homeabout uscontact us


Tutorial 7: Conclusion


In general, it is best to start identifying simpler patterns in the data first. This usually means using IBIS with single genes and Linear Discriminant Analysis (LDA) to begin with. Only if the accuracy or MSE values are unsatisfactory should you try Quadratic Discriminant Analysis (QDA) and Uniform/Gaussian Discriminant Analysis (UGDA) as well as gene pairs.

Remember that single gene IBIS searches are always relatively quick, even for tens of thousands of genes. However, when looking for patterns over gene pairs, the run time will be multiplied by the number of genes in the dataset again. For instance, if running 1D IBIS took 1 minute on 500 genes, then 2D IBIS will take about 500 minutes (8 hours) on the same data. Effective filtering of genes is an important step to make gene pair searches practical.

Use the minimum standard deviation to capture your estimate of the error in the measurements. With too small a value, you will find degenerate looking patterns that are not believable. With too large a value, you risk missing important patterns due to over-smoothing the classifier.

When you are finished, you can close all the open plots either by clicking on the 'x' box in the upper-right hand corner of each, or by selecting Close All from the Window menu.


Where To Go From Here