We study a data mining problem concerning the elastic peak detection in 2D liquid chromatography-mass spectrometry (LC-MS) data. These data can be modeled as time series, in which the X-axis represents time points and the Y-axis represents intensity values. A peak occurs in a set of 2D LC-MS data when the sum of the intensity values in a sliding time window exceeds a user-determined threshold. The elastic peak detection problem is to locate all peaks across multiple window sizes of interest in the data set. We propose a new data structure, called a Shifted Aggregation Tree or AggTree for short, and use the data structure to find the different peaks. Our method, called PeakID, solves the elastic peak detection problem in 2D LC-MS data yielding neither false positives nor false negatives. The method works by first constructing an AggTree in a bottom-up manner from the given data set, and then searching the AggTree for the peaks in a top-down manner. We describe a state-space algorithm for finding the topology and structure of an efficient AggTree to be used by PeakID. Our experimental results demonstrate the superiority of the proposed method over other methods on both synthetic and real-world data.

We study a new data mining problem concerning the discovery of frequent agreement subtrees (FASTs) from a set of phylogenetic trees. A phylogenetic tree, or phylogeny, is an unordered tree in which the order among siblings is unimportant. Furthermore, each leaf in the tree has a label representing a taxon (species or organism) name, whereas internal nodes are unlabeled. The tree may have a root, representing the common ancestor of all species in the tree, or may be unrooted. An unrooted phylogeny arises due to the lack of sufficient evidence to infer a common ancestor of the taxa in the tree. The FAST problem addressed here is a natural extension of the maximum agreement subtree (MAST) problem widely studied in the computational phylogenetics community. The paper establishes a framework for tackling the FAST problem for both rooted and unrooted phylogenetic trees using data mining techniques. We first develop a novel canonical form for rooted trees together with a phylogeny-aware tree expansion scheme for generating candidate subtrees level by level. Then, we present an efficient algorithm to find all FASTs in a given set of rooted trees, through an Apriori-like approach. We show the correctness and completeness of the proposed method. Finally, we discuss the extensions of the techniques to unrooted trees. Experimental results demonstrate that the proposed methods work well, and are capable of finding interesting patterns in both synthetic data and real phylogenetic trees.

We present a new data mining problem to discover frequent agreement subtree patterns from a database of rooted phylogenetic trees. This problem is a natural extension of the traditional MAST (maximum agreement subtree) problem. To solve the problem, we first present a novel canonical form for leaf-labeled trees and an efficient tree expansion algorithm for generating candidate subtrees level by level. We then show how to efficiently discover all frequent agreement subtrees from a given set of phylogenetic trees, through an Apriori-like data mining approach. We discuss the correctness and completeness of the proposed method. Experimental results demonstrate that the proposed method can discover interesting patterns from different phylogenetic trees for multiple species. The algorithms were implemented in C++ and integrated into an online toolkit, which is fully operational and accessible on the World Wide Web.

Currently, phylogenetic tree techniques are being used in multiple areas, from Tree of Life problems to pathogen recognition to drug discovery. With all of these applications for phylogenetic tree techniques, methods are needed to exploit the knowledge modeled in phylogenetic trees more thoroughly. One such information point of interest is the behavior of frequent patterns in phylogenetic trees. While there are many techniques that look at maximal, consensus and supertreepatterns, there are few techniques that look at frequent, but not maximal pattern. This demonstration paper presents PhyloMiner, a tool that automatically discovers frequent agreement subtrees from multiple phylogenies. It introduces this topic of frequent agreement subtrees and then concludes with describing the PhyloMiner tool that implements these concepts and is available freely on the World Wide Web.

Frequent structure mining (FSM) aims to discover and extract patterns frequently occurring in structural data, such as trees and graphs. FSM finds many applications in bioinformatics, XML processing, Web log analysis, and so on. We present a new FSM technique for finding patterns in rooted unordered labeled trees. The patterns of interest are cousin pairs in these trees. A cousin pair is a pair of nodes sharing the same parent, the same grandparent, or the same great-grandparent, etc. Given a tree T, our algorithm finds all interesting cousin pairs of T in O(|T|/sup 2/) time where |T| is the number of nodes in T. Experimental results on synthetic data and phylogenies show the scalability and effectiveness of the proposed technique. To demonstrate the usefulness of our approach, we discuss its applications to locating co-occurring patterns in multiple evolutionary trees, evaluating the consensus of equally parsimonious trees, and finding kernel trees of groups of phylogenies. We also describe extensions of our algorithms for undirected acyclic graphs (or free trees).

Phylogenetic trees are unordered labeled trees in which each leaf node has a label and the order among siblings is unimportant. In this paper we propose a new similarity measure, called TreeRank, for phylogenetic trees and present an algorithm for computing TreeRank scores. Given a query or pattern tree P and a data tree D, the TreeRank score from P to D is a measure of the topological relationships in P that are found to be the same or similar in D. The proposed algorithm calculates the TreeRank score in O(M/sup 2/ + N) time where M is the number of nodes appearing in both P and D, and N is the number of nodes in D. We then develop a search engine that, given a query or pattern tree P and a database of trees D, finds and ranks the nearest neighbors of P in D where the "nearness" is measured by the proposed similarity function. This structure-based search engine is fully operational and is available on the World Wide Web.

In this paper we present a new system, called GeneScout, for predicting gene structures in vertebrate genomic DNA. The system contains specially designed hidden Markov models (HMMs) for detecting functional sites including protein-translation start sites, mRNA splicing junction donor and acceptor sites, etc. Our main hypothesis is that, given a vertebrate genomic DNA sequence S, it is always possible to construct a directed acyclic graph G such that the path for the actual coding region of S is in the set of all paths on G. Thus, the gene detection problem is reduced to that of analyzing the paths in the graph G. A dynamic programming algorithm is used to find the optimal path in G. The proposed system is trained using an expectation-maximization (EM) algorithm and its performance on vertebrate gene prediction is evaluated using the 10-way cross-validation method. Experimental results show the good performance of the proposed system and its complementarity to a widely used gene detection system.

Modern search engines answer keyword-based queries extremely efficiently. The impressive speed is due to clever inverted index structures, caching, a domain-independent knowledge of strings, and thousands of machines. Several research efforts have attempted to generalize keyword search to keytree and keygraph searching, because trees and graphs have many applications in next-generation database systems. This paper surveys both algorithms and applications, giving some emphasis to our own work.

Phylogenetic trees are essential for understanding the relationships among organisms or taxa. Many of the current techniques for searching phylogenetic repositories allow the user to perform a keyword-type search or an aligned sequence data search, or to browse a hierarchical list of taxa. Here we describe a new search engine that allows the user to present an example phylogeny, or a query tree, and then searches a phylogenetic database for trees that contain the query structure. The presented search engine is fully operational and is available on the World Wide Web.

An unordered labeled tree is a tree in which each node has a string label and the parent-child relationship is significant, but the order among siblings is unimportant. This paper presents an approach to the nearest neighbor search problem for these trees. Given a database D of unordered labeled trees and a query tree Q, the goal is to find those trees in D that "approximately" contain Q. Our approach is based on storing the paths of the trees in a suffix array and then counting the number of mismatching paths between the query tree and a data tree. To speed up a search, we use a hash-based technique to filter out unqualified data trees at an early stage of the search. Experimental results obtained by running our techniques on phylogenetic trees and synthetic data demonstrate the good performance of the proposed approach. We also discuss the use of our work in XML and scientific database management.

Presents a method for finding patterns in 3D graphs. Each node in a graph is an undecomposable or atomic unit and has a label. Edges are links between the atomic units. Patterns are rigid substructures that may occur in a graph after allowing for an arbitrary number of whole-structure rotations and translations as well as a small number (specified by the user) of edit operations in the patterns or in the graph. (When a pattern appears in a graph only after the graph has been modified, we call that appearance "approximate occurrence.") The edit operations include relabeling a node, deleting a node and inserting a node. The proposed method is based on the geometric hashing technique, which hashes node-triplets of the graphs into a 3D table and compresses the label-triplets in the table. To demonstrate the utility of our algorithms, we discuss two applications of them in scientific data mining. First, we apply the method to locating frequently occurring motifs in two families of proteins pertaining to RNA-directed DNA polymerase and thymidylate synthase and use the motifs to classify the proteins. Then, we apply the method to clustering chemical compounds pertaining to aromatic compounds, bicyclicalkanes and photosynthesis. Experimental results indicate the good performance of our algorithms and high recall and precision rates for both classification and clustering.

Biological data mining aims to extract signi cant information from DNA, RNA and proteins.The signi cant information may refer to motifs, functional sites, clustering and classi cation rules.This paper presents an example of biological data mining: the classi cation of protein sequences using neural netw orks.We proposenew tec hniques to extract features from protein data and use them in combination with the Ba yesianneural network to classify protein sequences obtained from the PIR protein database maintained at the National Biomedical Research F oundation.T o evaluate the performance of the proposed approach, we c o mpare it with other protein classi ers built based on sequence alignment and machine learning methods.Experimental results sho w the high precision of the proposed classi er and the complementarity of the tools studied in the paper.

A distance-mapping algorithm takes a set of objects and a distance metric and then maps those objects to a Euclidean or pseudo-Euclidean space in such a way that the distances among objects are approximately preserved. Distancemapping algorithms are a useful tool for clustering and visualization in data intensive applications, because they replace expensive distance calculations by sum-of-square calculations. This can make clustering in large databases with expensive distance metrics practical. In this paper we present five distance-mapping algorithms and conduct experiments to compare their performance in data clustering applications. These include two algorithms called FastMap and MetricMap, and three hybrid heuristics that combine the two algorithms in different ways. Experimental results on both synthetic and RNA data show the superiority of the hybrid algorithms. The results imply that FastMap and MetricMap capture complementary information about distance metrics and therefore ca...

Structural matching and discovery in documents such as SGML and HTML is important for data warehousing [6], version management [7, 11], hypertext authoring, digital libraries [4] and Internet databases. As an example, a user of the World Wide Web may be interested in knowing changes in an HTML document [2, 5, 10]. Such changes can be detected by comparing the old and new version of the document (referred to as structural matching of documents). As another example, in hypertext authoring, a user may wish to find the common portions in the history list of a document or in a database of documents (referred to as structural discovery of documents). In SIGMOD 95 demo sessions, we exhibited a software package, called TreeDiff [13], for comparing two latex documents and showing their differences. Given two documents, the tool represents the documents as ordered labeled trees and finds an optimal sequence of edit operations to transform one document (tree) to the other. An edit operation could be an insert, delete, or change of a node in the trees. The tool is so named because documents are represented and compared using approximate tree matching techniques [9, 12, 14].

A knowledge-based tutoring system (KBTS) is a computer-based instructional system that uses artificial intelligence techniques to help people learn some subjects. We found that the knowledge communication process involving a KBTS and a human student can be decomposed into a series of communication cycles, where each cycle concentrates on one topic and contains four major phases: planning, discussing, evaluating and remedying. The major contributions of this work are the development of a generic architecture for supporting the knowledge communication between a KBTS and a student, and a graphical notation and schema for supporting the curriculum knowledge representation and manipulation during the planning phase of a tutoring process. The curriculum knowledge about a course can help a tutoring system determine the sequences in which the topics will be discussed with the students effectively and diagnose the students' mistakes. The curriculum knowledge base contains the goal structure of the course, prerequisite relations, and multiple ways of organizing topics, among others. As an example, we focus on developing SQL-TUTOR, a KBTS for the domain of SQL programming. This system has features such as an efficient control mechanism, explicit curriculum knowledge representation, and individualized private tutoring. For allowing the students relative freedom to decide how to study the domain knowledge about a subject, the system provides the students with a group of operators to hand-tailor the learning schedules according to their special backgrounds, requests, and interests.

Over the past several years we have created or borrowed algorithms for combinatorial pattern matching and pattern discovery on sequences [2] and trees.In matching problems, given a pattern, a set of data objects and a distance metric, we find the distance between the pattern and one or more data objects. In discovery problems by contrast, given a set of objects, a metric, and a distance, we seek a pattern that matches many of those objects within the given distance. (So, discovery is a lot like data mining.) Our toolkit performs both matching and discovery with current targeted applications in molecular biology and document comparison.

Suppose you are given a set of natural entities (e.g., proteins, organisms, weather patterns, etc.) that possess some important common externally observable properties. You also have a structural description of the entities (e.g., sequence, topological, or geometrical data) and a distance metric. Combinatorial pattern discovery is the activity of finding patterns in the structural data that might explain these common properties based on the metric.

Ordered, labeled trees are trees in which each node has a label and the left-to-right order of its children (if it has any) is fixed. Such trees have many applications in vision, pattern recognition, molecular biology, programming compilation, and natural language processing. Many of the applications involve comparing trees or retrieving/extracting information from a repository of trees. Examples include classification of unknown patterns, analysis of newly sequenced RNA structures, semantic taxonomy for dictionary definitions, generation of interpreters for nonprocedural programming languages, and automatic error recovery and correction for programming languages. Previous systems use exact matching (or generalized regular expression matching) for tree comparison. This paper presents a system, called approximate-tree-by-example (ATBE), which allows inexact matching of trees. The ATBE system interacts with the user through a simple but powerful query language; graphical devices are provided to facilitate inputing the queries. The paper describes the architecture of ATBE, illustrates its use and describes some aspects of ATBE implementation. We also discuss the underlying algorithms and provide some sample applications.>

Office information systems (OISs) are employed to support office workers in their management of information and to assist them in their daily work. In the OISs, document classification is one of the major functional capabilities. Classifying a document can be facilitated through the layout analysis of the document. A new approach to the layout analysis, called nested segmentation, is introduced. The layout relationships of components of a document are defined in terms of the adjacency of blocks. Given the adjacency of blocks, an adjacent block graph is introduced where the problem of the nested segmentation is transformed to a classic minimal cut problem for the graph. Also, an ordered labeled tree structure (L-S-Tree) is introduced to represent the segmented document for document classification.>

Consider the class of distributed database systems consisting of a set of nodes connected by a high bandwidth network. Each node consists of a processor, a random access memory, and a slower but much larger memory such as a disk. There is no shared memory among the nodes. The data are horizontally partitioned often using a hash function. Such a description characterizes many parallel or distributed database systems that have recently been proposed, both commercial and academic. We study the optimization problem that arises when the query processor must repartition the relations and intermediate results participating in a multijoin query. Using estimates of the sizes of intermediate relations, we show (1) optimum solutions for closed chain queries; (2) the NP-completeness of the optimization problem for star, tree, and general graph queries; and (3) effective heuristics for these hard cases. Our general approach and many of our results extend to other attribute partitioning schemes, for example, sort-partitioning on attributes, and to partitioned object databases.

A scheme to answer best-match queries from a file containing a collection of objects is described. A best-match query is to find the objects in the file that are closest (according to some (dis)similarity measure) to a given target. Previous work [5, 331] suggests that one can reduce the number of comparisons required to achieve the desired results using the triangle inequality, starting with a data structure for the file that reflects some precomputed intrafile distances. We generalize the technique to allow the optimum use of any given set of precomputed intrafile distances. Some empirical results are presented which illustrate the effectiveness of our scheme, and its performance relative to previous algorithms.