Weakly supervised learning of information structure of scientific abstracts--is it accurate enough to benefit real-world tasks in biomedicine?

MOTIVATION: Many practical tasks in biomedicine require accessing specific types of information in scientific literature; e.g. information about the methods, results or conclusions of the study in question. Several approaches have been developed to identify such information in scientific journal articles. The best of these have yielded promising results and proved useful for biomedical text mining tasks. However, relying on fully supervised machine learning (ml) and a large body of annotated data, existing approaches are expensive to develop and port to different tasks. A potential solution to this problem is to employ weakly supervised learning instead. In this article, we investigate a weakly supervised approach to identifying information structure according to a scheme called Argumentative Zoning (az). We apply four weakly supervised classifiers to biomedical abstracts and evaluate their performance both directly and in a real-life scenario in the context of cancer risk assessment. RESULTS: Our best weakly supervised classifier (based on the combination of active learning and self-training) performs well on the task, outperforming our best supervised classifier: it yields a high accuracy of 81% when just 10% of the labeled data is used for training. When cancer risk assessors are presented with the resulting annotated abstracts, they find relevant information in them significantly faster than when presented with unannotated abstracts. These results suggest that weakly supervised learning could be used to improve the practical usefulness of information structure for real-life tasks in biomedicine.     

Semi-supervised analysis of gene expression profiles for lineage-specific development in the Caenorhabditis elegans embryo

MOTIVATION: Gene expression profiling is a powerful approach to identify genes that may be involved in a specific biological process on a global scale. For example, gene expression profiling of mutant animals that lack or contain an excess of certain cell types is a common way to identify genes that are important for the development and maintenance of given cell types. However, it is difficult for traditional computational methods, including unsupervised and supervised learning methods, to detect relevant genes from a large collection of expression profiles with high sensitivity and specificity. Unsupervised methods group similar gene expressions together while ignoring important prior biological knowledge. Supervised methods utilize training data from prior biological knowledge to classify gene expression. However, for many biological problems, little prior knowledge is available, which limits the prediction performance of most supervised methods. RESULTS: We present a Bayesian semi-supervised learning method, called BGEN, that improves upon supervised and unsupervised methods by both capturing relevant expression profiles and using prior biological knowledge from literature and experimental validation. Unlike currently available semi-supervised learning methods, this new method trains a kernel classifier based on labeled and unlabeled gene expression examples. The semi-supervised trained classifier can then be used to efficiently classify the remaining genes in the dataset. Moreover, we model the confidence of microarray probes and probabilistically combine multiple probe predictions into gene predictions. We apply BGEN to identify genes involved in the development of a specific cell lineage in the C. elegans embryo, and to further identify the tissues in which these genes are enriched. Compared to K-means clustering and SVM classification, BGEN achieves higher sensitivity and specificity. We confirm certain predictions by biological experiments. AVAILABILITY: The results are available at http://www.csail.mit.edu/~alanqi/projects/BGEN.html.     

Monte Carlo feature selection for supervised classification

MOTIVATION: Pre-selection of informative features for supervised classification is a crucial, albeit delicate, task. It is desirable that feature selection provides the features that contribute most to the classification task per se and which should therefore be used by any classifier later used to produce classification rules. In this article, a conceptually simple but computer-intensive approach to this task is proposed. The reliability of the approach rests on multiple construction of a tree classifier for many training sets randomly chosen from the original sample set, where samples in each training set consist of only a fraction of all of the observed features. RESULTS: The resulting ranking of features may then be used to advantage for classification via a classifier of any type. The approach was validated using Golub et al. leukemia data and the Alizadeh et al. lymphoma data. Not surprisingly, we obtained a significantly different list of genes. Biological interpretation of the genes selected by our method showed that several of them are involved in precursors to different types of leukemia and lymphoma rather than being genes that are common to several forms of cancers, which is the case for the other methods. AVAILABILITY: Prototype available upon request.     

Automated Identification of Subcellular Organelles by Coherent Anti-Stokes Raman Scattering

Coherent anti-Stokes Raman scattering (CARS) is an emerging tool for label-free characterization of living cells. Here, unsupervised multivariate analysis of CARS datasets was used to visualize the subcellular compartments. In addition, a supervised learning algorithm based on the “random forest” ensemble learning method as a classifier, was trained with CARS spectra using immunofluorescence images as a reference. The supervised classifier was then used, to our knowledge for the first time, to automatically identify lipid droplets, nucleus, nucleoli, and endoplasmic reticulum in datasets that are not used for training. These four subcellular components were simultaneously and label-free monitored instead of using several fluorescent labels. These results open new avenues for label-free time-resolved investigation of subcellular components in different cells, especially cancer cells.     

Automated Identification of Subcellular Organelles by Coherent Anti-Stokes Raman Scattering

Coherent anti-Stokes Raman scattering (CARS) is an emerging tool for label-free characterization of living cells. Here, unsupervised multivariate analysis of CARS datasets was used to visualize the subcellular compartments. In addition, a supervised learning algorithm based on the “random forest” ensemble learning method as a classifier, was trained with CARS spectra using immunofluorescence images as a reference. The supervised classifier was then used, to our knowledge for the first time, to automatically identify lipid droplets, nucleus, nucleoli, and endoplasmic reticulum in datasets that are not used for training. These four subcellular components were simultaneously and label-free monitored instead of using several fluorescent labels. These results open new avenues for label-free time-resolved investigation of subcellular components in different cells, especially cancer cells.     

Classification of lactic acid bacteria with UV-resonance Raman spectroscopy

UV-resonance Raman spectroscopy is applied as a method for the identification of lactic acid bacteria from yogurt. Eight different strains of bacteria from Lactobacillus acidophilus, L. delbrueckii ssp. bulgaricus, and Streptococcus thermophilus were investigated. At an excitation wavelength of 244 nm signals from nucleic acids and proteins are selectively enhanced. Classification was accomplished using different chemometric methods. In a first attempt, the unsupervised methods hierarchical cluster analysis and principal component analysis were applied to investigate natural grouping in the data. In a second step the spectra were analyzed using several supervised methods: K-nearest neighbor classifier, nearest mean classifier, linear discriminant analysis, and support vector machines.     

Feature selection based on empirical-risk function to detect lesions in vascular computed tomography

ObjectiveThe overall goal of the study is to detect coronary artery lesions regardless their nature, calcified or hypo-dense. To avoid explicit modelling of heterogeneous lesions, we adopted an approach based on machine learning and using unsupervised or semi-supervised classifiers. The success of the classifiers based on machine learning strongly depends on the appropriate choice of features differentiating between lesions and regular appearance. The specific goal of this article is to propose a novel strategy devised to select the best feature set for the classifiers used, out of a given set of candidate features.Materials and methodsThe features are calculated in image planes orthogonal to the artery centerline, and the classifier assigns to each of these cross-sections a label “healthy” or “diseased”. The contribution of this article is a feature-selection strategy based on the empirical risk function that is used as a criterion in the initial feature ranking and in the selection process itself. We have assessed this strategy in association with two classifiers based on the density-level detection approach that seeks outliers from the distribution corresponding to the regular appearance. The method was evaluated using a total of 13,687 cross-sections extracted from 53 coronary arteries in 15 patients.ResultsUsing the feature subset selected by the risk-based strategy, balanced error rates achieved by the unsupervised and semi-supervised classifiers respectively were equal to 13.5% and 15.4%. These results were substantially better than the rates achieved using feature subsets selected by supervised strategies. The unsupervised and semi-supervised methods also outperformed supervised classifiers using feature subsets selected by the corresponding supervised strategies.DiscussionSupervised methods require large data sets annotated by experts, both to select the features and to train the classifiers, and collecting these annotations is time-consuming. With these methods, lesions whose appearance differs from the training data may remain undetected. Lesion-detection problem is highly imbalanced, since healthy cross-sections usually are much more numerous than the diseased ones. Training the classifiers based on the density-level detection approach needs a small number of annotations or no annotations at all. The same annotations are sufficient to compute the empirical risk and to perform the selection. Therefore, our strategy associated with an unsupervised or semi-supervised classifier requires a considerably smaller number of annotations as compared to conventional supervised selection strategies. The approach proposed is also better suited for highly imbalanced problems and can detect lesions differing from the training set.ConclusionThe risk-based selection strategy, associated with classifiers using the density-level detection approach, outperformed other strategies and classifiers when used to detect coronary artery lesions. It is well suited for highly imbalanced problems, where the lesions are represented as low-density regions of the feature space, and it can be used in other anomaly detection problems interpretable as a binary classification problem where the empirical risk can be calculated.     

Learning gene regulatory networks from only positive and unlabeled data

Background  

Recently, supervised learning methods have been exploited to reconstruct gene regulatory networks from gene expression data. The reconstruction of a network is modeled as a binary classification problem for each pair of genes. A statistical classifier is trained to recognize the relationships between the activation profiles of gene pairs. This approach has been proven to outperform previous unsupervised methods. However, the supervised approach raises open questions. In particular, although known regulatory connections can safely be assumed to be positive training examples, obtaining negative examples is not straightforward, because definite knowledge is typically not available that a given pair of genes do not interact.     

Adaptive prediction of epileptic seizures from intracranial recordings

Previous work has demonstrated that some dynamic properties of intracranial EEG signals are indicative of epileptic seizures and hence could be used for prediction in order to realize counter measures. However, most previous studies only investigated predictability via offline analysis of EEG signals as compared to actually predicting seizures in a setting applicable to implantable devices. Here we address this problem, which calls for simple and fast online methods, and based on previous offline analyses we hypothesize that prediction can be further improved when using multiple features to detect the preictal patterns. We propose a simple adaptive online method (an evolving neuro-fuzzy model) to adaptively learn such combined features. The classifier starts out with a simple structure and patient-independent parameters and then grows into a personal seizure predictor as recursive methods tune the model structure and parameters. We apply the adaptive classifier to a publicly available database of intracranial recordings from 21 patients and demonstrate that seizure prediction is improved with our online method as compared to offline non-adaptive techniques. We show that our method is robust with respect to those few model parameters, which are not adapted. Moreover, as we report the performance on data from a publicly available seizure database, our results can serve as a yardstick for future method developments.     

Gene selection for classification of cancers using probabilistic model building genetic algorithm

Recently, DNA microarray-based gene expression profiles have been used to correlate the clinical behavior of cancers with the differential gene expression levels in cancerous and normal tissues. To this end, after selection of some predictive genes based on signal-to-noise (S2N) ratio, unsupervised learning like clustering and supervised learning like k-nearest neighbor (k NN) classifier are widely used. Instead of S2N ratio, adaptive searches like Probabilistic Model Building Genetic Algorithm (PMBGA) can be applied for selection of a smaller size gene subset that would classify patient samples more accurately. In this paper, we propose a new PMBGA-based method for identification of informative genes from microarray data. By applying our proposed method to classification of three microarray data sets of binary and multi-type tumors, we demonstrate that the gene subsets selected with our technique yield better classification accuracy.     

Feature Selection and Classification of MAQC-II Breast Cancer and Multiple Myeloma Microarray Gene Expression Data

Microarray data has a high dimension of variables but available datasets usually have only a small number of samples, thereby making the study of such datasets interesting and challenging. In the task of analyzing microarray data for the purpose of, e.g., predicting gene-disease association, feature selection is very important because it provides a way to handle the high dimensionality by exploiting information redundancy induced by associations among genetic markers. Judicious feature selection in microarray data analysis can result in significant reduction of cost while maintaining or improving the classification or prediction accuracy of learning machines that are employed to sort out the datasets. In this paper, we propose a gene selection method called Recursive Feature Addition (RFA), which combines supervised learning and statistical similarity measures. We compare our method with the following gene selection methods:

• Support Vector Machine Recursive Feature Elimination (SVMRFE)
• Leave-One-Out Calculation Sequential Forward Selection (LOOCSFS)
• Gradient based Leave-one-out Gene Selection (GLGS)

To evaluate the performance of these gene selection methods, we employ several popular learning classifiers on the MicroArray Quality Control phase II on predictive modeling (MAQC-II) breast cancer dataset and the MAQC-II multiple myeloma dataset. Experimental results show that gene selection is strictly paired with learning classifier. Overall, our approach outperforms other compared methods. The biological functional analysis based on the MAQC-II breast cancer dataset convinced us to apply our method for phenotype prediction. Additionally, learning classifiers also play important roles in the classification of microarray data and our experimental results indicate that the Nearest Mean Scale Classifier (NMSC) is a good choice due to its prediction reliability and its stability across the three performance measurements: Testing accuracy, MCC values, and AUC errors.     

Penalized feature selection and classification in bioinformatics

In bioinformatics studies, supervised classification with high-dimensional input variables is frequently encountered. Examples routinely arise in genomic, epigenetic and proteomic studies. Feature selection can be employed along with classifier construction to avoid over-fitting, to generate more reliable classifier and to provide more insights into the underlying causal relationships. In this article, we provide a review of several recently developed penalized feature selection and classification techniques--which belong to the family of embedded feature selection methods--for bioinformatics studies with high-dimensional input. Classification objective functions, penalty functions and computational algorithms are discussed. Our goal is to make interested researchers aware of these feature selection and classification methods that are applicable to high-dimensional bioinformatics data.     

Automated Glioblastoma Segmentation Based on a Multiparametric Structured Unsupervised Classification

Automatic brain tumour segmentation has become a key component for the future of brain tumour treatment. Currently, most of brain tumour segmentation approaches arise from the supervised learning standpoint, which requires a labelled training dataset from which to infer the models of the classes. The performance of these models is directly determined by the size and quality of the training corpus, whose retrieval becomes a tedious and time-consuming task. On the other hand, unsupervised approaches avoid these limitations but often do not reach comparable results than the supervised methods. In this sense, we propose an automated unsupervised method for brain tumour segmentation based on anatomical Magnetic Resonance (MR) images. Four unsupervised classification algorithms, grouped by their structured or non-structured condition, were evaluated within our pipeline. Considering the non-structured algorithms, we evaluated K-means, Fuzzy K-means and Gaussian Mixture Model (GMM), whereas as structured classification algorithms we evaluated Gaussian Hidden Markov Random Field (GHMRF). An automated postprocess based on a statistical approach supported by tissue probability maps is proposed to automatically identify the tumour classes after the segmentations. We evaluated our brain tumour segmentation method with the public BRAin Tumor Segmentation (BRATS) 2013 Test and Leaderboard datasets. Our approach based on the GMM model improves the results obtained by most of the supervised methods evaluated with the Leaderboard set and reaches the second position in the ranking. Our variant based on the GHMRF achieves the first position in the Test ranking of the unsupervised approaches and the seventh position in the general Test ranking, which confirms the method as a viable alternative for brain tumour segmentation.     

Argumentation Based Joint Learning: A Novel Ensemble Learning Approach

Recently, ensemble learning methods have been widely used to improve classification performance in machine learning. In this paper, we present a novel ensemble learning method: argumentation based multi-agent joint learning (AMAJL), which integrates ideas from multi-agent argumentation, ensemble learning, and association rule mining. In AMAJL, argumentation technology is introduced as an ensemble strategy to integrate multiple base classifiers and generate a high performance ensemble classifier. We design an argumentation framework named Arena as a communication platform for knowledge integration. Through argumentation based joint learning, high quality individual knowledge can be extracted, and thus a refined global knowledge base can be generated and used independently for classification. We perform numerous experiments on multiple public datasets using AMAJL and other benchmark methods. The results demonstrate that our method can effectively extract high quality knowledge for ensemble classifier and improve the performance of classification.     

Computerized classification of Mediterranean vegetation using panchromatic aerial photographs

Abstract. Historical aerial photographs are an important source for data on medium- to long-term (10 - 50 yr) vegetation changes. Older photographs are panchromatic, and manual interpretation has traditionally been used to derive vegetation data from such photographs. We present a method for computerized analysis of panchromatic aerial photographs, which enables one to create high resolution, accurate vegetation maps. Our approach is exemplified using two aerial photographs (from 1964 and 1992) of a test area on Mt. Meron, Israel. Spatial resolution (pixel size) of the geo-rectified photos was 0.30 m and spatial accuracy (RMS error) ca. 1 m. An illumination adjustment prior to classification was found to be essential in reducing misclassification error rates. Two classification approaches were employed: a standard maximum-likelihood supervised classifier, and a modification of a supervised classification, which takes into account spectral properties of individual pixels as well as their neighbourhood characteristics. Accuracy of the maximum likelihood classification was 81 % in the 1992 image and 54 % in the 1964 image. The neighbour classifier increased accuracy to 89 % and 82 % respectively. The overall results suggest that computerized analysis of sequences of panchromatic aerial photographs may serve as a valuable tool for the quantification of medium-term vegetation changes.     

Predicting peptides that bind to MHC molecules using supervised learning of hidden markov models

The binding of a major histocompatibility complex (MHC) molecule to a peptide originating in an antigen is essential to recognizing antigens in immune systems, and it has proved to be important to use computers to predict the peptides that will bind to an MHC molecule. The purpose of this paper is twofold: First, we propose to apply supervised learning of hidden Markov models (HMMs) to this problem, which can surpass existing methods for the problem of predicting MHC-binding peptides. Second, we generate peptides that have high probabilities to bind to a certain MHC molecule, based on our proposed method using peptides binding to MHC molecules as a set of training data. From our experiments, in a type of cross-validation test, the discrimination accuracy of our supervised learning method is usually approximately 2–15% better than those of other methods, including backpropagation neural networks, which have been regarded as the most effective approach to this problem. Furthermore, using an HMM trained for HLA-A2, we present new peptide sequences that are provided with high binding probabilities by the HMM and that are thus expected to bind to HLA-A2 proteins. Peptide sequences not shown in this paper but with rather high binding probabilities can be obtained from the author (E-mail: mami@ccm.cl.nec.co.jp). Proteins 33:460–474, 1998. © 1998 Wiley-Liss, Inc.     

Microarray-based cancer diagnosis with artificial neural networks

In recent years, the advent of experimental methods to probe gene expression profiles of cancer on a genome-wide scale has led to widespread use of supervised machine learning algorithms to characterize these profiles. The main applications of these analysis methods range from assigning functional classes of previously uncharacterized genes to classification and prediction of different cancer tissues. This article surveys the application of machine learning algorithms to classification and diagnosis of cancer based on expression profiles. To exemplify the important issues of the classification procedure, the emphasis of this article is on one such method, namely artificial neural networks. In addition, methods to extract genes that are important for the performance of a classifier, as well as the influence of sample selection on prediction results are discussed.     

PACK: Profile Analysis using Clustering and Kurtosis to find molecular classifiers in cancer

MOTIVATION: Elucidating the molecular taxonomy of cancers and finding biological and clinical markers from microarray experiments is problematic due to the large number of variables being measured. Feature selection methods that can identify relevant classifiers or that can remove likely false positives prior to supervised analysis are therefore desirable. RESULTS: We present a novel feature selection procedure based on a mixture model and a non-gaussianity measure of a gene's expression profile. The method can be used to find genes that define either small outlier subgroups or major subdivisions, depending on the sign of kurtosis. The method can also be used as a filtering step, prior to supervised analysis, in order to reduce the false discovery rate. We validate our methodology using six independent datasets by rediscovering major classifiers in ER negative and ER positive breast cancer and in prostate cancer. Furthermore, our method finds two novel subtypes within the basal subgroup of ER negative breast tumours, associated with apoptotic and immune response functions respectively, and with statistically different clinical outcome. AVAILABILITY: An R-function pack that implements the methods used here has been added to vabayelMix, available from (www.cran.r-project.org). CONTACT: aet21@cam.ac.uk SUPPLEMENTARY INFORMATION: Supplementary information is available at Bioinformatics online.     

Multi-class cancer subtype classification based on gene expression signatures with reliability analysis

Differential diagnosis among a group of histologically similar cancers poses a challenging problem in clinical medicine. Constructing a classifier based on gene expression signatures comprising multiple discriminatory molecular markers derived from microarray data analysis is an emerging trend for cancer diagnosis. To identify the best genes for classification using a small number of samples relative to the genome size remains the bottleneck of this approach, despite its promise. We have devised a new method of gene selection with reliability analysis, and demonstrated that this method can identify a more compact set of genes than other methods for constructing a classifier with optimum predictive performance for both small round blue cell tumors and leukemia. High consensus between our result and the results produced by methods based on artificial neural networks and statistical techniques confers additional evidence of the validity of our method. This study suggests a way for implementing a reliable molecular cancer classifier based on gene expression signatures.