High-dimensional Biological Datasets Research Articles

Leveraging permutation testing to assess confidence in positive-unlabeled learning applied to high-dimensional biological datasets

BackgroundCompared to traditional supervised machine learning approaches employing fully labeled samples, positive-unlabeled (PU) learning techniques aim to classify “unlabeled” samples based on a smaller proportion of known positive examples. This more challenging modeling goal reflects many real-world scenarios in which negative examples are not available—posing direct challenges to defining prediction accuracy and robustness. While several studies have evaluated predictions learned from only definitive positive examples, few have investigated whether correct classification of a high proportion of known positives (KP) samples from among unlabeled samples can act as a surrogate to indicate model quality.ResultsIn this study, we report a novel methodology combining multiple established PU learning-based strategies with permutation testing to evaluate the potential of KP samples to accurately classify unlabeled samples without using “ground truth” positive and negative labels for validation. Multivariate synthetic and real-world high-dimensional benchmark datasets were employed to demonstrate the suitability of the proposed pipeline to provide evidence of model robustness across varied underlying ground truth class label compositions among the unlabeled set and with different proportions of KP examples. Comparisons between model performance with actual and permuted labels could be used to distinguish reliable from unreliable models.ConclusionsAs in fully supervised machine learning, permutation testing offers a means to set a baseline “no-information rate” benchmark in the context of semi-supervised PU learning inference tasks—providing a standard against which model performance can be compared.

Gene selection for high dimensional biological datasets using hybrid island binary artificial bee colony with chaos game optimization

Microarray technology, as applied to the fields of bioinformatics, biotechnology, and bioengineering, has made remarkable progress in both the treatment and prediction of many biological problems. However, this technology presents a critical challenge due to the size of the numerous genes present in the high-dimensional biological datasets associated with an experiment, which leads to a curse of dimensionality on biological data. Such high dimensionality of real biological data sets not only increases memory requirements and training costs, but also reduces the ability of learning algorithms to generalise. Consequently, multiple feature selection (FS) methods have been proposed by researchers to choose the most significant and precise subset of classified genes from gene expression datasets while maintaining high classification accuracy. In this research work, a novel binary method called iBABC-CGO based on the island model of the artificial bee colony algorithm, combined with the chaos game optimization algorithm and SVM classifier, is suggested for FS problems using gene expression data. Due to the binary nature of FS problems, two distinct transfer functions are employed for converting the continuous search space into a binary one, thus improving the efficiency of the exploration and exploitation phases. The suggested strategy is tested on a variety of biological datasets with different scales and compared to popular metaheuristic-based, filter-based, and hybrid FS methods. Experimental results supplemented with the statistical measures, box plots, Wilcoxon tests, Friedman tests, and radar plots demonstrate that compared to prior methods, the proposed iBABC-CGO exhibit competitive performance in terms of classification accuracy, selection of the most relevant subset of genes, data variability, and convergence rate. The suggested method is also proven to identify unique sets of informative, relevant genes successfully with the highest overall average accuracy in 15 tested biological datasets. Additionally, the biological interpretations of the selected genes by the proposed method are also provided in our research work.

Variable selection for nonlinear dimensionality reduction of biological datasets through bootstrapping of correlation networks

Identifying the most relevant variables or features in massive datasets for dimensionality reduction can lead to improved and more informative display, faster computation times, and more explainable models of complex systems. Despite significant advances and available algorithms, this task generally remains challenging, especially in unsupervised settings. In this work, we propose a method that constructs correlation networks using all intervening variables and then selects the most informative ones based on network bootstrapping. The method can be applied in both supervised and unsupervised scenarios. We demonstrate its functionality by applying Uniform Manifold Approximation and Projection for dimensionality reduction to several high-dimensional biological datasets, derived from 4D live imaging recordings of hundreds of morpho-kinetic variables, describing the dynamics of thousands of individual leukocytes at sites of prominent inflammation. We compare our method with other standard ones in the field, such as Principal Component Analysis and Elastic Net, showing that it outperforms them. The proposed method can be employed in a wide range of applications, encompassing data analysis and machine learning.

Efficient high-dimension feature selection based on enhanced equilibrium optimizer

Feature selection (FS) is an important task in any classification process and aims to choose the smallest features number that yields higher classification accuracy. FS can be formulated as a combinatorial NP-hard problem for which robust metaheuristics are used as efficient wrapper-based FS approaches. However, when applied for high dimensional datasets that present large features number and few samples, the effectiveness of such wrapper-metaheuristics degraded, and their computation costs increased. To tackle this problem, we propose in this paper a hybrid FS approach based on the ReliefF filter method and a novel metaheuristic Equilibrium Optimizer (EO). The proposed method, called RBEO-LS, is composed of two phases. In the first phase, the ReliefF algorithm is used as a preprocessing step to assign weights for features, which estimate their relevance to the classification task. In the second phase, the binary EO (BEO) is used as a wrapper search approach. The features are ranked according to their weights and are used for the initialization of the BEO population. We embedded the BEO with a local search strategy to improve its performance by adding relevant features and removing redundant ones from the features subset guided by the features ranking and the Pearson coefficient correlation. The performance of the developed algorithm has been evaluated on sixteen UCI datasets and ten high dimensional biological datasets. The UCI datasets contain a high number of samples and a small or medium number of features. The biological datasets present a high number of features with few samples. The results demonstrate that the use of the ReliefF algorithm and the local search strategy improves the performance of the EO algorithm. The results also show the superiority of the RBEO-LS, among other state-of-the-art approaches.

Stochastic Mutual Information Gradient Estimation for Dimensionality Reduction Networks.

Feature ranking and selection is a widely used approach in various applications of supervised dimensionality reduction in discriminative machine learning. Nevertheless there exists significant evidence on feature ranking and selection algorithms based on any criterion leading to potentially sub-optimal solutions for class separability. In that regard, we introduce emerging information theoretic feature transformation protocols as an end-to-end neural network training approach. We present a dimensionality reduction network (MMINet) training procedure based on the stochastic estimate of the mutual information gradient. The network projects high-dimensional features onto an output feature space where lower dimensional representations of features carry maximum mutual information with their associated class labels. Furthermore, we formulate the training objective to be estimated non-parametrically with no distributional assumptions. We experimentally evaluate our method with applications to high-dimensional biological data sets, and relate it to conventional feature selection algorithms to form a special case of our approach.

Distributed load balancing frequent colossal closed itemset mining algorithm for high dimensional dataset

The focus of extracting colossal closed itemsets from high dimensional biological datasets has been great in recent times. A massive set of short and average sized mined itemsets do not confine complete and valuable information for decision making. But, the traditional itemset mining algorithms expend a gigantic measure of time in mining a massive set of short and average sized itemsets. The greater interest of research in the field of bioinformatics and the abundant data across the variety of domains paved the way for the generation of the high dimensional dataset. These datasets are depicted by an extensive number of features and a smaller number of rows. Colossal closed itemsets are very significant for numerous applications including the field of bioinformatics and are influential during the decision making. Extracting a huge amount of information and knowledge from the high dimensional dataset is a nontrivial task. The existing colossal closed itemsets mining algorithms for the high dimensional dataset are sequential and computationally expensive. Distributed and parallel computing is a good strategy to overcome the inefficiency of the existing sequential algorithm. Balanced Distributed Parallel Frequent Colossal Closed Itemset Mining (BDPFCCIM) algorithm is designed for high dimensional datasets. An efficient closeness checking method to check the closeness of the rowset and an efficient pruning strategy to snip the row enumeration mining search space is enclosed with the proposed BDPFCCIM algorithm. The proposed BDPFCCIM algorithm is the first distributed load balancing algorithm to mine frequent colossal closed itemsets from high dimensional biological datasets. The experimental results demonstrate the efficient performance of the proposed BDPFCCIM algorithm in comparison with the state-of-the-art algorithms.

PyBDA: a command line tool for automated analysis of big biological data sets

BackgroundAnalysing large and high-dimensional biological data sets poses significant computational difficulties for bioinformaticians due to lack of accessible tools that scale to hundreds of millions of data points.ResultsWe developed a novel machine learning command line tool called PyBDA for automated, distributed analysis of big biological data sets. By using Apache Spark in the backend, PyBDA scales to data sets beyond the size of current applications. It uses Snakemake in order to automatically schedule jobs to a high-performance computing cluster. We demonstrate the utility of the software by analyzing image-based RNA interference data of 150 million single cells.ConclusionPyBDA allows automated, easy-to-use data analysis using common statistical methods and machine learning algorithms. It can be used with simple command line calls entirely making it accessible to a broad user base. PyBDA is available at https://pybda.rtfd.io.

Penalized estimation of sparse concentration matrices based on prior knowledge with applications to placenta elemental data

Identifying patterns of association or dependency among high-dimensional biological datasets with sparse precision matrices remains a challenge. In this paper, we introduce a weighted sparse Gaussian graphical model that can incorporate prior knowledge to infer the structure of the network of trace element concentrations, including essential elements as well as toxic metals and metaloids measured in the human placentas. We present the weighted L1 penalized regularization procedure for estimating the sparse precision matrix in the setting of Gaussian graphical models. First, we use simulation models to demonstrate that the proposed method yields a better estimate of the precision matrix than the procedures that fail to account for the prior knowledge of the network structure. Then, we apply this method to estimate sparse element concentration matrices of placental biopsies from the New Hampshire Birth Cohort Study. The chemical architecture for elements is complex; thus, the method proposed herein was applied to infer the dependency structures of the elements using prior knowledge of their biological roles.

New tools for the visualization of biological pathways.

This paper presents several geometrically motivated techniques for the visualization of high-dimensional biological data sets. The Grassmann manifold provides a robust framework for measuring data similarity in a subspace context. Sparse radial basis function classification as a visualization technique leverages recent advances in radial basis function learning via convex optimization. In the spirit of deep belief networks, supervised centroid-encoding is proposed as a way to exploit class label information. These methods are compared to linear and nonlinear principal component analysis (autoencoders) in the context of data visualization; these approaches may perform poorly for visualization when the variance of the data is spread across more than three dimensions. In contrast, the proposed methods are shown to capture significant data structure in two or three dimensions, even when the information in the data lives in higher dimensional subspaces. To illustrate these ideas, the visualization techniques are applied to gene expression data sets that capture the host immune system's response to infection by the Ebola virus in non-human primate and collaborative cross mouse.

A method for learning a sparse classifier in the presence of missing data for high-dimensional biological datasets.

This work addresses two common issues in building classification models for biological or medical studies: learning a sparse model, where only a subset of a large number of possible predictors is used, and training in the presence of missing data. This work focuses on supervised generative binary classification models, specifically linear discriminant analysis (LDA). The parameters are determined using an expectation maximization algorithm to both address missing data and introduce priors to promote sparsity. The proposed algorithm, expectation-maximization sparse discriminant analysis (EM-SDA), produces a sparse LDA model for datasets with and without missing data. EM-SDA is tested via simulations and case studies. In the simulations, EM-SDA is compared with nearest shrunken centroids (NSCs) and sparse discriminant analysis (SDA) with k-nearest neighbors for imputation for varying mechanism and amount of missing data. In three case studies using published biomedical data, the results are compared with NSC and SDA models with four different types of imputation, all of which are common approaches in the field. EM-SDA is more accurate and sparse than competing methods both with and without missing data in most of the experiments. Furthermore, the EM-SDA results are mostly consistent between the missing and full cases. Biological relevance of the resulting models, as quantified via a literature search, is also presented. A Matlab implementation published under GNU GPL v.3 license is available at http://web.mit.edu/braatzgroup/links.html . braatz@mit.edu. Supplementary data are available at Bioinformatics online.

Finding causative genes from high-dimensional data: an appraisal of statistical and machine learning approaches.

Modern biological experiments often involve high-dimensional data with thousands or more variables. A challenging problem is to identify the key variables that are related to a specific disease. Confounding this task is the vast number of statistical methods available for variable selection. For this reason, we set out to develop a framework to investigate the variable selection capability of statistical methods that are commonly applied to analyze high-dimensional biological datasets. Specifically, we designed six simulated cancers (based on benchmark colon and prostate cancer data) where we know precisely which genes cause a dataset to be classified as cancerous or normal - we call these causative genes. We found that not one statistical method tested could identify all the causative genes for all of the simulated cancers, even though increasing the sample size does improve the variable selection capabilities in most cases. Furthermore, certain statistical tools can classify our simulated data with a low error rate, yet the variables being used for classification are not necessarily the causative genes.

DATA REDUCTION TECHNIQUES FOR HIGH DIMENSIONAL BIOLOGICAL DATA

High dimensional biological datasets in recent years has been growing rapidly. Extracting the knowledge and analyzing highdimensional biological data is one the key challenges in which variety and veracity are the two distinct characteristics. The question that arises now is, how to perform dimensionality reduction for this heterogeneous data and how to develop a high performance platform to efficiently analyze high dimensional biological data and how to find the useful things from this data. To deeply discuss this issue, this paper begins with a brief introduction to data analytics available for biological data, followed by the discussions of big data analytics and then a survey on various data reduction methods for biological data. We propose a dense clustering algorithm for standard high dimensional biological data.

一种融合T-Rank和Softmax的特征提取算法研究

本文针对高维生物数据特征提出了一种融合T-Rank和Softmax的特征提取算法。该方法比传统特征提取方法在处理高维生物数据更加有效，不仅提取的特征个数较少，而且计算速度快。利用算法本文对高维银屑病基因表达谱数据进行了研究，得到了分类准确率较高的疾病诊断模型。 The paper proposed a new feature extraction algorithm by integrating T-rank and Softmax for the high dimensional biological data sets, which is more effective than traditional method when dealing with high dimensional data. It can not only extract a very few number of features, but also have fast computing speed. By using of this new algorithm, the paper obtains a high accuracy diagnosis model for psoriasis.

Independent Principal Component Analysis for biologically meaningful dimension reduction of large biological data sets

BackgroundA key question when analyzing high throughput data is whether the information provided by the measured biological entities (gene, metabolite expression for example) is related to the experimental conditions, or, rather, to some interfering signals, such as experimental bias or artefacts. Visualization tools are therefore useful to better understand the underlying structure of the data in a 'blind' (unsupervised) way. A well-established technique to do so is Principal Component Analysis (PCA). PCA is particularly powerful if the biological question is related to the highest variance. Independent Component Analysis (ICA) has been proposed as an alternative to PCA as it optimizes an independence condition to give more meaningful components. However, neither PCA nor ICA can overcome both the high dimensionality and noisy characteristics of biological data.ResultsWe propose Independent Principal Component Analysis (IPCA) that combines the advantages of both PCA and ICA. It uses ICA as a denoising process of the loading vectors produced by PCA to better highlight the important biological entities and reveal insightful patterns in the data. The result is a better clustering of the biological samples on graphical representations. In addition, a sparse version is proposed that performs an internal variable selection to identify biologically relevant features (sIPCA).ConclusionsOn simulation studies and real data sets, we showed that IPCA offers a better visualization of the data than ICA and with a smaller number of components than PCA. Furthermore, a preliminary investigation of the list of genes selected with sIPCA demonstrate that the approach is well able to highlight relevant genes in the data with respect to the biological experiment.IPCA and sIPCA are both implemented in the R package mixomics dedicated to the analysis and exploration of high dimensional biological data sets, and on mixomics' web-interface.