Training statistical gene models based on high-density oligonucleotide tiling microarray data. (A) Large-scale signal data from tiling array experiments can be used to train statistical models to score the hits, and a small/medium proportion of these results can be further validated by experiments or other biological knowledge via runs of iterations and optimizations. (B) Different strategies can be used to select genomic regions for validation; e.g., (1) select only the regions with high signals, (2) select regions randomly, or (3) select those that have the maximum signal entropies, which usually contain “borders” of high and low signals. One question worth asking is whether an optimal way of selection exists to best help in training the statistical model.
