BINARY PREDICTION TREE MODELING WITH MANY PREDICTORS AND ITS USES IN CLINICAL AND GENOMIC APPLICATIONS

摘要

The tree-model that statistical analysis technique is prediction statistics is provided. The model first screen gene is to reduce noise, using relevant k- mean clusters, then using singular value decomposition to extract single remarkable indicator (principal component) from each cluster. This generates factor single derived from statistically significant amount of cluster, we are known as metagenes, and the expression of characterization multiple pattern gene passes through sample. The strategy, which is intended to extract a variety of sizes for but reducing this pattern and eliminates gene-specific noises, passes through the cluster in set. With the Bayes's classification tree analysis of the formal forecast analysis of these metagenes. This generates multiple recursive subdivision samples to subgroup (' retain ' tree), and be associated with Bayes prediction result probability and each subgroup. Always predict then individual sample passes through consensus forecast, using appropriate weight, in many such tree-models. The model includes going out sample mutually using iteration and confirming prediction to carry out repacking model, actual prediction and the mirror image prediction that new case occurs within a context, because they are main targets.