For diseases, we used the Gene Expression Omnibus (GEO) repository to generate microarray data for disease samples and control samples.
The first approach prioritized the genes based on their differential expression behavior in cells treated with the drug or in disease samples compared to normal samples.
In many cancer studies, a gene may be expressed in some but not all of the disease samples, reflecting the complexity of the underlying disease.
Surprisingly FTD and PiD disease samples still clustered as a single large group.
All disease samples were obtained from University of California, San Francisco.
However, the patterns were dominated by CA-125 and had poor performance when evaluated on early stage disease samples.
However when new Alzheimer's disease samples were also included, they did not cluster together with matching samples from the previous experiment, see Figure 2C.
In general, outlier analysis offers a unique and powerful approach for the identification of key pathogenetic genes involved in a subset of disease samples.
The data from the different disease group also showed 100% 955% CI: 98.5 100) specificity as none of the 250 non-leishmanial disease samples were PCR positive.
A gene showing over-expression in a subgroup of disease samples based on a cut-off threshold is defined as an outlier (Figure 1).
When the proportion of outliers among the disease samples becomes very small, all existing methods become increasingly insensitive to the presence of small outlier populations.
