IBIS (Integrated Bayesian Inference System) is a system that is able to predict class membership for a gene expression dataset containing measurements for the same phenomenon as the dataset used to train the IBIS classifier. One of the major strengths of the IBIS method is its ability to reveal nonlinear and non-monotonic associations between pairs of genes and their concerted response to a particular stimulus such as a drug. Three types of classifiers are available in GeneLinkerô: Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), and Uniform/Gaussian Discriminant Analysis (UGDA). Different classifiers predict different responses to a stimulus for a gene or pair of genes. Each prediction has an associated accuracy percentage and an MSE value.
The concept that gene expression levels for a single gene can be used to predict stimulus response in every case is quite primitive. Although LDA classifiers are able to capture this relationship, there are certainly associations in which the response to a particular stimulus fluctuates as a function of the products of multiple genes. QDA and UGDA classifiers are able to uncover such associations.
IBIS requires a complete dataset with an associated variable. The variable must contain more than one class value with at least three observations each (meaning the dataset must have at least six samples). Also, the variable cannot include the class 'unknown'. Generating IBIS classifiers can be time and resource intensive, so filtering to remove genes of no interest first is recommended.
LDA can be used to discover linear association between pairs of genes.
QDA can be used to discover non-linear associations between pairs of genes.
UGDA can be used to discover nonlinear, non-monotonic associations between pairs of genes.
In general, it is best to start by creating classifiers using LDA and single genes. Only if the accuracy and MSE values are unsatisfactory should you try QDA/UGDA as well as gene pairs.
If you do not have a specific gene or gene pair in mind, the first step is to search the dataset for a gene or gene pair that would act as a good classifier. The IBIS Search process does this generating a set of proto-classifiers with accuracy and MSE statistics. The results of this process can be viewed in the IBIS Search Results Viewer and in the Classifier Gradient Plot.
Next, create a classifier from one of the proto-classifiers or using the gene or gene pair that is of particular interest to you. The results of this step can be visualized in the Classifier Gradient Plot.
Finally a dataset can be classified using the IBIS classifier and the results of that classification can be visualized in the Classification Plot or in the Classifier Gradient Plot.
Create IBIS Classifier From IBIS Search Results
Create IBIS Classifier Using a Gene or Gene Pair