Support vector machine apply on gene expression data

Question

[sana777munquad]

Hii, Currently I am working on Geo datasets of gene expression(cancer) and I made a matrix of sample vs genes and add class label(Normal vs tumor) for SVM. Samples are in the rows and genes in the coloumn and the last coluomn is Class label.

Is this correct matrix for Support vector machine.??

[shoujun.gu]

SVM is just a ML method. There is no so called correct matrix.

Answer 1

[Mensur Dlakic]

It depends on SVM implementation you are using. Class label is typically in first column if you are using LibSVM format.

[sana777munquad]

Thanks for your information for class label. actually I have a gene expression data of cancer which consists of 6 observations with 20000 variables(genes). I got the data from geo datasets and normalize with affy. and got 20000 about genes. how i can minimize the size of matrix?

[Mensur Dlakic]

There is no need to reduce the size of your matrix unless you feel like throwing away the data. Any SVM implementation will handle a 20000 x 6 matrix without problems.

You should scale rather than normalize data. Those two operations may lead to similar numbers, but normalization typically gives numbers that are not bound in either positive or negative direction. SVMs work best with small numbers in [-1, +1] or [0, 1] ranges. If after normalization your largest number in each column is not much bigger than 1 and the smallest not much smaller than -1, you might be OK with normalization. Still, it is just as easy to scale the data as it is to normalize, and it will take care of potential numerical issues. This page and this file explain the concept in more detail.