NIA Array Analysis Tool

NIA Array Analysis Tool

Normalization of Input File

Our software can normalize data coming from 1-color arrays (e.g., with radioactive label). Although the same normalization method can be used for 2-color arrays, we never tested it with 2-color arrays. Most companies that produce 2-color arrays also provide software for image analysis and data normalization. It is better to use commercial software for normalization of 2-color arrays than our tool.

Our normalization tool implements the non-parametric method that equalizes multiple quantiles of the probability distribution of gene expression.
Step1: log-transform all the data
Step2: in each column estimate 15 quantiles that correspond to ratios: 1/30, 3/30, ... 29/30.
Step3: estimate 15 target quantiles as average quantiles across all columns.
Step4: transform data using a piece-linear function that converts actual quantiles in each column into target quantiles. Data above the highest quantile is transformed based on the linear function between two highest quantiles.
Step5: back-transform data with exponent function.

In some cases original data has non-uniform distribution at the lower end; thus estimated quantiles may be not reliable. Thus, for each column we determine the lowest reliable quantile based on the following condition: the quantile is reliable if the difference between it and the next higher quantile is not >2 times greater and not <2 times smaller than between corresponding target quantiles. Then transformation of data below the smallest reliable quantile is based on the linear function for this quantile.

Split-Normalization of Input File

Split-normalization means that mormalization is done for a subset of genes at a time. The major reason for using split-normalization is that all genes do not fit on one array. In this case the full set of genes may be represented by 2 or more arrays that require independent normalization. For split-normalization the input file must have array indexes in the second column as shown below:

Example of an input file ready for split-normalization

Geneid	Array index	Control rep1	Control rep2	Control rep3	Treatment1 rep1	Treatment1 rep2	Treatment1 rep3	Treatment2 rep1	Treatment2 rep2	Treatment2 rep3
1		1	180	314	296	433	182	311	397	566	361
2		1	6780	17085	9223	18468	8623	15019	17588	24026	21732
3		1	16592	15161	10476	16790	9752	10316	13885	19448	14564
4		1	2896	239	101	59	29	53	185	198	124
5		2	5496	4283	6635	4912	4459	3175	8050	5973	6297
6		2	9708	18958	7701	19469	5279	13767	9876	23492	15672
7		2	5342	1548	2494	1222	1490	891	2566	1438	3155

If you suspect a gradient of background (e.g. from left to right) on one array, then you can try splitting this array into several portions (e.g., vertical sections) and do split-normalization. However, if you have non-uniform background we suggest you using other software for normalization.