory failure by day 7. Those with prolonged acute hypoxemic respiratory failure had a longer ICU stay (16.5 vs 4.0 d; p < 0.001) and higher mortality (13.4% vs 1.0%). A multivariable model using random forests with 10 admission and eight day 2 variables performed best (0.93 area under the receiver operating characteristic curve; 95 CI%: 0.90-0.95) where respiratory rate, Fio2, and pH on day 2 were the most important factors. In this prospective multicentric study, most children with influenza virus-related respiratory failure with prolonged acute hypoxemic respiratory failure can be identified early in their hospital course applying machine learning onto routine clinical data. Further validation is needed prior to bedside implementation.

Weighted support vector machine algorithm for efficient classification and prediction of binary response data

This paper proposes a weighted Support Vector Machine (w-SVM) method for efficient class prediction in binary response data sets. The proposed method was obtained by introducing weights which utilizes the point biserial correlation between each of the predictors and the dichotomized response variable into the standard SVM algorithm to maximize the classification accuracy. The optimal value of the proposed w-SVM cost and each of the kernels parameters were determined by grid search in a 10-fold cross validation resampling method. Monte-Carlo Cross Validation method was employed to examine the predictive power of the proposed method by partitioning the data into train and test samples using different sampling splitting ratios. Application of the proposed method on the simulated data sets yielded high prediction accuracy on the test sample. Results from other performance indices further gave credence to the efficiency of the proposed method. The performance of the proposed method was compared with three of the state-of-the art machine learning methods including the standard SVM and the result showed the superiority of this method over others. Finally, the results generally show that the modified algorithm with Radial Basis Function (RBF) Kernel perform excellently and achieved the best predictive performance than any of the existing classifiers considered.

Revealing hub pathway cross-talk for premature newborns with bronchopulmonary dysplasia by the integration of pathway analysis and Monte Carlo Cross-Validation.

The objective of this study was to reveal hub pathway cross-talk for premature newborns with bronchopulmonary dysplasia (BPD) based on the pathway enrichment analysis and Monte Carlo Cross-Validation (MCCV) method. The inference of key pathway cross-talk consisted of four parts: i) Identifying differentially expressed genes (DEGs); ii) detecting differentially expressed pathways (DEPs); iii) computing discriminating score (DS) for each pair of DEPs or cross-talk and investigating seed cross-talk through the random forest (RF) algorithm and iv) extracting hub cross-talk dependent on the MCCV method. The results showed that a total of 132 DEGs and 137 DEPs were obtained across BPD patients and normal controls. Using the DS and RF algorithm, 10 seed DEP cross-talk were detected. By conducting the MCCV on seed cross-talk, 3 hub cross-talk for BPD were uncovered: i) The pair of pathways role of interleukin-17F (IL-17F) in allergic inflammatory airway diseases and role of IL-17A in psoriasis; ii) the pair of pathways role of IL random forest 17A in psoriasis and IL-17A signaling in fibroblasts and ii) the pair of pathways IL-17A signaling in airway cells and role of hypercytokinemia/hyperchemokinemia in the pathogenesis of influenza. These 3 hub cross-talk among DEPs might give an insight to reveal the molecular mechanism underlying the premature newborns with BPD.

Predicting pathway cross-talks in ankylosing spondylitis through investigating the interactions among pathways.

Given that the pathogenesis of ankylosing spondylitis (AS) remains unclear, the aim of this study was to detect the potentially functional pathway cross-talk in AS to further reveal the pathogenesis of this disease. Using microarray profile of AS and biological pathways as study objects, Monte Carlo cross-validation method was used to identify the significant pathway cross-talks. In the process of Monte Carlo cross-validation, all steps were iterated 50 times. For each run, detection of differentially expressed genes (DEGs) between two groups was conducted. The extraction of the potential disrupted pathways enriched by DEGs was then implemented. Subsequently, we established a discriminating score (DS) for each pathway pair according to the distribution of gene expression levels. After that, we utilized random forest (RF) classification model to screen out the top 10 paired pathways with the highest area under the curve (AUCs), which was computed using 10-fold cross-validation approach. After 50 bootstrap, the best pairs of pathways were identified. According to their AUC values, the pair of pathways, antigen presentation pathway and fMLP signaling in neutrophils, achieved the best AUC value of 1.000, which indicated that this pathway cross-talk could distinguish AS patients from normal subjects. Moreover, the paired pathways of SAPK/JNK signaling and mitochondrial dysfunction were involved in 5 bootstraps. Two paired pathways (antigen presentation pathway and fMLP signaling in neutrophil, as well as SAPK/JNK signaling and mitochondrial dysfunction) can accurately distinguish AS and control samples. These paired pathways may be helpful to identify patients with AS for early intervention.

Identification of significant pathway cross-talk in rheumatoid arthritis by the Monte Carlo cross-validation method.

We attempted to identify significant pathway cross-talk in rheumatoid arthritis (RA) by the Monte Carlo cross-validation (MCCV) method. We therefore obtained and preprocessed the gene expression profile of RA. MCCV involves identifying differentially expressed genes (DEGs), identifying differential pathways (DPs), calculating the discriminating score (DS) of the pathway cross-talk, and random forest (RF) classification. We carried out 50 bootstrap iterations of MCCV to identify the key instances of pathway cross-talk involved in RA. We identified a total of 17 significant DEGs and 15 significant DPs by comparing RA samples and normal controls. We found the most significant difference between RA and the normal controls in the eIF4 and p70S6K signaling regulation pathway. Furthermore, we identified 10 instances of pathway cross-talk with the best classification performance for RA and normal controls, using the RF classification model. All of the top 10 pathway pairs involved cross-talk with eIF4 and p70S6K signaling regulation, and the other 10 pathways were immune-related. By MCCV, we identified one critical DP and 10 significant instances of pathway cross-talk in RA. We propose that the eIF4 and p70S6K signaling regulation pathway and the other significant instances of pathway cross-talk play key roles in the occurrence and development of RA, and are potential predictive and prognostic markers for RA.

Abstract P3-06-19: Gene expression profiles accompanying phenotypic changes during non-malignant breast epithelial cells acini formation to explain MRI phenotypes

Abstract Magnetic resonance imaging (MRI) captures the three dimensional way in which tumors are organized in the breast, defined as imaging phenotypes in the I-SPY 1 trial. We developed a gene set based on the way in which breast epithelial cells aggregate and organize in three dimensional cultures. We investigated whether these organizational genes correspond to the imaging phenotypes. MRI phenotypes have been shown to correspond to the degree of response to neoadjuvant chemotherapy, and are used to predict the ability to achieve breast conservation treatment (Mukhtar et al, Ann Surg Oncol, 20: 3823–3830, 2013). We hypothesized that the molecular profile accompanying phenotypic changes occurring during the organization process of non-malignant acini may explain the molecular basis of MRI tumor phenotypes. We have developed prediction models for MRI phenotypes based on expression profiles identified during the organization process of non-malignant breast epithelial cells in three-dimensional laminin-rich extracellular matrix (Fournier et al. Cancer Res, 66, 7095-102, 2006). We analyzed a subset of 324 organizational genes in 147 samples from stage II-III breast cancer in the I-SPY 1 trial cohort with Agilent microarrays and MRI annotation (CALGB 150007/150012; ACRIN 6657). MRI phenotype of the index lesion was assessed by the site radiologists using the following radiographic criteria: A) well defined unicentric mass; B) well-defined multilobulated mass; C) area enhancement with nodularity; D) area enhancement without nodularity; E) Septal spreading. The distribution of phenotypes in I-SPY 1 was: A: 16%, B: 33%, C: 30%, D: 15%, E: 7%. We developed predictors for MRI phenotypes dichotomized as either "well-defined" (A and B) or "non-well-defined" (C, D and E). Using patent-pending algorithms we selected a subset of the organizational genes with the greatest predictive power to identify MRI "well-defined" phenotypes in the I-SPY 1 cohort. Then prediction models were developed using the 50 top ranking genes and logistic regression methods. We followed Monte-Carlo cross validation method to make sure that the performance of a model is not a result of over-fitting the model to the sample data using separate datasets to support the modeling. The samples were randomly partitioned 10,000 times in a 85% training, and 15% test ratio to test models performance. The resulting performance of predictive models in the test set had an average classification accuracy of greater than 80% (ROC statistics AUC&amp;gt;0.8) using gene expression measurements from between 16 to 20 genes. For random lists of 16-20 genes, the accuracy was approximately 50% (ROC AUC∼0.5). Amongst the MRI models were several genes that are known to regulate key cellular pathways such as cell division, metabolism, and migration using MetaCore pathway analysis. Taken together, the results suggest that the MRI phenotypes may be a manifestation of the organization genes that determine behavior in three dimensional culture. Future research includes confirming these results using an independent dataset, defining the potential drivers of MRI phenotypes, and determining if putative drivers provide a key contribution to tumor subtypes. Citation Format: Marcia V Fournier, Alan Derr, Alex Margulis, Kevin Reid, Sara Brumbaugh, Richard Anderson, W Fraser Symmans, Laura Esserman, Nola Hylton. Gene expression profiles accompanying phenotypic changes during non-malignant breast epithelial cells acini formation to explain MRI phenotypes [abstract]. In: Proceedings of the Thirty-Seventh Annual CTRC-AACR San Antonio Breast Cancer Symposium: 2014 Dec 9-13; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2015;75(9 Suppl):Abstract nr P3-06-19.