Jongwook Woo

Prediction of reversibly oxidized protein cysteine thiols using protein structure properties.

Protein cysteine thiols can be divided into four groups based on their reactivities: those that form permanent structural disulfide bonds, those that coordinate with metals, those that remain in the reduced state, and those that are susceptible to reversible oxidation. Physicochemical parameters of oxidation‐susceptible protein thiols were organized into a database named the Balanced Oxidation Susceptible Cysteine Thiol Database (BALOSCTdb). BALOSCTdb contains 161 cysteine thiols that undergo reversible oxidation and 161 cysteine thiols that are not susceptible to oxidation. Each cysteine was represented by a set of 12 parameters, one of which was a label (1/0) to indicate whether its thiol moiety is susceptible to oxidation. A computer program (the C4.5 decision tree classifier re‐implemented as the J48 classifier) segregated cysteines into oxidation‐susceptible and oxidation‐non‐susceptible classes. The classifier selected three parameters critical for prediction of thiol oxidation susceptibility: (1) distance to the nearest cysteine sulfur atom, (2) solvent accessibility, and (3) pKa. The classifier was optimized to correctly predict 136 of the 161 cysteine thiols susceptible to oxidation. Leave‐one‐out cross‐validation analysis showed that the percent of correctly classified cysteines was 80.1% and that 16.1% of the oxidation‐susceptible cysteine thiols were incorrectly classified. The algorithm developed from these parameters, named the Cysteine Oxidation Prediction Algorithm (COPA), is presented here. COPA prediction of oxidation‐susceptible sites can be utilized to locate protein cysteines susceptible to redox‐mediated regulation and identify possible enzyme catalytic sites with reactive cysteine thiols.

Flow-sensitive alias analysis with referred-set representation for Java

Alias analysis refers to the determination of objects that may be accessed by two or more names. Many alias analysis algorithms have been proposed for several programming languages. In this paper, we propose a flow-sensitive alias analysis algorithm for Java, an object-oriented language. This algorithm is more efficient and precise than previous algorithms for C++, another object-oriented language. For the efficiency, we define a referred-set representation of an alias that is proper for Java, while the conventional representations in C++ cause an inefficient analysis for Java. We also present a data-flow equation based on propagation rules for the referred-set. The equation computes alias information more efficiently and precisely by removing redundant alias information at a call statement. Finally, we propose our algorithm that uses an iterative looping method for an alias analysis with a structural traverse of a CFG to improve its efficiency.

Alias analysis in Java with reference-set representation for high-performance computing

In this paper, a flow-sensitive, context-insensitive alias analysis in Java is proposed. It is more efficient and precise than previous analyses for C++, and it does not negatively affect the safety of aliased references. To this end, we first present a reference-set alias representation. Second, data-flow equations based on the propagation rules for the reference-set alias representation are introduced. The equations compute alias information more efficiently and precisely than previous analyses for C++. Third, for the constant time complexity of the type determination, a type table is introduced with reference variables and all possible types for each reference variable. Fourth, an alias analysis algorithm is proposed, which uses a popular iterative loop method for an alias analysis. Finally, running times of benchmark codes are compared for reference-set and existing object-pair representation.

Scalable, incremental learning with MapReduce parallelization for cell detection in high-resolution 3D microscopy data

Accurate estimation of neuronal count and distribution is central to the understanding of the organization and layout of cortical maps in the brain, and changes in the cell population induced by brain disorders. High-throughput 3D microscopy techniques such as Knife-Edge Scanning Microscopy (KESM) are enabling whole-brain survey of neuronal distributions. Data from such techniques pose serious challenges to quantitative analysis due to the massive, growing, and sparsely labeled nature of the data. In this paper, we present a scalable, incremental learning algorithm for cell body detection that can address these issues. Our algorithm is computationally efficient (linear mapping, non-iterative) and does not require retraining (unlike gradient-based approaches) or retention of old raw data (unlike instance-based learning). We tested our algorithm on our rat brain Nissl data set, showing superior performance compared to an artificial neural network-based benchmark, and also demonstrated robust performance in a scenario where the data set is rapidly growing in size. Our algorithm is also highly parallelizable due to its incremental nature, and we demonstrated this empirically using a MapReduce-based implementation of the algorithm. We expect our scalable, incremental learning approach to be widely applicable to medical imaging domains where there is a constant flux of new data.

Market Basket Analysis algorithms with MapReduce

The MapReduce approach has been popular in computing large scale data since Google implemented its platform on Google Distributed File Systems (GFS) followed by Amazon Web Service (AWS) providing the Apache Hadoop platform in inexpensive computing nodes. Map/Reduce motivates to redesign and convert the existing sequential algorithms to MapReduce as restricted parallel programming so that the paper proposes Market Basket Analysis algorithm with MapReduce as well as apriority property. Two algorithms are proposed by adapting an existing Apriori‐algorithm and building a simple algorithm that sorts data sets and converts it to (key, value) pairs to fit with MapReduce. It is executed on Amazon EC2 Map/Reduce platform. The experimental results show that the Apriori‐algorithm does not perform as well as the simple algorithm. Using the simple algorithm, the code with Map/Reduce increases the performance by adding more nodes, but at a certain point there is a bottleneck that does not allow further performance gain. It is believed that the operations of distributing, aggregating, and reducing data in Map/Reduce, cause the bottleneck. WIREs Data Mining Knowl Discov 2013, 3:445–452. doi: 10.1002/widm.1107

Load balancing strategies for symbolic vision computations

Most intermediate and high-level vision algorithms manipulate symbolic features. A key operation in these vision algorithms is to search symbolic features satisfying certain geometric constraints. Parallelizing this symbolic search needs a non-trivial algorithmic technique due to the unpredictable workload. In this paper, we propose load balancing strategies for parallelizing symbolic search operations on distributed memory machines. By using an initial workload estimate, we first partition the computations such that the workload is distributed evenly across the processors. In addition, we perform fast migrations dynamically to adapt to the evolving workload. To demonstrate the usefulness of our load balancing strategies, experiments were conducted on an IBM SP2 and a Cray T3D. Our results show that our task migration strategy can balance the unpredictable workload with little overhead. Our code using C and MPI is portable onto other high performance computing platforms.

MapReduce Example with HBase for Association Rule

The paper illustrates how to store and compute association sets of Big Transaction Data using Hadoop and HBase and then, shows the experimental result of a MapReduce algorithm using HBase to find out association in transaction data, which is a Market Basket Analysis algorithm of Association Rule in Business Intelligence. The algorithm sorts and converts the transaction data of HBase to data set with (key, value) pair, and stores the associated data to the HBase. The algorithm and HBase run on Amazon EC2 service using Apache Whirr. The experimental results show that the algorithm increases the performance as adding more nodes till a certain number of transaction data. However, it loses control and connection when there are too many IOs with more than 3.5 millions of transaction data in HBase.

Traversal pattern mining in web environment

There have been researches about analyzing the information retrieval patterns of log file to obtain users’ information search patterns in web environment. Algorithms that find the frequently traversed path pattern from search path inputs are suggested mainly. But one of the existing works’ problems is to provide inadequate solution for complex, that is, general topological patterns. This paper suggests an efficient algorithm for deriving the maximal frequent traversal pattern from general paths.