Genome-wide analysis and classification of focal deletions in cancer.

<p>Genes that had 15 or more focal deletions (4,823) were analyzed by the unsupervised learning method of Principal Component Analysis (PCA) and classified by two supervised learning methods, Support Vector Machine (SVM) and Random Forest (RF). (A) Graph of the 4,826 genes based on their values for the first three principal components derived from the seven deletion properties that distinguish common fragile site genes from tumor suppressor genes. Known common fragile site genes are shown in blue, known tumor suppressors are shown in red, and all others are shown in green. Most known tumor suppressors are buried within the major group of genes, whereas the known common fragile site genes, with one exception (<i>LARGE</i>), are clearly separated from the major group of genes by their first and second principal component values. There are only a handful of other genes that are located near the known common fragile site genes. (B) Graphical representation of the classification of the 4,826 genes based on their probability of being common fragile site genes by Random Forest (<i>y</i>-axis) or Support Vector Machine (<i>x</i>-axis) classifiers. Both classifiers were trained on the 35 known CFS genes or tumor suppressors (<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0066264#pone.0066264.s005" target="_blank">Table S4</a>). The probability of being a tumor suppressor is one minus the CFS probability. Most genes are clustered together in the lower left quadrant; therefore, both classifiers predict with high probability that most of these genes are tumor suppressors.