Random Forest Score Research Articles

The impact of tobacco exposure on tumor microenvironment and prognosis in lung adenocarcinoma by integrative analysis of multi-omics data

BackgroundImmune checkpoint inhibitors (ICIs) are standard therapies for patients with advanced lung adenocarcinoma and significantly improve treatment outcomes. The effect of tobacco smoking on the response of immune checkpoint inhibitors is somewhat diverging. Here, we assessed the impact of tobacco exposure on the tumor microenvironment and developed a feasible tool for predicting prognosis. MethodsWhole exon sequence data and the corresponding clinical information were downloaded from the Cancer Genome Atlas. The signature was developed by the Random Forest algorithm. CIBERSORTx online tool was used to estimate immune infiltration. Functional assays were performed to assess the roles of tobacco exposure in cancer cells. Immunohistochemistry (IHC) was performed to identify and validate the immune activation status. ResultsThe TMB of lifelong non-smoker, current reformed smoker for over 15 years, current reformed smoker<15 years and current smoker had a significantly increasing trend in LUAD patients. In vitro tobacco exposure promoted the expression of PD-L1 and malignant phenotype of LUAD cells. In addition, patients with high Random Forest score (RFscore) had a poorer prognosis than those with low RFscore. The ROC curve analysis of RFscore revealed a promising prognostic capability. Memory activated CD4 + T cells, CD8 + t cells and memory B cells were noticeably enriched in the high RFscore group and PDCD1 appreciably upregulated in the high RFscore group as well. Furthermore, IHC results suggested that patients with high RFscore remained an immune activation status, indicating a positive correlation between RFscore and patient’s immune status. ConclusionOur analysis provides further insight into the profound impacts of tobacco exposure on tumor immune microenvironment and envisions integrative predictive models of RFscore, predicting the prognosis of smoking lung adenocarcinoma, which might help to understand the potential mechanism of smoking exposure on tumor immune microenvironment.

Ensemble‐ and distance‐based feature ranking for unsupervised learning

In this work, we propose two novel (groups of) methods for unsupervised feature ranking and selection. The first group includes feature ranking scores (Genie3 score, RandomForest score) that are computed from ensembles of predictive clustering trees. The second method is URelief, the unsupervised extension of the Relief family of feature ranking algorithms. Using 26 benchmark data sets and 5 baselines, we show that both the Genie3 score (computed from the ensemble of extra trees) and the URelief method outperform the existing methods and that Genie3 performs best overall, in terms of predictive power of the top-ranked features. Additionally, we analyze the influence of the hyper-parameters of the proposed methods on their performance, and show that for the Genie3 score the highest quality is achieved by the most efficient parameter configuration. Finally, we propose a way of discovering the location of the features in the ranking, which are the most relevant in reality.

ILBE for Computational Identification of Linear B-cell Epitopes by Integrating Sequence and Evolutionary Features

Linear B-cell epitopes are critically important for immunological applications, such as vaccine design, immunodiagnostic test, and antibody production, as well as disease diagnosis and therapy. The accurate identification of linear B-cell epitopes remains challenging despite several decades of research. In this work, we have developed a novel predictor, Identification of Linear B-cell Epitope (iLBE), by integrating evolutionary and sequence-based features. The successive feature vectors were optimized by a Wilcoxon-rank sum test. Then the random forest (RF) algorithm using the optimal consecutive feature vectors was applied to predict linear B-cell epitopes. We combined the RF scores by the logistic regression to enhance the prediction accuracy. iLBE yielded an area under curve score of 0.809 on the training dataset and outperformed other prediction models on a comprehensive independent dataset. iLBE is a powerful computational tool to identify the linear B-cell epitopes and would help to develop penetrating diagnostic tests. A web application with curated datasets for iLBE is freely accessible at http://kurata14.bio.kyutech.ac.jp/iLBE/.

Comparison of three similarity scores for bullet LEA matching

Recent advances in microscopy have made it possible to collect 3D topographic data, enabling more precise virtual comparisons based on the collected 3D data as a supplement to traditional comparison microscopy and 2D photography. Automatic comparison algorithms have been introduced for various scenarios, such as matching cartridge cases [1,2] or matching bullet striae [3–5]. One key aspect of validating these automatic comparison algorithms is to evaluate the performance of the algorithm on external tests, that is, using data which were not used to train the algorithm. Here, we present a discussion of the performance of the matching algorithm [6] in three studies conducted using different Ruger weapons. We consider the performance of three scoring measures: random forest score, cross correlation, and consecutive matching striae (CMS) at the land-to-land level and, using Sequential Average Maxima scores, also at the bullet-to bullet level. Cross correlation and random forest scores both result in perfect discrimination of same-source and different-source bullets. At the land-to-land level, discrimination for both cross correlation and random forest scores (based on area under the curve, AUC) is excellent (≥0.90).

574P - A metabolomic recurrence score for risk-stratification of elderly patients (pts) with early colorectal cancer (eCRC)

BackgroundAdjuvant treatment decisions for pts with eCRC are currently based on suboptimal risk stratification factors, especially for elderly pts. Metabolomics measures multiple cancer-related metabolites with potential to identify new biomarkers in this setting. We have shown that serum metabolomics can discriminate pts with eCRC from pts with metastatic CRC (mCRC). We hypothesized that a metabolomic score derived from pts with mCRC may identify pts with eCRC with an increased risk of relapse. This hypothesis was tested in a cohort of elderly pts with eCRC. MethodsSerum samples from 103 pts aged ≥70 were collected from four clinical trials with 5 years follow up. These samples were derived from 55 pts with eCRC, and 48 with mCRC. Samples were retrospectively analyzed via proton nuclear magnetic resonance (1H NMR), to assess their metabolomic fingerprints. A Random Forest (RF) classification model was built using a training set of 30 eCRC pts free from relapse at 5 years and all mCRC pts (N=48). This model was then applied to a validation set constituted by the remaining eCRC pts (10 relapse-free and 15 relapsed). A risk-of-recurrence score was built on the basis of the likelihood of the sample being misclassified as metastatic. ResultsIn the eCRC group, 44% (n=24) received adjuvant chemotherapy, and 27% (n=15) experienced relapse. In the training set, the RF model discriminated eCRC from mCRC with an accuracy of 74.4%. The RF risk of recurrence score correlated with relapse, with an AUC of 0.754 in ROC analysis. In the training set, by maximizing specificity and sensitivity, a threshold for the RF score was set at 0.55. In the validation set, using this threshold, an AUC of 0.727 in ROC analysis, and a prediction accuracy of 76% (73.3% sensitivity, 80% specificity) were obtained in predicting relapse. ConclusionsSerum metabolomics performed on post-operative samples of elderly pts with eCRC identifies pts with higher risk of relapse with good accuracy. This may represent a tool to refine risk stratification in this population, to maximize the benefit from adjuvant chemotherapy. Based on these results, a prospective trial is ongoing (LIquid BIopsy and METabolomics in CRC - LIBIMET). Legal entity responsible for the studyThe authors. FundingFondazione Sandro Pitigliani per la Lotto Contro i Tumori - ONLUS. DisclosureS. Di Donato: Advisory / Consultancy, Travel / Accommodation / Expenses: Amgen; Advisory / Consultancy, Travel / Accommodation / Expenses: Lilly; Advisory / Consultancy, Travel / Accommodation / Expenses: Roche; Advisory / Consultancy, Travel / Accommodation / Expenses: Servier; Travel / Accommodation / Expenses: Celgene; Advisory / Consultancy, Travel / Accommodation / Expenses: Sanofi; Advisory / Consultancy, Travel / Accommodation / Expenses: Bayer. E. Mori: Travel / Accommodation / Expenses: Bayer. L. Malorni: Honoraria (self): AstraZeneca; Advisory / Consultancy: Novartis; Travel / Accommodation / Expenses: Janssen; Honoraria (self), Advisory / Consultancy, Research grant / Funding (self): Pfizer; Travel / Accommodation / Expenses: Roche. D. Becheri: Travel / Accommodation / Expenses: Daiichi-Sankyo; Travel / Accommodation / Expenses: Bristol-Myers Squibb. A. Di Leo: Honoraria (self), Advisory / Consultancy, Research grant / Funding (institution), Travel / Accommodation / Expenses: AstraZeneca; Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Bayer; Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Celgene; Advisory / Consultancy, Travel / Accommodation / Expenses: Daiichi-Sankyo; Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Eisai; Advisory / Consultancy: Genentech; Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Genomic Health; Honoraria (self), Advisory / Consultancy, Research grant / Funding (self): Lilly; Honoraria (self), Advisory / Consultancy, Research grant / Funding (self), Travel / Accommodation / Expenses: Novartis; Honoraria (self), Advisory / Consultancy, Research grant / Funding (self), Travel / Accommodation / Expenses: Pfizer; Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Pierre Fabre; Advisory / Consultancy, Travel / Accommodation / Expenses: Puma Biotechnology; Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Roche. L. Biganzoli: Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Astrazeneca; Honoraria (self), Advisory / Consultancy, Research grant / Funding (self), Travel / Accommodation / Expenses: Celgene; Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Eisai; Advisory / Consultancy, Research grant / Funding (self), Travel / Accommodation / Expenses: Genomic Health; Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Ipsen; Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Lilly; Honoraria (self), Advisory / Consultancy, Research grant / Funding (self), Travel / Accommodation / Expenses: Novartis; Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Pfizer; Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Pierre Fabre; Honoraria (self), Advisory / Consultancy, Travel / Accommodation / Expenses: Roche. All other authors have declared no conflicts of interest.

Feature ranking for multi-target regression

In this work, we address the task of feature ranking for multi-target regression (MTR). The task of MTR concerns problems with multiple continuous dependent/target variables, where the goal is to learn a model for predicting all of them simultaneously. This task is receiving an increasing attention from the research community, but performing feature ranking in the context of MTR has not been studied thus far. Here, we study two groups of feature ranking scores for MTR: scores (Symbolic, Genie3 and Random Forest score) based on ensembles (bagging, random forests, extra trees) of predictive clustering trees, and a score derived as an extension of the RReliefF method. We also propose a generic data-transformation approach to MTR feature ranking and thus have two versions of each score. For both groups of feature ranking scores, we analyze their theoretical computational complexity. For the extension of the RReliefF method, we additionally derive some theoretical properties of the scores. Next, we extensively evaluate the scores on 24 benchmark MTR datasets, in terms of the quality of the ranking and the computational complexity of producing it. The results identify the parameters that influence the quality of the rankings, reveal that both groups of methods produce relevant feature rankings, and show that the Symbolic and Genie3 score, coupled with random forest ensembles, yield the best rankings.

Efficient computational model for identification of antitubercular peptides by integrating amino acid patterns and properties.

Tuberculosis (TB) is a leading killer caused by Mycobacteriumtuberculosis. Recently, anti-TB peptides have provided an alternative approach to combat antibiotic tolerance. We have developed an effective computational predictor, identification of antitubercular peptides (iAntiTB), by the integration of multiple feature vectors deriving from the amino acid sequences via random forest (RF) and support vector machine (SVM) classifiers. The iAntiTB combines the RF and SVM scores via linear regression to enhance the prediction accuracy. To make a robust and accurate predictor, we prepared the two datasets with different types of negative samples. The iAntiTB achieved area under the ROC curve values of 0.896 and 0.946 on the training datasets of the first and second datasets, respectively. The iAntiTB outperformed the other existing predictors.

An integrated molecular diagnostic report for heart transplant biopsies using an ensemble of diagnostic algorithms.

We previously reported a microarray-based diagnostic system for heart transplant endomyocardial biopsies (EMBs), using either 3-archetype (3AA) or 4-archetype (4AA) unsupervised algorithms to estimate rejection. In the present study we examined the stability of machine-learning algorithms in new biopsies, compared 3AA vs 4AA algorithms, assessed supervised binary classifiers trained on histologic or molecular diagnoses, created a report combining many scores into an ensemble of estimates, and examined possible automated sign-outs. We studied 889 EMBs from 454 transplant recipients at 8 centers: the initial cohort (N = 331) and a new cohort (N = 558). Published 3AA algorithms derived in Cohort 331 were tested in Cohort 558, the 3AA and 4AA models were compared, and supervised binary classifiers were created. A`lgorithms derived in Cohort 331 performed similarly in new biopsies despite differences in case mix. In the combined cohort, the 4AA model, including a parenchymal injury score, retained correlations with histologic rejection and DSA similar to the 3AA model. Supervised molecular classifiers predicted molecular rejection (areas under the curve [AUCs] >0.87) better than histologic rejection (AUCs <0.78), even when trained on histology diagnoses. A report incorporating many AA and binary classifier scores interpreted by 1 expert showed highly significant agreement with histology (p < 0.001), but with many discrepancies, as expected from the known noise in histology. An automated random forest score closely predicted expert diagnoses, confirming potential for automated signouts. Molecular algorithms are stable in new populations and can be assembled into an ensemble that combines many supervised and unsupervised estimates of the molecular disease states.

NMR-based metabolomics identifies patients at high risk of death within two years after acute myocardial infarction in the AMI-Florence II cohort

BackgroundRisk stratification and management of acute myocardial infarction patients continue to be challenging despite considerable efforts made in the last decades by many clinicians and researchers. The aim of this study was to investigate the metabolomic fingerprint of acute myocardial infarction using nuclear magnetic resonance spectroscopy on patient serum samples and to evaluate the possible role of metabolomics in the prognostic stratification of acute myocardial infarction patients.MethodsIn total, 978 acute myocardial infarction patients were enrolled in this study; of these, 146 died and 832 survived during 2 years of follow-up after the acute myocardial infarction. Serum samples were analyzed via high-resolution 1H-nuclear magnetic resonance spectroscopy and the spectra were used to characterize the metabolic fingerprint of patients. Multivariate statistics were used to create a prognostic model for the prediction of death within 2 years after the cardiovascular event.ResultsIn the training set, metabolomics showed significant differential clustering of the two outcomes cohorts. A prognostic risk model predicted death with 76.9% sensitivity, 79.5% specificity, and 78.2% accuracy, and an area under the receiver operating characteristics curve of 0.859. These results were reproduced in the validation set, obtaining 72.6% sensitivity, 72.6% specificity, and 72.6% accuracy. Cox models were used to compare the known prognostic factors (for example, Global Registry of Acute Coronary Events score, age, sex, Killip class) with the metabolomic random forest risk score. In the univariate analysis, many prognostic factors were statistically associated with the outcomes; among them, the random forest score calculated from the nuclear magnetic resonance data showed a statistically relevant hazard ratio of 6.45 (p = 2.16×10−16). Moreover, in the multivariate regression only age, dyslipidemia, previous cerebrovascular disease, Killip class, and random forest score remained statistically significant, demonstrating their independence from the other variables.ConclusionsFor the first time, metabolomic profiling technologies were used to discriminate between patients with different outcomes after an acute myocardial infarction. These technologies seem to be a valid and accurate addition to standard stratification based on clinical and biohumoral parameters.

Predicting peptide presentation by major histocompatibility complex class I: an improved machine learning approach to the immunopeptidome

BackgroundTo further our understanding of immunopeptidomics, improved tools are needed to identify peptides presented by major histocompatibility complex class I (MHC-I). Many existing tools are limited by their reliance upon chemical affinity data, which is less biologically relevant than sampling by mass spectrometry, and other tools are limited by incomplete exploration of machine learning approaches. Herein, we assemble publicly available data describing human peptides discovered by sampling the MHC-I immunopeptidome with mass spectrometry and use this database to train random forest classifiers (ForestMHC) to predict presentation by MHC-I.ResultsAs measured by precision in the top 1% of predictions, our method outperforms NetMHC and NetMHCpan on test sets, and it outperforms both these methods and MixMHCpred on new data from an ovarian carcinoma cell line. We also find that random forest scores correlate monotonically, but not linearly, with known chemical binding affinities, and an information-based analysis of classifier features shows the importance of anchor positions for our classification. The random-forest approach also outperforms a deep neural network and a convolutional neural network trained on identical data. Finally, we use our large database to confirm that gene expression partially determines peptide presentation.ConclusionsForestMHC is a promising method to identify peptides bound by MHC-I. We have demonstrated the utility of random forest-based approaches in predicting peptide presentation by MHC-I, assembled the largest known database of MS binding data, and mined this database to show the effect of gene expression on peptide presentation. ForestMHC has potential applicability to basic immunology, rational vaccine design, and neoantigen binding prediction for cancer immunotherapy. This method is publicly available for applications and further validation.

GPSuc: Global Prediction of Generic and Species-specific Succinylation Sites by aggregating multiple sequence features.

Lysine succinylation is one of the dominant post-translational modification of the protein that contributes to many biological processes including cell cycle, growth and signal transduction pathways. Identification of succinylation sites is an important step for understanding the function of proteins. The complicated sequence patterns of protein succinylation revealed by proteomic studies highlight the necessity of developing effective species-specific in silico strategies for global prediction succinylation sites. Here we have developed the generic and nine species-specific succinylation site classifiers through aggregating multiple complementary features. We optimized the consecutive features using the Wilcoxon-rank feature selection scheme. The final feature vectors were trained by a random forest (RF) classifier. With an integration of RF scores via logistic regression, the resulting predictor termed GPSuc achieved better performance than other existing generic and species-specific succinylation site predictors. To reveal the mechanism of succinylation and assist hypothesis-driven experimental design, our predictor serves as a valuable resource. To provide a promising performance in large-scale datasets, a web application was developed at http://kurata14.bio.kyutech.ac.jp/GPSuc/.

Causal inference for multi-level treatments with machine-learned propensity scores

Propensity score-based methods have been widely developed to adjust for confounders in observational studies to estimate causal treatment effect for binary treatments. We generalize these causal inference methods to the multi-level treatment case. We review the generalized causal inference framework and several propensity score estimation methods. We conduct a comprehensive simulation study to evaluate the performance of multinomial logistic regression, generalized boosted models, random forest and data adaptive matching score for estimating propensity scores based on inverse probability of treatment weighting. From our findings, multinomial logistic regression is susceptible to yielding extreme weights while a mis-specified model is assumed, which results in poor performance of the inverse probability weighted estimator. On the other hand, machine-learned propensity scores tend to have less biased and more stable performance, and the data adaptive matching score tends to perform the best overall. The above-mentioned propensity score based methods are applied to the Taobao dataset to evaluate the causal effect of reputation on sales.

Prediction Error Reduction Function as a Variable Importance Score

This paper introduces and develops a novel variable importance score function in the context of ensemble learning, and demonstrates its appeal empirically. Our proposed score function is simple and more straightforward than its counterpart proposed in the context of random forest, and by avoiding permutations, it is by design computationally more efficient than the random forest variable importance function. Just like the random forest variable importance function, our score handles both regression and classification seamlessly. One of the distinct advantage of our proposed score is the fact that it offers a natural cut off at zero, with all the positive scores indicating importance and significance, while the negative scores are deemed indications of insignificance. An extra advantage of our proposed score lies in the fact it works very well beyond ensemble of trees and can seamlessly be used with any base learners in the random subspace learning context. Our examples, both simulated and real, demonstrate that our proposed score does compete mostly favorably with the random forest score.

Construction of an Indonesian herbal constituents database and its use in Random Forest modelling in a search for inhibitors of aldose reductase

Data on phytochemical constituents of plants commonly used in traditional Indonesian medicine have been compiled as a computer database. This database (the Indonesian Herbal constituents database, IHD) currently contains details on ∼1,000 compounds found in 33 different plants. For each entry, the IHD gives details of chemical structure, trivial and systematic name, CAS registry number, pharmacology (where known), toxicology (LD50), botanical species, the part(s) of the plant(s) where the compounds are found, typical dosage(s) and reference(s). A second database has been also been compiled for plant-derived compounds with known activity against the enzyme, aldose reductase (AR). This database (the aldose reductase inhibitors database, ARID) contains the same details as the IHD, and currently comprises information on 120 different AR inhibitors. Virtual screening of all compounds in the IHD has been performed using Random Forest (RF) modelling, in a search for novel leads active against AR—to provide for new forms of symptomatic relief in diabetic patients. For the RF modelling, a set of simple 2D chemical descriptors were employed to classify all compounds in the combined ARID and IHD databases as either active or inactive as AR inhibitors. The resulting RF models (which gave misclassification rates of 21%) were used to identify putative new AR inhibitors in the IHD, with such compounds being identified as those giving RF scores >0.5 (in each of the three different RF models developed). In vitro assays were subsequently performed for four of the compounds obtained as hits in this in silico screening, to determine their inhibitory activity against human recombinant AR. The two compounds having the highest RF scores (prunetin and ononin) were shown to have the highest activities experimentally (giving ∼58% and ∼52% inhibition at a concentration of 15μM, respectively), while the compounds with lowest RF scores (vanillic acid and cinnamic acid) showed the lowest activities experimentally (giving ∼29% and ∼44% inhibition at a concentration of 15μM, respectively). These simple virtual screening studies were thus helpful in identifying novel inhibitors of AR, but yielded compounds with only very modest (micromolar) potency.