Search PubMed⌕ Search

SEARCH · Search PubMed

Results for “unsupervised clustering”

Search indexed PubMed citations on genomics, clinical trials, systematic reviews and public health. Explore titles, authors and supplied subject terms, then open the PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 469 records · Page 26Linked to original sources
PubMed1996

Self-organizing neural networks--an alternative way of cluster analysis in clinical chemistry.

Supervised learning schemes have been employed by several workers for training neural networks designed to solve clinical problems. We demonstrate that unsupervised techniques can also produce interesting and meaningful results. Using a data set on the chemical composition of milk from 22 different mammals, we demonstrate that self-organizing feature maps (Kohonen networks) as well as a modified version of error backpropagation technique yield results mimicking conventional cluster analysis. Both techniques are able to project a potentially multi-dimensional input vector onto a two-dimensional space whereby neighborhood relationships remain conserved. Thus, these techniques can be used for reducing dimensionality of complicated data sets and for enhancing comprehensibility of features hidden in the data matrix.

Animals↗
PubMed2004

A hierarchical mixture of Markov models for finding biologically active metabolic paths using gene expression and protein classes.

With the recent development of experimental high-throughput techniques, the type and volume of accumulating biological data have extremely increased these few years. Mining from different types of data might lead us to find new biological insights. We present a new methodology for systematically combining three different datasets to find biologically active metabolic paths/patterns. This method consists of two steps: First it synthesizes metabolic paths from a given set of chemical reactions, which are already known and whose enzymes are co-expressed, in an efficient manner. It then represents the obtained metabolic paths in a more comprehensible way through estimating parameters of a probabilistic model by using these synthesized paths. This model is built upon an assumption that an entire set of chemical reactions corresponds to a Markov state transition diagram. Furthermore, this model is a hierarchical latent variable model, containing a set of protein classes as a latent variable, for clustering input paths in terms of existing knowledge of protein classes. We tested the performance of our method using a main pathway of glycolysis, and found that our method achieved higher predictive performance for the issue of classifying gene expressions than those obtained by other unsupervised methods. We further analyzed the estimated parameters of our probabilistic models, and found that biologically active paths were clustered into only two or three patterns for each expression experiment type, and each pattern suggested some new long-range relations in the glycolysis pathway.

Computer Simulation↗
PubMed1992

Clustering proteins into families using artificial neural networks.

An artificial neural network was used to cluster proteins into families. The network, composed of 7 x 7 neurons, was trained with the Kohonen unsupervised learning algorithm using, as inputs, matrix patterns derived from the bipeptide composition of 447 proteins, belonging to 13 different families. As a result of the training, and without any a priori indication of the number or composition of the expected families, the network self-organized the activation of its neurons into topologically ordered maps in which almost all the proteins (96.7%) were correctly clustered into the corresponding families. In a second computational experiment, a similar network was trained with one family of the previous learning set (76 cytochrome c sequences). The new neural map clustered these proteins into 25 different neurons (five in the first experiment), wherein phylogenetically related sequences were positioned close to each other. This result shows that the network can adapt the clustering resolution to the complexity of the learning set, a useful feature when working with an unknown number of clusters. Although the learning stage is time consuming, once the topological map is obtained, the classification of new proteins is very fast. Altogether, our results suggest that this novel approach may be a useful tool to organize the search for homologies in large macromolecular databases.

Algorithms↗
PubMedDate not supplied

Unsupervised measurement of brain tumor volume on MR images.

We examined unsupervised methods of segmentation of MR images of the brain for measuring tumor volume in response to treatment. Two clustering methods were used: fuzzy c-means and a nonfuzzy clustering algorithm. Results were compared with volume segmentations by two supervised methods, k-nearest neighbors and region growing, and all results were compared with manual labelings. Results of individual segmentations are presented as well as comparisons on the application of the different methods with 10 data sets of patients with brain tumors. Unsupervised segmentation is preferred for measuring tumor volumes in response to treatment, as it eliminates operator dependency and may be adequate for delineation of the target volume in radiation therapy. Some obstacles need to be overcome, in particular regarding the detection of anatomically relevant tissue classes. This study shows that these improvements are possible.

Adult↗
PubMedDate not supplied

Variable selection and specification of robust QSAR models from multicollinear data: arylpiperazinyl derivatives with affinity and selectivity for alpha2-adrenoceptors.

Two QSAR models have been identified that predict the affinity and selectivity of arylpiperazinyl derivatives for alpha1 and alpha2 adrenoceptors (ARs). The models have been specified and validated using 108 compounds whose structures and inhibition constants (Ki) are available in the literature [Barbaro et al., J. Med. Chem., 44 (2001) 2118; Betti et al., J. Med. Chem., 45 (2002) 3603; Barbaro et al., Bioorg. Med. Chem., 10 (2002) 361; Betti et al., J. Med. Chem., 46 (2003) 3555]. One hundred and forty-seven predictors have been calculated using the Cerius 2 software available from Accelrys. This set of variables exhibited redundancy and severe multicollinearity, which had to be identified and removed as appropriate in order to obtain robust regression models free of inflated errors for the beta estimates - so-called bouncing betas. Those predictors that contained information relevant to the alpha2 response were identified on the basis of their pairwise linear correlations with affinity (-log Ki) for alpha2 adrenoceptors; the remaining variables were discarded. Subsequent variable selection made use of Factor Analysis (FA) and Unsupervised Variable Selection (UzFS). The data was divided into test and training sets using cluster analysis. These two sets were characterised by similar and consistent distributions of compounds in a high dimensional, but relevant predictor space. Multiple regression was then used to determine a subset of predictors from which to determine QSAR models for affinity to alpha2-ARs. Two multivariate procedures, Continuum Regression (the Portsmouth formulation) and Canonical Correlation Analysis (CCA), have been used to specify models for affinity and selectivity, respectively. Reasonable predictions were obtained using these in silico screening tools.

Models, Molecular↗
PubMed2005

Self-Organizing Map (SOM) unveils and visualizes hidden sequence characteristics of a wide range of eukaryote genomes.

Novel tools are needed for comprehensive comparisons of interspecies characteristics of massive amounts of genomic sequences currently available. An unsupervised neural network algorithm, Self-Organizing Map (SOM), is an effective tool for clustering and visualizing high-dimensional complex data on a single map. We modified the conventional SOM, on the basis of batch-learning SOM, for genome informatics making the learning process and resulting map independent of the order of data input. We generated the SOMs for tri- and tetranucleotide frequencies in 10- and 100-kb sequence fragments from 38 eukaryotes for which almost complete genome sequences are available. SOM recognized species-specific characteristics (key combinations of oligonucleotide frequencies) in the genomic sequences, permitting species-specific classification of the sequences without any information regarding the species. We also generated the SOM for tetranucleotide frequencies in 1-kb sequence fragments from the human genome and found sequences for four functional categories (5' and 3' UTRs, CDSs and introns) were classified primarily according to the categories. Because the classification and visualization power is very high, SOM is an efficient and powerful tool for extracting a wide range of genome information.

3' Untranslated Regions↗
PubMed2006

Unsupervised multiscale color image segmentation based on MDL principle.

We present an unsupervised multiscale color image segmentation algorithm. The basic idea is to apply mean shift clustering to obtain an over-segmentation and then merge regions at multiple scales to minimize the minimum description length criterion. The performance on the Berkeley segmentation benchmark campares favorably with some existing approaches.

Algorithms↗
PubMed2005

Protein expression profiling identifies subclasses of breast cancer and predicts prognosis.

Breast cancer is a heterogeneous disease whose evolution is difficult to predict by using classic histoclinical prognostic factors. Prognostic classification can benefit from molecular analyses such as large-scale expression profiling. Using immunohistochemistry on tissue microarrays, we have monitored the expression of 26 selected proteins in more than 1,600 cancer samples from 552 consecutive patients with early breast cancer. Both an unsupervised approach and a new supervised method were used to analyze these profiles. Hierarchical clustering identified relevant clusters of coexpressed proteins and clusters of tumors. We delineated protein clusters associated with the estrogen receptor and with proliferation. Tumor clusters correlated with several histoclinical features of samples, including 5-year metastasis-free survival (MFS), and with the recently proposed pathophysiologic taxonomy of disease. The supervised method identified a set of 21 proteins whose combined expression significantly correlated to MFS in a learning set of 368 patients (P < 0.0001) and in a validation set of 184 patients (P < 0.0001). Among the 552 patients, the 5-year MFS was 90% for patients classified in the "good-prognosis class" and 61% for those classified in the "poor-prognosis class" (P < 0.0001). This difference remained significant when the molecular grouping was applied according to lymph node or estrogen receptor status, as well as the type of adjuvant systemic therapy. In multivariate analysis, the 21-protein set was the strongest independent predictor of clinical outcome. These results show that protein expression profiling may be a clinically useful approach to assess breast cancer heterogeneity and prognosis in stage I, II, or III disease.

Adult↗
PubMed2005

An incremental network for on-line unsupervised classification and topology learning.

This paper presents an on-line unsupervised learning mechanism for unlabeled data that are polluted by noise. Using a similarity threshold-based and a local error-based insertion criterion, the system is able to grow incrementally and to accommodate input patterns of on-line non-stationary data distribution. A definition of a utility parameter, the error-radius, allows this system to learn the number of nodes needed to solve a task. The use of a new technique for removing nodes in low probability density regions can separate clusters with low-density overlaps and dynamically eliminate noise in the input data. The design of two-layer neural network enables this system to represent the topological structure of unsupervised on-line data, report the reasonable number of clusters, and give typical prototype patterns of every cluster without prior conditions such as a suitable number of nodes or a good initial codebook.

Algorithms↗
PubMed2004

Ontogeny and oncogenesis balance the transcriptional profile of renal cell cancer.

Global transcript analysis is increasingly used to describe cancer taxonomies beyond the microscopic reach of the eye. Diagnostic and prognostic portraits are formulated by ranking cancers according to transcriptional proximity. However, the role that distinct biological factors play in defining these portraits remains undefined. It is likely that the transcriptional repertoire of cancers depends, on one hand, on the anamnestic retention of their ontogenesis and, on the other, on the emergence of novel expression patterns related to oncogenesis. We compared the transcriptional profile of primary renal cell cancers (RCCs) with that of normal kidney tissue and several epithelial cancers of nonrenal origin to weigh the contribution that ontogeny and oncogenesis make in molding their genetic profile. Unsupervised global transcript analysis demonstrated that RCCs retain transcriptional signatures related to their ontogeny and cluster close to normal renal epithelium. When renal lineage-associated genes are removed from the analysis and cancer-specific genes are analyzed, RCCs segregate with other cancers with limited lineage specificity underlying a predominance of the oncogenic process over lineage specificity. However, a RCC-specific set of oncogenesis-related genes was identified and surprisingly shared by sarcomas. In summary, the transcriptional portrait of primary RCCs is largely dominated by ontogeny. Genes responsible for lineage specificity may represent poor molecular targets for immune or drug therapy. Most genes associated with oncogenesis are shared with other cancers and may represent better therapeutic targets. Finally, a small subset of genes is associated with lineage-specific oncogenesis, and these may provide information regarding the biological behavior of RCCs and facilitate diagnostic classification of RCCs.

Adenocarcinoma, Clear Cell↗
PubMed2004

An accurate and robust method for unsupervised assessment of abdominal fat by MRI.

PURPOSE: To describe and evaluate an automatic and unsupervised method for assessing the quantity and distribution of abdominal adipose tissue by MRI. MATERIAL AND METHODS: A total of 20 patients underwent whole-abdomen MRI. A total of 32 transverse T1-weighted images were acquired from each subject. The data collected were transferred to a dedicated workstation and analyzed by both our unsupervised method and a manual procedure. The proposed methodology allows the automatic processing of MRI axial images, segmenting the adipose tissue by fuzzy clustering approach. The use of an active contour algorithm on image masks provided by the fuzzy clustering algorithm allows the separation of subcutaneous fat from visceral fat. Finally, an automated procedure based on automatic image histogram analysis identifies the visceral fat. RESULTS: The accuracy, reproducibility, and speed of our automatic method were compared with the state-of-the-art manual approach. The unsupervised analysis correlated well with the manual analysis, and was significantly faster than manual tracing. Moreover, the unsupervised method was not affected by intraobserver and interobserver variability. CONCLUSION: The results obtained demonstrate that the proposed method can provide the volume of subcutaneous adipose tissue, visceral adipose tissue, global adipose tissue, and the ratio between subcutaneous and visceral fat in an unsupervised and effective manner.

Abdominal Cavity↗
PubMed1996

Segmentation of multispectral magnetic resonance image using penalized fuzzy competitive learning network.

Segmentation (tissue classification) of the medical images obtained from Magnetic resonance (MR) images is a primary step in most applications of computer vision to medical image analysis. This paper describes a penalized fuzzy competitive learning network designed to segment multispectral MR spin echo images. The proposed approach is a new unsupervised and winner-takes-all scheme based on a neural network using the penalized fuzzy clustering technique. Its implementation consists of the combination of a competitive learning network and penalized fuzzy clustering methods in order to make parallel implementation feasible. The penalized fuzzy competitive learning network could provide an acceptable result for medical image segmentation in parallel processing using the hardware implementation. The experimental results show that a promising solution can be obtained using the penalized fuzzy competitive learning neural network based on least squares criteria.

Adult↗
PubMed1999

An automatic classification technique for attenuation correction in positron emission tomography.

In this paper a clustering technique is proposed for attenuation correction (AC) in positron emission tomography (PET). The method is unsupervised and adaptive with respect to counting statistics in the transmission (TR) images. The technique allows the classification of pre- or post-injection TR images into main tissue components in terms of attenuation coefficients. The classified TR images are then forward projected to generate new TR sinograms to be used for AC in the reconstruction of the corresponding emission (EM) data. The technique has been tested on phantoms and clinical data of brain, heart and whole-body PET studies. The method allows: (a) reduction of noise propagation from TR into EM images, (b) reduction of TR scanning to a few minutes (3 min) with maintenance of the quantitative accuracy (within 6%) of longer acquisition scans (15-20 min), (c) reduction of the radiation dose to the patient, (d) performance of quantitative whole-body studies.

Brain↗
PubMed2004

Transcriptional response to corticotropin-releasing factor in AtT-20 cells.

Corticotropin-releasing factor (CRF) plays a central role in the regulation of the hypothalamic-pituitary-adrenal axis, mediating endocrine and behavioral responses to various stressors. Two high-affinity receptors for CRF have been described. Although many of the intracellular signaling pathways activated by CRF have been studied extensively, our knowledge of transcriptional responses downstream of the CRF receptor 1 (CRFR1) is still limited. To elucidate gene networks regulated by CRF and CRFR1, we applied microarray technology to explore transcriptional response to CRF stimulation. Therefore, mouse pituitary-derived AtT-20 cells were exposed continuously to CRF either in the presence or absence of the specific CRFR1 antagonist R121919. Transcriptional responses to different treatments were studied in a time course ranging from 0.5 to 24 h. Microarray data were analyzed using classic microarray data analysis tools such as correspondence factor analysis, cluster analysis, and fold-change filtering. Furthermore, spectral map analysis was applied, a recently introduced unsupervised multivariate analysis method. A broad and transient transcriptional response to CRF was identified that could be blocked by the antagonist. This way, several known CRF-induced target genes and novel CRF responsive genes were identified. These include transcription factors such as cAMP-responsive element modulator (7x increased), secreted peptides such as cholecystokinin (1.5x), and proteins involved in modulating intracellular signaling, such as regulator of G-protein signaling 2 (11x). Up-regulation of many of these genes can be explained as negative feedback, attenuating CRF-activated pathways. In addition, spectral map analysis proved to be a promising new tool for microarray data analysis.

Animals↗
PubMed2006

Magnetoencephalographic artifact identification and automatic removal based on independent component analysis and categorization approaches.

Artifact signals from eye movements, heart beat and muscle activity contaminate magnetoencephalographic (MEG) signals generated from the neural activities inside the brain. Rejection of contaminated trials not only causes data loss, but can also significantly increase the experimental time or even prevent the analysis of highly contaminated or noisy data. We combined the use of independent component analysis (ICA) and clustering methods to isolate the artifacts from MEG signals. Threshold-based clustering analyses based on the topographic pattern, statistical aspects and power spectral patterns of independent components (ICs) successfully identified ICs related to certain types of artifacts. Unsupervised neural network based on the Adaptive Resonance Theory (ART) also categorized the artifact ICs, albeit with lower accuracy. Performance of the identification methods were evaluated with measurements of underestimation and overestimation of the target artifactual ICs. The combination of threshold-based clustering and ART-2 neural network categorization methods demonstrated the best identification performance. Comparison between contaminated and artifact-cleaned MEG signal waveforms showed the efficiency of the proposed methods of artifacts rejection. The analysis of the artifact components suggested the possibility of automatic artifact removal based on general templates.

Algorithms↗
PubMed1999

Estimating the number of clusters in multivariate data by self-organizing maps.

Determining the structure of data without prior knowledge of the number of clusters or any information about their composition is a problem of interest in many fields, such as image analysis, astrophysics, biology, etc. Partitioning a set of n patterns in a p-dimensional feature space must be done such that those in a given cluster are more similar to each other than the rest. As there are approximately Kn/K! possible ways of partitioning the patterns among K clusters, finding the best solution is very hard when n is large. The search space is increased when we have no a priori number of partitions. Although the self-organizing feature map (SOM) can be used to visualize clusters, the automation of knowledge discovery by SOM is a difficult task. This paper proposes region-based image processing methods to post-processing the U-matrix obtained after the unsupervised learning performed by SOM. Mathematical morphology is applied to identify regions of neurons that are similar. The number of regions and their labels are automatically found and they are related to the number of clusters in a multivariate data set. New data can be classified by labeling it according to the best match neuron. Simulations using data sets drawn from finite mixtures of p-variate normal densities are presented as well as related advantages and drawbacks of the method.

Algorithms↗
PubMed2004

Molecular cytogenetic identification of subgroups of grade III invasive ductal breast carcinomas with different clinical outcomes.

Tumor grade is an established indicator of breast cancer outcome, although considerable heterogeneity exists even within-grade. Around 25% of grade III invasive ductal breast carcinomas are associated with a "basal" phenotype, and these tumors are reported to be a distinct subgroup. We have investigated whether this group of breast cancers has a distinguishing pattern of genetic alterations and which of these may relate to the different clinical outcome of these patients. We performed comparative genomic hybridization (CGH) analysis on 43 grade III invasive ductal breast carcinomas positive for basal cytokeratin 14, as well as 43 grade- and age-matched CK14-negative controls, all with up to 25 years (median, 7 years) of clinical follow-up. Significant differences in CGH alterations were seen between the two groups in terms of mean number of changes (CK14+ve - 6.5, CK14-ve - 10.3; P = 0.0012) and types of alterations at chromosomes 4q, 7q, 8q, 9p, 13q, 16p, 17p, 17q, 19p, 19q, 20p, 20q and Xp. Supervised and unsupervised algorithms separated the two groups on CGH data alone with 76% and 74% accuracy, respectively. Hierarchical clustering revealed distinct subgroups, one of which contained 18 (42%) of the CK14+ve tumors. This subgroup had significantly shorter overall survival (P=0.0414) than other grade III tumors, regardless of CK14 status, and was an independent prognostic marker (P=0.031). These data provide evidence that the "basal" phenotype on its own does not convey a poor prognosis. Basal tumors are also heterogeneous with only a subset, identifiable by pattern of genetic alterations, exhibiting a shorter overall survival. Robust characterization of this basal group is necessary if it is to have a major impact on management of patients with breast cancer.

Algorithms↗
PubMed1999

A unifying review of linear gaussian models.

Factor analysis, principal component analysis, mixtures of gaussian clusters, vector quantization, Kalman filter models, and hidden Markov models can all be unified as variations of unsupervised learning under a single basic generative model. This is achieved by collecting together disparate observations and derivations made by many previous authors and introducing a new way of linking discrete and continuous state models using a simple nonlinearity. Through the use of other nonlinearities, we show how independent component analysis is also a variation of the same basic generative model. We show that factor analysis and mixtures of gaussians can be implemented in autoencoder neural networks and learned using squared error plus the same regularization term. We introduce a new model for static data, known as sensible principal component analysis, as well as a novel concept of spatially adaptive observation noise. We also review some of the literature involving global and local mixtures of the basic models and provide pseudocode for inference and learning for all the basic models.

Algorithms↗