Fatih Altiparmak

Incremental Maintenance of Online Summaries Over Multiple Streams

We propose a novel approach based on predictive quantization (PQ) for online summarization of multiple time-varying data streams. A synopsis over a sliding window of most recent entries is computed in one pass and dynamically updated in constant time. The correlation between consecutive data elements is effectively taken into account without the need for preprocessing. We extend PQ to multiple streams and propose structures for real-time summarization and querying of a massive number of streams. Queries on any subsequence of a sliding window over multiple streams are processed in real time. We examine each component of the proposed approach, prediction, and quantization separately and investigate the space-accuracy trade-off for synopsis generation. Complementing the theoretical optimality of PQ-based approaches, we show that the proposed technique, even for very short prediction windows, significantly outperforms the current techniques for a wide variety of query types on both synthetic and real data sets.

Relationship preserving feature selection for unlabelled clinical trials time-series

Feature selection has been widely studied in supervised data mining applications, where the typical goal is to create clusters through the selection of a reduced attribute set that maximizes classification accuracies. Such a goal may not be appropriate for preserving inter-attribute relationships of unlabelled time-series, such as the case of clinical trials data. In this paper, we select the features based on the time-series relationships of attributes by measuring their inter-attribute movement. We present performance measures and methods for feature selection over unlabelled time-series with the aim of preserving inter-attribute relationships. The performance metrics estimate the effectiveness of a given feature set with respect to representation quality by measuring the nearest neighbors before and after feature selection. We provide techniques to combine and compare data from non-standard variable-length time-series sources and provide a mechanism to inject expert opinion into the feature selection process. The methodologies and comparative results are presented in the context of a real pharmaceutical database application.

A Multi-metric Similarity Based Analysis of Microarray Data

Clustering has been shown to be effective in analyzing functional relationships of genes. However, no single clustering method with single distance metric is capable of capturing all types of relationships that a gene may have with other genes. In this paper we introduce a framework which groups genes around a query gene, and ranks them in order corresponding to different levels of similarity utilizing multiple metrics. The focus of our efforts is to create gene centric clusters. The notion of Strong Group (SG) is presented as a cluster definition where no two genes are distant from each other, greater than a threshold value. The genes are then ranked on their frequency of co-occurrence. The grouping and rankings are drawn by applying set operations over results of multiple distance metrics, each capturing particular similarities such as shifted relationships, negative correlations and strong positive relationships. The effectiveness of the algorithm is demonstrated on two case studies. In the first one, a single yeast cell cycle dataset is used. It is shown that different combination of set operations reveals different kinds of interactions between genes. Finally, to provide further analysis on our techniques, we have tested them on multiple microarray datasets obtained from Stanford Microarray Database.

PREDICTING THE BINDING AFFINITY OF MHC CLASS II PEPTIDES

MHC (Major Histocompatibility Complex) proteins are categorized under the heterodimeric integral membrane proteins. The MHC molecules are divided into 2 subclasses, class I and class II. Two classes differ from each other in size of their binding pockets. Predicting the affinity of these peptides is important for vaccine design. It is also vital for understanding the roles of immune system in various diseases. Due to the variability of the locations of the class II peptide binding cores, predicting the affinity of these peptides is difficult. In this paper, we proposed a new method for predicting the affinity of the MHC Class II binding peptides based on their sequences. Our method classifies peptides as binding and non-binding. Our prediction method is based on a 3-step algorithm. In the first step we identify the informative n-grams based on their frequencies for both classes. In the next step, the alphabet size is reduced. At the last step, by utilizing the informative n-grams, the class of a given sequence is predicted. We have tested our method on the MHC Bench IV-b data set [13], and compared with various other methods in the literature.

Radar fingerprint extraction via variational mode decomposition

In iMs paper, a novel method for extracting radar fingerprint using the unintentional modulation on radar signals is proposed. Proposed technique decomposes the unintentional modulations into its components using Variational Mode Decomposition (VMD) technique. Then, features that characterize each component are calculated. Simulations using real radar data show that proposed technique can classify radars in the dataset with high performance.