The program selects most correlated genes for specified gene set.

Program is provided with viewer.

Data specification

The expression data for the set of genes is represented as a table, consisting of rows (usually corresponding to genes) and columns (or fields, usually corresponding to samples/tissues/experiments). Each row corresponds to expression measurements for the gene. Columns correspond to experiments/samples/tissues. However, this table may include not only expression data, but also other information related to genes, for example gene names, classifiers, etc. Therefore we will call the table columns as 'fields' in general case. In general, columns of the table could be of four basic types:
IVALUE signed integer value;
FVALUE floating point value;
WORD text without spaces inside (single word);
STRING text with spaces inside allowed.

Fields are completely defined by their basic types and names.

SelTag Input file basic format

Basic input file format should be as follows:


; May contain comment starting from the semicolon in any line of the file
NAME<tab>WORD
GENEID<tab>IVALUE
TISSUECANCER0<tab>FVALUE
TISSUECANCER1<tab>FVALUE
TISSUENORMAL0<tab>FVALUE
TISSUENORMAL1<tab>FVALUE
TISSUENORMAL2<tab>FVALUE
#GROUP<tab>Cancer tissues
TISSUECANCER0
TISSUECANCER1
#ENDGROUP
#GROUP<tab>Arbitrary group
TISSUECANCER1
TISSUECANCER2
TISSUENORMAL0
TISSUENORMAL1
#ENDGROUP
END
DATA
GENE04675<tab>402<tab>6.00<tab>5.60<tab>5.97<tab>6.00<tab>6.00
GENE46890<tab>794<tab>2.77<tab>3.22<tab>5.65<tab>5.68<tab>5.68
GENE23794<tab>404<tab>5.97<tab>5.97<tab>6.00<tab>5.60<tab>5.97

In this example <tab> implies 'Tab' character symbol.

First lines (up to the "DATA" line) contain data format description. In this part of the file each line describes field description: field name and field basic type.

After the "DATA" line - data on each gene are represented. Each line correspond single cards. Field data are separated by 'tab' symbol. Double 'tab' is interpreted as missed data.

It is assumed in SetTag program that the expression data in the file are normalized and the expression levels of genes in experiments are comparable.

Selection files

MolQuest version of the SelTag program can also operates with other types of files, namely, selection files. These files contain information about some selected genes or samples from the large data file in SelTag format. The selection file contain: the data file name from which selection was obtained; type of selection data (genes of samples), list of selected objects (their indices in the large data file). The selection files are in the XML format. Two examples are below.

Selection for some genes.


<?xml version="1.0" encoding="ISO-8859-5"?>
<SELECTION>
	<HEADER name="cc_Selection5">
		<DATA source="c:/data/cc.txt"/>
		<COMMENT><![CDATA["$F1 == "GEN14263" | $F12 >= 300"]]></COMMENT>
	</HEADER>
	<ELEMENTS type="GENES" count="9">
	<![CDATA[0;1;2;10;14;15;17;26;30]]>
	</ELEMENTS>
</SELECTION>

Selection for some fields (samples).


<?xml version="1.0" encoding="ISO-8859-5"?>
<SELECTION>
	<HEADER name="notterman2001_set1">
		<DATA source="c:/data/notterman2001_set1.txt"/>
		<COMMENT><![CDATA["From cc.txt data file."]]></COMMENT>
	</HEADER>
	<ELEMENTS type="FIELDS" count="10">
	  <![CDATA[0;1;2;3;5;6;7;18;19;30]]>
	</ELEMENTS>
</SELECTION>

Selection files may be selected during the SelTag execution and also used by SelTag for calculation and/or visualization. Note, each selection file is linked to large data file by its name. Selection data cannot be applied to another data file.

Algorithm

The SelTag:SelCorr program allows selecting genes which have expression profiles highly correlated to the profile of the user-defined gene(s).
User should provide list of fields to calculate correlation.

Three types of correlation are possible:
Pearson's r - Pearson's correlation coefficient. The Pearson product moment correlation coefficient between expression profiles i and j is calculated as follows:

where yki is the expression level of gene i in the experiment k; is the mean expression level of the gene i. Positive correlation implies that the expression levels of genes i,j are related positively, the higher expression of gene i, the higher expression of gene j. Negative correlation means that the expression levels of genes i,j are related negatively, the higher expression of gene i, the lower expression of gene j. If the rij is close to zero, two expression profiles are unrelated.

Spearman r - Spearman's correlation coefficient.
This correlation coefficient is computed for ranks. Let Rki is the rank of the expression level in the experiment k of gene i (relatively to other experiments), Rkj is the rank of the expression level in the experiment k of gene j. Then Spearman's correlation coefficient is calculated by the formula

Kendall's t - Kendall's tau correlation coefficient. To calculate Kendall's t K for data points (yki; ykj) 2K(K - 1) pairs considered (without self-pairing, the points in either order count as one pair). Pairs in which yki > ymi and ykj > ymj or yki < ymi and ykj < ymj are called concordant pairs (agreement between ranks), pairs with rank disagreement are called discordant pairs. In general, t is calculated as
t = ([number of concordant] - [number of discordant]) / total number of pairs

For the specified gene user can select other genes that have correlation coefficient between target gene expression profile greater than threshold. There are several threshold types: "Best N" - select N most correlated genes from set; "Best %" - select a fraction (in %) of most correlated genes from set; "Value" - select the genes with the absolute correlation value equal or higher than the threshold; "All" - select all genes from list.

If a number of genes are selected in target list, several options exist how to treat the correlation of profile with this groups of profiles: "Max. correlation value to select" - when comparing genes, the key parameter is the maximum coefficient of correlation of a gene from Set 1 with genes from Set 2; "Aver. correlation value to select" - when comparing genes from Set 1, the key parameter is the average coefficient of the correlation of a gene from Set 1 with genes from Set 2; "Corr. for aver. field values to select" - when comparing genes from Set 1, the key parameter is the coefficient of correlation of a gene from Set 2 with an average profile of genes from Set 2. This means that the program creates an "imaginary" average gene from Set 2 and uses this average value to calculate the correlation coefficient.

Example of the output data


status=Correlation matrix for cards...
status=Correlation matrix calculation...
status=done [0.0 sec]
List of selected genes [30 total]:
1	6718	X54232
2	4575	R81175
3	7132	X79981
4	5493	T78432
5	3454	R06627
6	5166	T59895
7	6042	U14394
8	6690	X52947

Some lines starting from "status=" just output the status of the calculation and can be ignored. Then the result information (with the number of selected genes) is output. Then list of selected genes with their indices in data file and gene names are printed out.