Alfred Ultsch

Data Mining and Knowledge Discovery with Emergent Self-Organizing Feature Maps for Multivariate Time Series

Publisher Summary Data mining aims to discover hitherto unknown knowledge in large datasets. The most important step thereby is the transition from sub-symbolic to symbolic knowledge. Self-organizing feature maps (SOFM), when used appropriately, can exhibit emergent phenomena. SOFM with only a few neurons limit this ability, therefore emergent feature maps need to have thousands of neurons. The structures of emergent feature maps can be visualized using u-matrix methods. U-matrices lead to the construction of self-organizing classifiers possessing the ability to classify new data points. This sub-symbolic knowledge can be converted to a symbolic form which is understandable for humans. All these steps were combined into a system for neuronal data mining. This system has been applied successfully for knowledge discovery in multivariate time series.

Optimizing time series discretization for knowledge discovery

Knowledge Discovery in time series usually requires symbolic time series. Many discretization methods that convert numeric time series to symbolic time series ignore the temporal order of values. This often leads to symbols that do not correspond to states of the process generating the time series and cannot be interpreted meaningfully. We propose a new method for meaningful unsupervised discretization of numeric time series called Persist. The algorithm is based on the Kullback-Leibler divergence between the marginal and the self-transition probability distributions of the discretization symbols. Its performance is evaluated on both artificial and real life data in comparison to the most common discretization methods. Persist achieves significantly higher accuracy than existing static methods and is robust against noise. It also outperforms Hidden Markov Models for all but very simple cases.

Targeting lipid metabolism by the lipoprotein lipase inhibitor orlistat results in apoptosis of B-cell chronic lymphocytic leukemia cells

Constitutively activated pathways contribute to apoptosis resistance in chronic lymphocytic leukemia (CLL). Little is known about the metabolism of lipids and function of lipases in CLL cells. Performing gene expression profiling including B-cell receptor (BCR) stimulation of CLL cells in comparison to healthy donor CD5+ B cells, we found significant overexpression of lipases and phospholipases in CLL cells. In addition, we observed that the recently defined prognostic factor lipoprotein lipase (LPL) is induced by stimulation of BCR in CLL cells but not in CD5+ normal B cells. CLL cellular lysates exhibited significantly higher lipase activity compared to healthy donor controls. Incubation of primary CLL cells (n=26) with the lipase inhibitor orlistat resulted in induction of apoptosis, with a half-maximal dose (IC50) of 2.35 μM. In healthy B cells a significantly higher mean IC50 of 148.5 μM of orlistat was observed, while no apoptosis was induced in healthy peripheral blood mononuclear cells (PBMCs; P<0.001). Orlistat-mediated cytotoxicity was decreased by BCR stimulation. Finally, the cytotoxic effects of orlistat on primary CLL cells were enhanced by the simultaneous incubation with fludarabine (P=0.003). In summary, alterations of lipid metabolism are involved in CLL pathogenesis and might represent a novel therapeutic target in CLL.

Efficient mining of understandable patterns from multivariate interval time series

We present a new method for the understandable description of local temporal relationships in multivariate data, called Time Series Knowledge Mining (TSKM). We define the Time Series Knowledge Representation (TSKR) as a new language for expressing temporal knowledge in time interval data. The patterns have a hierarchical structure, with levels corresponding to the temporal concepts duration, coincidence, and partial order. The patterns are very compact, but offer details for each element on demand. In comparison with related approaches, the TSKR is shown to have advantages in robustness, expressivity, and comprehensibility. The search for coincidence and partial order in interval data can be formulated as instances of the well known frequent itemset problem. Efficient algorithms for the discovery of the patterns are adapted accordingly. A novel form of search space pruning effectively reduces the size of the mining result to ease interpretation and speed up the algorithms. Human interaction is used during the mining to analyze and validate partial results as early as possible and guide further processing steps. The efficacy of the methods is demonstrated using two real life data sets. In an application to sports medicine the results were recognized as valid and useful by an expert of the field.

Knowledge Extraction from Self-Organizing Neural Networks

In this work we present the integration of neural networks with a rule based expert system. The system realizes the automatic acquisition of knowledge out of a set of examples. It enhances the reasoning capabilities of classical expert systems with the ability of generalise and the handling of incomplete cases. It uses neural nets with unsupervised learning algorithms to extract regularities out of case data. A symbolic rule generator transforms these regularities into Prolog rules. The generated rules and the trained neural nets are embedded into the expert system as knowledge bases. In the system’s diagnosis phase it is possible to use these knowledge bases together with human expert’s knowledge bases in order to diagnose a unknown case. Furthermore the system is able to diagnose and to complete inconsistent data using the trained neural nets exploiting their ability to generalise.

Advances in Data Analysis, Data Handling and Business Intelligence

Invited.- Semi-supervised Probabilistic Distance Clustering and the Uncertainty of Classification.- Strategies of Model Construction for the Analysis of Judgment Data.- Clustering of High-Dimensional Data via Finite Mixture Models.- Clustering and Dimensionality Reduction to Discover Interesting Patterns in Binary Data.- Kernel Methods for Detecting the Direction of Time Series.- Statistical Processes Under Change: Enhancing Data Quality with Pretests.- Clustering and Classification.- Evaluation Strategies for Learning Algorithms of Hierarchies.- Fuzzy Subspace Clustering.- Motif-Based Classification of Time Series with Bayesian Networks and SVMs.- A Novel Approach to Construct Discrete Support Vector Machine Classifiers.- Predictive Classification Trees.- Isolated Vertices in Random Intersection Graphs.- Strengths and Weaknesses of Ant Colony Clustering.- Variable Selection for Kernel Classifiers: A Feature-to-Input Space Approach.- Finite Mixture and Genetic Algorithm Segmentation in Partial Least Squares Path Modeling: Identification of Multiple Segments in Complex Path Models.- Cluster Ensemble Based on Co-occurrence Data.- Localized Logistic Regression for Categorical Influential Factors.- Clustering Association Rules with Fuzzy Concepts.- Clustering with Repulsive Prototypes.- Mixture Analysis.- Weakly Homoscedastic Constraints for Mixtures of -Distributions.- Bayesian Methods for Graph Clustering.- Determining the Number of Components in Mixture Models for Hierarchical Data.- Testing Mixed Distributions when the Mixing Distribution Is Known.- Classification with a Mixture Model Having an Increasing Number of Components.- Nonparametric Fine Tuning of Mixtures: Application to Non-Life Insurance Claims Distribution Estimation.- Linguistics and Text Analysis.- Classification of Text Processing Components: The Tesla Role System.- Nonparametric Distribution Analysis for Text Mining.- Linear Coding of Non-linear Hierarchies: Revitalization of an Ancient Classification Method.- Automatic Dictionary Expansion Using Non-parallel Corpora.- Multilingual Knowledge-Based Concept Recognition in Textual Data.- Pattern Recognition and Machine Learning.- A Diversified Investment Strategy Using Autonomous Agents.- Classification with Kernel Mahalanobis Distance Classifiers.- Identifying Influential Cases in Kernel Fisher Discriminant Analysis by Using the Smallest Enclosing Hypersphere.- Self-Organising Maps for Image Segmentation.- Image Based Mail Piece Identification Using Unsupervised Learning.- Statistical Musicology.- Statistical Analysis of Human Body Movement and Group Interactions in Response to Music.- Applying Statistical Models and Parametric Distance Measures for Music Similarity Search.- Finding Music Fads by Clustering Online Radio Data with Emergent Self Organizing Maps.- Analysis of Polyphonic Musical Time Series.- Banking and Finance.- Hedge Funds and Asset Allocation: Investor Confidence, Diversification Benefits, and a Change in Investment Style Composition.- Mixture Hidden Markov Models in Finance Research.- Multivariate Comparative Analysis of Stock Exchanges: The European Perspective.- Empirical Examination of Fundamental Indexation in the German Market.- The Analysis of Power For Some Chosen Backtesting Procedures: Simulation Approach.- Extreme Unconditional Dependence Vs. Multivariate GARCH Effect in the Analysis of Dependence Between High Losses on Polish and German Stock Indexes.- Is Log Ratio a Good Value for Measuring Return in Stock Investments?.- Marketing, Management Science and Economics.- Designing Products Using Quality Function Deployment and Conjoint Analysis: A Comparison in a Market for Elderly People.- Analyzing the Stability of Price Response Functions: Measuring the Influence of Different Parameters in a Monte Carlo Comparison.- Real Options in the Assessment of New Products.- Exploring the Interaction Structure of Weblogs.- Analyzing Preference Rankings when There Are Too Many Alternatives.- Considerations on the Impact of Ill-Conditioned Configurations in the CML Approach.- Dyadic Interactions in Service Encounter: Bayesian SEM Approach.- Archaeology and Spatial Planning.- Estimating the Number of Buildings in Germany.- Mapping Findspots of Roman Military Brickstamps in (Mainz) and Archaeometrical Analysis.- Analysis of Guarantor and Warrantee Relationships Among Government Officials in the Eighth Century in the Old Capital of Japan by Using Asymmetric Multidimensional Scaling.- Analysis of Massive Emigration from Poland: The Model-Based Clustering Approach.- Bio- and Health Sciences.- Systematics of Short-range Correlations in Eukaryotic Genomes.- On Classification of Molecules and Species of Representation Rings.- The Precise and Efficient Identification of Medical Order Forms Using Shape Trees.- On the Prognostic Value of Gene Expression Signatures for Censored Data.- Quality-Based Clustering of Functional Data: Applications to Time Course Microarray Data.- A Comparison of Algorithms to Find Differentially Expressed Genes in Microarray Data.- Exploratory Data Analysis, Modeling and Applications.- Data Compression and Regression Based on Local Principal Curves.- Optimization of Centrifugal Impeller Using Evolutionary Strategies and Artificial Neural Networks.- Efficient Media Exploitation Towards Collective Intelligence.- Multi-Class Extension of Verifiable Ensemble Models for Safety-Related Applications.- Dynamic Disturbances in BTA Deep-Hole Drilling: Modelling Chatter and Spiralling as Regenerative Effects.- Nonnegative Matrix Factorization for Binary Data to Extract Elementary Failure Maps from Wafer Test Images.- Collective Intelligence Generation from User Contributed Content.- Computation of the Molenaar Sijtsma Statistic.

Modeling timbre distance with temporal statistics from polyphonic music

Timbre distance and similarity are expressions of the phenomenon that some music appears similar while other songs sound very different to us. The notion of genre is often used to categorize music, but songs from a single genre do not necessarily sound similar and vice versa. In this work, we analyze and compare a large amount of different audio features and psychoacoustic variants thereof for the purpose of modeling timbre distance. The sound of polyphonic music is commonly described by extracting audio features on short time windows during which the sound is assumed to be stationary. The resulting down sampled time series are aggregated to form a high-level feature vector describing the music. We generated high-level features by systematically applying static and temporal statistics for aggregation. The temporal structure of features in particular has previously been largely neglected. A novel supervised feature selection method is applied to the huge set of possible features. The distances of the selected feature correspond to timbre differences in music. The features show few redundancies and have high potential for explaining possible clusters. They outperform seven other previously proposed feature sets on several datasets with respect to the separation of the known groups of timbrally different music.

Functional genomics of pain in analgesic drug development and therapy

Advances in genomic research have led to the clarification of the detailed involvement of gene products in biological pathways and these are being increasingly exploited in strategies for drug discovery and repurposing. Concomitant developments in informatics have resulted in the acquisition of complex gene information through the application of computational analysis of molecular interaction networks. This approach enables the acquired knowledge on hundreds of genes to be used to view molecular disease mechanisms from a genetic point of view. By analyzing 410 genes which control the complex process of pain, we show by computational analysis, based on functional annotations to pain-related genes, that 12 clearly circumscribed functional areas are essential for pain perception and thus for analgesic drug development. The genetics perspective revealed that future development strategies should focus on substances modulating intracellular signal transduction, ion transport and anatomical structure development. These processes are involved in the genetic-based absence of pain and therefore, provide promising fields for curative or preventive treatments. In contrast, interactions with G-protein coupled receptor pathways seem merely to provide symptomatic, not preventative relief of pain. In addition, biological functions accessed either by analgesic drugs or microRNAs suggest that synergistic therapies may be a future direction for drug development. With modern computational functional genomics, it is possible to exploit genetic information from increasingly available data sets on complex diseases, such as pain, and offers a new insight into drug development and therapy which is complementary to pathway-centered approaches.

Overexpression of TOSO in CLL is triggered by B-cell receptor signaling and associated with progressive disease

Resistance toward apoptotic stimuli mediated by overexpression of antiapoptotic factors or extracellular survival signals is considered to be responsible for accumulation of malignant B cells in chronic lymphocytic leukemia (CLL). TOSO was identified as overexpressed candidate gene in CLL, applying unit-transformation assays of publicly available microarray datasets. Based on CLL samples from 106 patients, TOSO was identified to exhibit elevated relative expression (RE) of 6.8 compared with healthy donor B cells using quantitative real-time polymerase chain reaction (PCR; P = .004). High levels of TOSO expression in CLL correlated with high leukocyte count, advanced Binet stage, previous need for chemotherapy, and unmutated IgV(H) status. CD38(+) CLL subsets harboring proliferative activity showed enhanced TOSO expression. We evaluated functional mechanisms of aberrant TOSO expression and identified TOSO expression significantly induced by B-cell receptor (BCR) stimulation compared with control cells (RE; 8.25 vs 4.86; P = .01). In contrast, CD40L signaling significantly reduced TOSO expression (RE, 2.60; P = .01). In summary, we show that the antiapoptotic factor TOSO is associated with progressive disease and enhanced in the proliferative CD38(+) CLL subset. Both association with unmutated IgV(H) and the specific induction of TOSO via the BCR suggest autoreactive BCR signaling as a key mediator of apoptosis resistance in CLL.

Pareto Density Estimation: A Density Estimation for Knowledge Discovery

Pareto Density Estimation (PDE) as defined in this work is a method for the estimation of probability density functions using hyperspheres. The radius of the hyperspheres is derived from optimizing information while minimizing set size. It is shown, that PDE is a very good estimate for data containing clusters of Gaussian structure. The behavior of the method is demonstrated with respect to cluster overlap, number of clusters, different variances in different clusters and application to high dimensional data. For high dimensional data PDE is found to be appropriate for the purpose of cluster analysis. The method is tested successfully on a difficult high dimensional real world problem: stock picking in falling markets.