The advent of genomic microarray technology enables simultaneously measuring the expressions of thousands of genes in massive experiments, and hence provides scientists, for the first time, the opportunity of observing complex relationships between various genes in a genome. In order to extract biologically meaningful insights from a plethora of data generated from microarray experiments, knowledge discovery techniques, which discover patterns, statistical or predictive models, and relationships among massive data, have been widely applied in microarray data analysis. For example, clustering can be applied to identify groups of genes that are regulated in a similar manner under a number of experimental conditions or groups of samples that show similar expression patterns across a number of genes (Jiang, Tang, & Zhang, 2004). Classification can be performed to characterize the cellular difference between different samples, such as between normal and cancer cells or between cancer cells with different responses to treatment, and can potentially be used to predict the classes of samples based on their gene expression patterns (Statnikov, Aliferis, Tsamardinos, Hardin, & Levy, 2005). Feature selection or gene selection can help identify among thousands of genes a small fraction of genes that are relevant for discriminating between different sample types, and may potentially lead to the identification of a few biologically relevant “marker” genes for subsequent biological validation (Saeys, Inza, & Larranaga, 2007). This article provides a brief introduction to the field of knowledge discovery and its applications in discovering useful knowledge from genomic microarray data. It describes common knowledge discovery tasks for genomic microarray data, presents representative methods for each task, and identifies emerging challenges and trends in knowledge discovery from genomic microarray data.
Knowledge discovery from data refers to the overall process of converting raw data into useful information, which consists of data preprocessing, data mining, and postprocessing of data mining results (Tan, Steinbach, & Kumar, 2005). The purpose of data preprocessing is to transform the raw input data into an appropriate format for subsequent data mining process. The tasks involved in data preprocessing include cleaning data to remove noise, duplicate, inconsistent, or missing information, integrating data from multiple sources, transforming data values into the right scale and format, and selecting instances and features that are relevant to the data mining task at hand. Data mining is the automatic process of extracting interesting patterns or knowledge from data. Data mining tasks can be generally divided into two major categories: predictive tasks and descriptive tasks. The objective of predictive tasks is to predict the values of a particular feature, based on the values of other features. The objective of descriptive tasks is to derive patterns that summarize the underlying relationships in data. These tasks are often exploratory in nature and frequently require postprocessing techniques which validate and explain the results. For example, visualizing the patterns allows analysts to explore the result from multiple viewpoints. Statistical tests can be used to validate the significance of the results and eliminate patterns that are generated by chance. For a comprehensive discussion on various knowledge discovery tasks, please refer to widely adopted text books on data mining (Han & Kamber, 2005; Tan et al., 2005). This article focuses on knowledge discovery tasks that are commonly performed on genomic microarray data introduced next.
Key Terms in this Chapter
Class ification: A process of predicting the classes of unseen instances based on patterns learned from available instances with predefined classes.
Genome: All of the genetic information or hereditary material possessed by an organism.
Knowledge Discovery: The overall process of converting raw data into useful information, consisting of data preprocessing, data mining, and postprocessing of data mining results.
Clustering: A process of grouping instances into clusters so that instances are similar to one another within a cluster but dissimilar to instances in other clusters.
Feature Selection: A process of choosing an optimal subset of features from original features according to a certain criterion.
Gene Expression Microarrays: Silicon chips that simultaneously measure the expression levels of thousands of genes.
Gene: A hereditary unit consisting of a sequence of DNA that contains all information necessary to produce a molecule that performs some biological function.
Data Mining: the automatic process of extracting interesting patterns or knowledge from data.