Figure 4

Screenshots of the input and output interfaces of CAGER. (A). The input consists of four steps. First, the user provides ORF identifies or promoter sequences for positive and negative genes. Second, the user selects a type of features or some of their combinations. Third, the user can change parameters for decision tree learning. The user then provides an email address for notification of results and finally submits the job. (B). On the top of the output page is a graphical representation of a decision tree. Each oval of the decision tree represents an internal node and each box represents a leaf node. The text inside an internal node is the name of a feature, and the text associated with an edge is a test of the feature. The text inside a leaf node gives the predicted label (p: positive, n: negative) for genes inside the leaf, and the number of supporting and counter-instances, if any. A path from the root to a leaf node defines a possible regulatory rule. For example, the rightmost path can be read as "if the binding affinity of Mbpl to a gene's promoter is at least 1.75, the gene is positive (i.e., up-regulated under the condition that the decision tree models)." The numbers "16/1" enclosed in parenthesis means that 16 training instances have feature Mbpl_bind ≥ 1.75, of which 15 are positive and one is negative. On the bottom of the output page are related statistics for training and cross-validation of the decision tree model.