ShrnaAlgorithms

Use christof's metric for potent shrnas, and beat prediction correlation of 0.78 !
 * Goal**:

1. Our naive algorithm is the use Gunnar's shogun svm package with the weighted degree kernel 1.1. Add the entire 50 mer sequence. 2. We would like to extend step 1 to include early stopping for the classifier which takes into account the processing steps of the shrna in the cell.
 * Our approach:**

The following figure shows the concept, sub-window should read shrna. Each shrna goes through the first classifier, which learns only a subtask, will the shrna pass the first processing stage. If it passes, then the goes to the second classifier and so on, until it passes all the classifiers, and is deemed a potent shrna. If an shrna doesnt pass any classifier, its processing is stopped, and it is rejected.



Classification results will be added next week for both steps 1,2.