I got a message from user (Scott Taylor:
email@example.com) regarding new algorithm requests:
Thanks again for the information.
I have just one more question concerning the analysis
capabilities of MeV. We are conducting and
investigation into genes that are good predictors of
patient survival for brain tumors. The SAM censored
survival module provides a useful way to obtain a panel
of genes relavent to survial. The next step we would
like to take is to use these genes to predict the class
membership of unknown samples into some relavent, but
arbitrary, catagories such as survival after 1 yr or 2
yrs. While this can be done with the k-NN module,
Discriminant analysis, or support vectors, the issue I
face is that I am trying to force an arbitrary
(dicotomous or tricotomous) decision rule on a
continuous set of data (SAM output). Thus, error rates
in trying to fit the data into these arbitrary
categories can be appreciable.
I believe, but am not sure, that one alternative would
be to used techinques such as bagging, boosting or
relative risk forests. Are you guys working on
implimenting any of these algortims in subsequent
versions of MeV?
Do you have a suggestion that I may use to work around
my difficulties in the current version of MeV?