detect tissue-specific (or tissue-selective) patterns from microarray data with many kinds of samples

Share:

Description

ROKU is a method for detecting tissue-specific (or tissue-selective) patterns from gene expression data for many tissues (or samples). ROKU (i) ranks genes according to their overall tissue-specificity using Shannon entropy after data processing and (ii) detects tissues specific to each gene if any exist using an Akaike's information criterion (AIC) procedure.

Usage

1
ROKU(data, upper.limit = 0.25, sort = FALSE)

Arguments

data

numeric matrix or data frame containing microarray data (on log2 scale), where each row indicates the gene or probeset ID, each column indicates the tissue, and each cell indicates a (log2-transformed) expression value of the gene in the tissue. Numeric vector can also be accepted for a single gene expression vector.

upper.limit

numeric value (between 0 and 1) specifying the maximum percentage of tissues (or samples) as outliers to each gene.

sort

logical. If TRUE, results are sorted in descending order of the entropy scores.

Details

As shown in Figure 1 in the original study of ROKU (Kadota et al., 2006), Shannon entropy H of a gene expression vector (x_{1}, x_{2}, ..., x_{N}) for N tissues can range from zero to log_{2}N, with the value 0 for genes expressed in a single tissue and log_{2}N for genes expressed uniformly in all the tissues. Researchers therefore rely on the low entropy score for the identification of tissue-specific patterns. However, direct calculation of the entropy for raw gene expression vector works well only for detecting tissue-specific patterns when over-expressed in a small number of tissues but unexpressed or slightly expressed in others: The H scores of tissue-specific patterns such as (8,8,2,8,8,8,8,8,8,8) for the 3rd tissue-specific down-regulation (see the Figure 1e) are close to the maximum value (log_{2}N=3.32 when N=10) and cannot identify such patterns as tissue-specific. To detect various kinds of tissue-specific patterns by low entropy score, ROKU processes the original gene expression vector and makes a new vector (x_{1'}, x_{2'}, ..., x_{N'}). The data processing is done by subtracting the one-step Tukey biweight and by taking the absolute value. In case of the above example, ROKU calculates the H score from the processed vector (0,0,6,0,0,0,0,0,0,0), giving very low score (from H = 3.26 before processing to H' = 0 after processing). A major characteristic of ROKU is, therefore, to be able to rank various tissue-specific patterns by using the modified entropy scores.

Note that the modified entropy does not explain to which tissue a gene is specific, only measuring the degree of overall tissue specificity of the gene. ROKU employs an AIC-based outlier detection method (Ueda, 1996). Consider, for example, a hypothetical mixed-type of tissue-selective expression pattern (1.2, 5.1, 5.2, 5.4, 5.7, 5.9, 6.0, 6.3, 8.5, 8.8) where we imagine a total of three tissues are specific (down-regulated in tissue1; up-regulated in tissues 9 and 10). The method first normalize the expression values by subtracting the mean and dividing by the standard deviation (i.e., z-score transformation), then sorted in order of increasing magnitude by
(-2.221, -0.342, -0.294, -0.198, -0.053, 0.043, 0.092, 0.236, 1.296, 1.441). The method evaluates various combinations of outlier candidates starting from both sides of the values: model1 for non-outlier, model2 for one outlier for high-side, model3 for two outliers for high-side, ..., modelx for one outlier for down-side, ..., modely for two outliers for both up- and down sides, and so on. Then, it calculates AIC-like statistic (called U) for each combination of model and search the best combination that achieves the lowest U value and is termed the minimum AIC estimate (MAICE). Since the upper.limit value corresponds to the maximum number of the outlier candidates, it decides the number of combinations. The AIC-based method output a vector (1 for up-regulated outliers, -1 for down-regulated outliers, and 0 for non-outliers) that corresponds to the input vector. For example, the method outputs a vector (-1, 0, 0, 0, 0, 0, 0, 0, 1, 1) when using upper.limit = 0.5 and (-1, 0, 0, 0, 0, 0, 0, 0, 0, 0) when using upper.limit = 0.25 (as default). See the Kadota et al., 2007 for detailed discussion about the effect of different parameter settings.

Value

A list containing following fields:

outlier

A numeric matrix when the input data are data frame or matrix. A numeric vector when the input data are numeric vector. Both matrix or vector consist of 1, -1, and 0: 1 for over-expressed outliers, -1 for under-expressed outliers, and 0 for non-outliers.

H

A numeric vector when the input data are data frame or matrix. A numeric scalar when the input data are numeric vector. Both vector or scalar consist of original entropy (H) score(s) calculated from an original gene expression vector.

modH

A numeric vector when the input data are data frame or matrix. A numeric scalar when the input data are numeric vector. Both vector or scalar consist of modified entropy (H') score(s) calculated from a processed gene expression vector.

rank

A numeric vector or scalar consisting of the rank(s) of modH.

Tbw

a numeric vector or scalar consisting of one-step Tukey's biweight as an iteratively reweighted measure of central tendency. This value is in general similar to median value and the same as the output of tukey.biweight with default parameter settings in affy package. The data processing is done by subtracting this value for each gene expression vector and by taking the absolute value.

References

Kadota K, Konishi T, Shimizu K: Evaluation of two outlier-detection-based methods for detecting tissue-selective genes from microarray data. Gene Regulation and Systems Biology 2007, 1: 9-15.

Kadota K, Ye J, Nakai Y, Terada T, Shimizu K: ROKU: a novel method for identification of tissue-specific genes. BMC Bioinformatics 2006, 7: 294.

Kadota K, Nishimura SI, Bono H, Nakamura S, Hayashizaki Y, Okazaki Y, Takahashi K: Detection of genes with tissue-specific expression patterns using Akaike's Information Criterion (AIC) procedure. Physiol Genomics 2003, 12: 251-259.

Ueda T. Simple method for the detection of outliers. Japanese J Appl Stat 1996, 25: 17-26.

Examples

1
2
3