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 523 records · Page 29Linked to original sources
PubMed2004

Unsupervised learning of an atlas from unlabeled point-sets.

One of the key challenges in deformable shape modeling is the problem of estimating a meaningful average or mean shape from a set of unlabeled shapes. We present a new joint clustering and matching algorithm that is capable of computing such a mean shape from multiple shape samples which are represented by unlabeled point-sets. An iterative bootstrap process is used wherein multiple shape sample point-sets are nonrigidly deformed to the emerging mean shape, with subsequent estimation of the mean shape based on these nonrigid alignments. The process is entirely symmetric with no bias toward any of the original shape sample point-sets. We believe that this method can be especially useful for creating atlases of various shapes present in medical images. We have applied the method to create mean shapes from nine hand-segmented 2D corpus callosum data sets and 10 hippocampal 3D point-sets.

Algorithms↗
PubMed1997

Learning retina implants with epiretinal contacts.

Retina implants are currently being developed by several interdisciplinary research consortia worldwide for blind humans with various retinal degenerative diseases. It is the aim of our retina implant project to develop a novel type of visual prosthesis to regain a moderate amount of vision such as perception of location and shape of large objects in the first stage and to approach reading quality in a subsequent stage. In our planned retina implant, a retina encoder (RE) outside the eye has to replace the information processing of the retina. A retina stimulator (RS), implanted adjacently to the retinal ganglion cell layer, has to contact a sufficient number of retinal ganglion cells/fibers for electrical elicitation of spikes. A wireless signal and energy transmission system has to provide the communication between the RE and RS. This paper outlines the retina implant project of our consortium of 14 expert groups and describes first results of the learning RE. The RE approximates the typical receptive field (RF) properties of primate retinal ganglion cells by means of individually tunable spatiotemporal RF filters. The RE as a cluster of RF filters maps visual patterns onto spike trains for a number of contacted ganglion cells. A concept is presented to train the individual RF filters in an unsupervised learning process, which employs neural networks in a dialog with the individual human subject. The desired aim of this dialog is an optimization of the visual perception by matching the various RF filter properties with those 'expected' by the central visual system for each contacted ganglion cell.

Animals↗
PubMed2005

Molecular alterations in primary prostate cancer after androgen ablation therapy.

PURPOSE: After an initial response to androgen ablation, most prostate tumors recur, ultimately progressing to highly aggressive androgen-independent cancer. The molecular mechanisms underlying progression are not well known in part due to the rarity of androgen-independent samples from primary and metastatic sites. EXPERIMENTAL DESIGN: We compared the gene expression profiles of 10 androgen-independent primary prostate tumor biopsies with 10 primary, untreated androgen-dependent tumors. Samples were laser capture microdissected, the RNA was amplified, and gene expression was assessed using Affymetrix Human Genome U133A GeneChip. Differential expression was examined with principal component analysis, hierarchical clustering, and Student's t testing. Analysis of gene ontology was done with Expression Analysis Systematic Explorer and gene expression data were integrated with genomic alterations with Differential Gene Locus Mapping. RESULTS: Unsupervised principal component analysis showed that the androgen-dependent and androgen-independent tumors segregated from one another. After filtering the data, 239 differentially expressed genes were identified. Two main gene ontologies were found discordant between androgen-independent and androgen-dependent tumors: macromolecule biosynthesis was down-regulated and cell adhesion was up-regulated in androgen-independent tumors. Other differentially expressed genes were related to interleukin-6 signaling as well as angiogenesis, cell adhesion, apoptosis, oxidative stress, and hormone response. The Differential Gene Locus Mapping analysis identified nine regions of potential chromosomal deletion in the androgen-independent tumors, including 1p36, 3p21, 6p21, 8p21, 11p15, 11q12, 12q23, 16q12, and 16q21. CONCLUSIONS: Taken together, these data identify several unique characteristics of androgen-independent prostate cancer that may hold potential for the development of targeted therapeutic intervention.

Aged↗
PubMed2003

On learning to estimate the block directional image of a fingerprint using a hierarchical neural network.

This paper presents a hierarchical neural network architecture for computing fingerprints block directional images. Two separately trained neural networks are connected in series. First, the fingerprint image is divided into 16x16 blocks, each block is submitted to the first network which is a back propagation neural network. It has four counters in its output layer one for each direction to count the main directional codes in each fingerprint block. The output of this network is considered the feature vector for the fingerprint block, which is then submitted to the second network. The second network is a self-organized feature maps neural network uses an unsupervised learning strategy to group the fingerprint blocks into distinct directional classes. In this scheme, there is more than one sub-class for each directional class, an agglomerative hierarchical cluster algorithm for merging two clusters is used to merge two classes if their corresponding distances are below a specified threshold. Results obtained with a real world data set indicate the effectiveness of the proposed architecture.

Algorithms↗
PubMedDate not supplied

Quantitative self-organizing maps for clustering electron tomograms.

Tomography emerges as a powerful methodology for determining the complex architectures of biological specimens that are better regarded from the structural point of view as singular entities. However, once the structure of a sufficiently large number of singular specimens is solved, quite possibly structural patterns start to emerge. This latter situation is addressed here, where the clustering of a set of 3D reconstructions using a novel quantitative approach is presented. In general terms, we propose a new variant of a self-organizing neural network for the unsupervised classification of 3D reconstructions. The novelty of the algorithm lies in its rigorous mathematical formulation that, starting from a large set of noisy input data, finds a set of "representative" items, organized onto an ordered output map, such that the probability density of this set of representative items resembles at its possible best the probability density of the input data. In this study, we evaluate the feasibility of application of the proposed neural approach to the problem of identifying similar 3D motifs within tomograms of insect flight muscle. Our experimental results prove that this technique is suitable for this type of problem, providing the electron microscopy community with a new tool for exploring large sets of tomogram data to find complex patterns.

Algorithms↗
PubMed2001

Best harmony, unified RPCL and automated model selection for unsupervised and supervised learning on Gaussian mixtures, three-layer nets and ME-RBF-SVM models.

After introducing the fundamentals of BYY system and harmony learning, which has been developed in past several years as a unified statistical framework for parameter learning, regularization and model selection, we systematically discuss this BYY harmony learning on systems with discrete inner-representations. First, we shown that one special case leads to unsupervised learning on Gaussian mixture. We show how harmony learning not only leads us to the EM algorithm for maximum likelihood (ML) learning and the corresponding extended KMEAN algorithms for Mahalanobis clustering with criteria for selecting the number of Gaussians or clusters, but also provides us two new regularization techniques and a unified scheme that includes the previous rival penalized competitive learning (RPCL) as well as its various variants and extensions that performs model selection automatically during parameter learning. Moreover, as a by-product, we also get a new approach for determining a set of 'supporting vectors' for Parzen window density estimation. Second, we shown that other special cases lead to three typical supervised learning models with several new results. On three layer net, we get (i) a new regularized ML learning, (ii) a new criterion for selecting the number of hidden units, and (iii) a family of EM-like algorithms that combines harmony learning with new techniques of regularization. On the original and alternative models of mixture-of-expert (ME) as well as radial basis function (RBF) nets, we get not only a new type of criteria for selecting the number of experts or basis functions but also a new type of the EM-like algorithms that combines regularization techniques and RPCL learning for parameter learning with either least complexity nature on the original ME model or automated model selection on the alternative ME model and RBF nets. Moreover, all the results for the alternative ME model are also applied to other two popular nonparametric statistical approaches, namely kernel regression and supporting vector machine. Particularly, not only we get an easily implemented approach for determining the smoothing parameter in kernel regression, but also we get an alternative approach for deciding the set of supporting vectors in supporting vector machine.

Algorithms↗
PubMed1999

Artificial neural networks in laboratory medicine and medical outcome prediction.

Since the early nineties the number of scientific papers reporting on artificial neural network (ANN) applications in medicine has been quickly increasing. In the present paper, we describe in some detail the architecture of network types used most frequently in ANN applications in the broad field of laboratory medicine and clinical chemistry, present a technique-structured review about the recent ANN applications in the field, and give information about the improvements of available ANN software packages. ANN applications are divided into two main classes: supervised and unsupervised methods. Most of the described supervised applications belong to the fields of medical diagnosis (n = 7) and outcome prediction (n = 9). Laboratory and clinical data are presented to multilayer feed-forward ANNs which are trained by the back propagation algorithm. Results are often better than those of traditional techniques such as linear discriminant analysis, classification and regression trees (CART), Cox regression analysis, logistic regression, clinical judgement or expert systems. Unsupervised ANN applications provide the ability of reducing the dimensionality of a dataset. Low-dimensional plots can be generated and visually understood and compared. Results are very similar to that of cluster analysis and factor analysis. The ability of Kohonen's self-organizing maps to generate 2D maps of molecule surface properties was successfully applied in drug design.

Chemistry, Clinical↗
PubMed2006

Gene expression profiles of primary breast carcinomas from patients at high risk for local recurrence after breast-conserving therapy.

PURPOSE: Several risk factors for local recurrence of breast cancer after breast-conserving therapy (BCT) have been identified. The identification of additional risk factors would be very useful in guiding optimal therapy and also in improving understanding of the mechanisms underlying local recurrence. We used cDNA microarray analysis to identify gene expression profiles associated with local recurrence. EXPERIMENTAL DESIGN: Using 18K cDNA microarrays, gene expression profiles were obtained from 50 patients who underwent BCT. Of these 50 patients, 19 developed a local recurrence; the remaining 31 patients were selected as controls as they were free of local recurrence at least 11 years after treatment. For 9 of 19 patients, the local recurrence was also available for gene expression profiling. Unsupervised and supervised methods of classification were used to separate patients in groups corresponding to disease outcome and to study the overall gene expression pattern of primary tumors and their recurrences. RESULTS: Hierarchical clustering of patients did not show any grouping reflecting local recurrence status. Supervised analysis revealed no significant set of genes that was able to distinguish recurring tumors from nonrecurring tumors. Paired-data analysis of primary tumors and local recurrences showed a remarkable similarity in gene expression profile between primary tumors and their recurrences. CONCLUSIONS: No significant differences in gene expression between primary breast cancer tumors in patients with or without local recurrence after BCT were identified. Furthermore, analyses of primary tumors and local recurrences show a preservation of the overall gene expression pattern in the local recurrence, even after radiotherapy.

Adult↗
PubMed2005

Non-linear mapping for exploratory data analysis in functional genomics.

BACKGROUND: Several supervised and unsupervised learning tools are available to classify functional genomics data. However, relatively less attention has been given to exploratory, visualisation-driven approaches. Such approaches should satisfy the following factors: Support for intuitive cluster visualisation, user-friendly and robust application, computational efficiency and generation of biologically meaningful outcomes. This research assesses a relaxation method for non-linear mapping that addresses these concerns. Its applications to gene expression and protein-protein interaction data analyses are investigated. RESULTS: Publicly available expression data originating from leukaemia, round blue-cell tumours and Parkinson disease studies were analysed. The method distinguished relevant clusters and critical analysis areas. The system does not require assumptions about the inherent class structure of the data, its mapping process is controlled by only one parameter and the resulting transformations offer intuitive, meaningful visual displays. Comparisons with traditional mapping models are presented. As a way of promoting potential, alternative applications of the methodology presented, an example of exploratory data analysis of interactome networks is illustrated. Data from the C. elegans interactome were analysed. Results suggest that this method might represent an effective solution for detecting key network hubs and for clustering biologically meaningful groups of proteins. CONCLUSION: A relaxation method for non-linear mapping provided the basis for visualisation-driven analyses using different types of data. This study indicates that such a system may represent a user-friendly and robust approach to exploratory data analysis. It may allow users to gain better insights into the underlying data structure, detect potential outliers and assess assumptions about the cluster composition of the data.

Algorithms↗
PubMed1987

Multispectral magnetic resonance image analysis.

Multiecho magnetic resonance (MR) scanning produces tomographic images with approximately equal morphologic information but varying gray scales at the same anatomic level. Multispectral image classification techniques, originally developed for satellite imaging, have recently been applied to MR tissue characterization. Statistical assessment of multispectral tissue classification techniques has been used to select the most promising of several alternative methods. MR examinations of the head and body, obtained with a 0.35, 0.5, or 1.5T imager, comprised data sets with at least two pulse sequences yielding three images at each anatomical level: (1) TR = 0.3 sec, TE = 30 msec, (2) TR = 1.5, TE = 30, (3) TR = 1.5, TE = 120. Normal and pathological images have been analyzed using multispectral analysis and image classification. MR image data are first subjected to radiometric and geometric corrections to reduce error resulting from (1) instrumental variations in data acquisition, (2) image noise, and (3) misregistration. Training regions of interest (ROI) are outlined in areas of normal (gray and white matter, CSF) and pathological tissue. Statistics are extracted from these ROIs and classification maps generated using table lookup, minimum distance to means, maximum likelihood, and cluster analysis. These synthetic maps are then compared pixel by pixel with manually prepared classification maps of the same MR images. Using these methods, the authors have found that: (1) both supervised and unsupervised classification techniques yielded theme maps (class maps) which demonstrated tissue characteristic signatures and (2) tissue classification errors found in computer-generated theme maps were due to subtle gray scale changes present in the original MR data sets arising from radiometric inhomogeneity and spatial nonuniformity.

Brain↗
PubMed2006

Spectral analysis of two-signed microarray expression data.

We give a simple and informative derivation of a spectral algorithm for clustering and reordering complementary DNA microarray expression data. Here, expression levels of a set of genes are recorded simultaneously across a number of samples, with a positive weight reflecting up-regulation and a negative weight reflecting down-regulation. We give theoretical support for the algorithm based on a biologically justified hypothesis about the structure of the data, and illustrate its use on public domain data in the context of unsupervised tumour classification. The algorithm is derived by considering a discrete optimization problem and then relaxing to the continuous realm. We prove that in the case where the data have an inherent 'checkerboard' sign pattern, the algorithm will automatically reveal that pattern. Further, our derivation shows that the algorithm may be regarded as imposing a random graph model on the expression levels and then clustering from a maximum likelihood perspective. This indicates that the output will be tolerant to perturbations and will reveal 'near-checkerboard' patterns when these are present in the data. It is interesting to note that the checkerboard structure is revealed by the first (dominant) singular vectors--previous work on spectral methods has focussed on the case of nonnegative edge weights, where only the second and higher singular vectors are relevant. We illustrate the algorithm on real and synthetic data, and then use it in a tumour classification context on three different cancer data sets. Our results show that respecting the two-signed nature of the data (thereby distinguishing between up-regulation and down-regulation) reveals structures that cannot be gleaned from the absolute value data (where up- and down-regulation are both regarded as 'changes').

Algorithms↗
PubMed2003

CASTOR: clustering algorithm for sequence taxonomical organization and relationships.

Given a set of related proteins, two important problems in biology are the inference of protein subsets such that members of one subset share a common function and the identification of protein regions that possess functional significance. The former is typically approached by hierarchical bottom-up clustering based on pairwise sequence similarity and various linkage rules. The latter is typically approached in a supervised manner, based on global multiple sequence alignment. However, the two problems are inextricably linked, since functional subsets are usually characterized by distinctive functional regions. This paper introduces CASTOR, an automatic and unsupervised system that addresses both problems simultaneously and efficiently. It identifies protein regions that are likely to have functional significance by discovering and refining statistically significant motifs. It infers likely functional protein subsets and their relationships based on the presence of the discovered motifs in a top-down and recursive manner, allowing the identification of both hierarchical and nonhierarchical subset relationships. This is, to our knowledge, the first system that approaches both problems simultaneously in a top-down, systematic manner. CASTOR's performance is evaluated against the G-protein coupled receptor superfamily. The identified protein regions lead to a taxonomical organization of this superfamily that is in remarkable agreement with a biologically motivated one and which outperforms those produced by bottom-up clustering methods. We also find that conventional hierarchical representations may fail to accurately describe the complexity of evolutionary development responsible for the final organization of a complex protein family. In particular, many functional relationships governing distant subfamilies of such a protein family may not be represented hierarchically.

Algorithms↗
PubMed2025

DNABERT-S: pioneering species differentiation with species-aware DNA embeddings.

SUMMARY: We introduce DNABERT-S, a tailored genome model that develops species-aware embeddings to naturally cluster and segregate DNA sequences of different species in the embedding space. Differentiating species from genomic sequences (i.e. DNA and RNA) is vital yet challenging, since many real-world species remain uncharacterized, lacking known genomes for reference. Embedding-based methods are therefore used to differentiate species in an unsupervised manner. DNABERT-S builds upon a pre-trained genome foundation model named DNABERT-2. To encourage effective embeddings to error-prone long-read DNA sequences, we introduce Manifold Instance Mixup (MI-Mix), a contrastive objective that mixes the hidden representations of DNA sequences at randomly selected layers and trains the model to recognize and differentiate these mixed proportions at the output layer. We further enhance it with the proposed Curriculum Contrastive Learning (C2LR) strategy. Empirical results on 28 diverse datasets show DNABERT-S's effectiveness, especially in realistic label-scarce scenarios. For example, it identifies twice more species from a mixture of unlabeled genomic sequences, doubles the Adjusted Rand Index (ARI) in species clustering, and outperforms the top baseline's performance in 10-shot species classification with just a 2-shot training. AVAILABILITY AND IMPLEMENTATION: Model, codes, and data are publically available at https://github.com/MAGICS-LAB/DNABERT_S.

Sequence Analysis, DNA↗
PubMed2024

DNABERT-S: Pioneering Species Differentiation with Species-Aware DNA Embeddings.

We introduce DNABERT-S, a tailored genome model that develops species-aware embeddings to naturally cluster and segregate DNA sequences of different species in the embedding space. Differentiating species from genomic sequences (i.e., DNA and RNA) is vital yet challenging, since many real-world species remain uncharacterized, lacking known genomes for reference. Embedding-based methods are therefore used to differentiate species in an unsupervised manner. DNABERT-S builds upon a pre-trained genome foundation model named DNABERT-2. To encourage effective embeddings to error-prone long-read DNA sequences, we introduce Manifold Instance Mixup (MI-Mix), a contrastive objective that mixes the hidden representations of DNA sequences at randomly selected layers and trains the model to recognize and differentiate these mixed proportions at the output layer. We further enhance it with the proposed Curriculum Contrastive Learning (C2LR) strategy. Empirical results on 23 diverse datasets show DNABERT-S's effectiveness, especially in realistic label-scarce scenarios. For example, it identifies twice more species from a mixture of unlabeled genomic sequences, doubles the Adjusted Rand Index (ARI) in species clustering, and outperforms the top baseline's performance in 10-shot species classification with just a 2-shot training. Model, codes, and data is publicly available at https://github.com/MAGlCS-LAB/DNABERT_S.

Journal Article↗
PubMed1996

Sub-species discrimination, using pyrolysis mass spectrometry and self-organising neural networks, of Propionibacterium acnes isolated from normal human skin.

Curie-point pyrolysis mass spectra were obtained from 30 Propionibacterium acnes strains isolated from the foreheads of six healthy humans. Multivariate analyses and Kohonen artificial neural networks (KANNs), employing unsupervised learning, were used successfully to discriminate between the P.acnes isolates from different individual hosts. The classification of the isolates by KANNs was compared with the more classical multivariate techniques of canonical variates analysis and hierarchical cluster analysis and found to give similar groupings. The combination of pyrolysis mass spectrometry with these numerical methods also showed that more than one strain of P.acnes had been isolated from three of the human hosts.

Adult↗
PubMed2026

Nested co-expression network analysis identifies compact gene clusters in a black box.

MOTIVATION: Digital analysis of biological systems requires methods capable of identifying both broad and nested gene modules reflecting complex biological processes. Existing transcriptomic methods often miss compact gene sets corresponding to subprocesses in specialized cell types, limiting insights into functional heterogeneity. RESULTS: We present Nested-WGCNA, a two-stage unsupervised network analysis algorithm designed to identify coarse-grained and fine-grained gene modules. Applied to bulk RNA-Seq data, Nested-WGCNA reveals stable modules reproducible across datasets. When validated against scRNA-Seq data, these modules correspond to both major and minor immune cell subtypes. Application to immunotherapy response datasets uncovers predictive and prognostic biomarkers, highlighting its utility in treatment stratification and biomarker discovery. AVAILABILITY: The NestedWGCNA source code and analysis pipeline are available on GitHub (https://github.com/ilyada/NestedWGCNA) and archived on Zenodo (https://doi.org/10.5281/zenodo.18959244).

Algorithms↗
PubMed1998

Slow stochastic Hebbian learning of classes of stimuli in a recurrent neural network.

We study unsupervised Hebbian learning in a recurrent network in which synapses have a finite number of stable states. Stimuli received by the network are drawn at random at each presentation from a set of classes. Each class is defined as a cluster in stimulus space, centred on the class prototype. The presentation protocol is chosen to mimic the protocols of visual memory experiments in which a set of stimuli is presented repeatedly in a random way. The statistics of the input stream may be stationary, or changing. Each stimulus induces, in a stochastic way, transitions between stable synaptic states. Learning dynamics is studied analytically in the slow learning limit, in which a given stimulus has to be presented many times before it is memorized, i.e. before synaptic modifications enable a pattern of activity correlated with the stimulus to become an attractor of the recurrent network. We show that in this limit the synaptic matrix becomes more correlated with the class prototypes than with any of the instances of the class. We also show that the number of classes that can be learned increases sharply when the coding level decreases, and determine the speeds of learning and forgetting of classes in the case of changes in the statistics of the input stream.

Generalization, Stimulus↗
PubMed2006

Category-specificity can emerge from bottom-up visual characteristics: evidence from a modular neural network.

The role of bottom-up visual processes in category-specific object recognition has been largely unexplored. We examined the role of low-level visual characteristics in category specific recognition using a modular neural network comprising both unsupervised and supervised components. One hundred standardised pictures from ten different categories (five living and five nonliving, including body parts and musical instruments) were presented to a Kohonen self-organising map (SOM) which re-represents the visual stimuli by clustering them within a smaller number of dimensions. The SOM representations were then used to train an attractor network to learn the superordinate category of each item. The ease with which the model acquired the category mappings was investigated with respect to emerging category effects. We found that the superordinates could be separated by very low-level visual factors (as extracted by the SOM). The model also accounted for the well documented atypicality of body parts and musical instrument superordinates. The model has clear relevance to human object recognition since the model was quicker to learn more typical category exemplars and finally the model also accounted for more than 20% of the naming variance in a sample of 57 brain injured subjects. We conclude that purely bottom-up visual characteristics can explain some important features of category-specific phenomena.

Female↗