Identifying and Assessing Interesting Subgroups in a Heterogeneous Population

Algorithm 1

CAMS.

while       do

 Shuffle the genes

 Partition the genes into disjoint subsets

 for    to    do

  Perform hierarchical clustering on the th subset of the genes

  for the number of clusters ()    do

   Cut the dendrogram from the hierarchical clustering to yield clusters.

   for    to    do

    Take th cluster (a subtype identifier)

    Run hierarchical clustering on patients using only the genes

    in th cluster (Assume that this step yields clusters).

    for    to    do

     Perform two-sample -tests (e.g. relapse yes vs. no)

     with the individuals in th cluster and the whole genes.

     Fit the null distribution of the two sample -statistics

     with known functional forms.

     if

    (Normal approximation for the null distribution is acceptable) then

      Compute the FDR estimate based on the normal approximation.

     end if

     if

    (Normal approximation for the null distribution is not acceptable) then

      Compute the FDR estimate based on permutations.

     end if

     Compute our proposed FDR-based measure ()

     to assess the resulting cluster.

     For a given subtype, compute the -value of by permuting group labels.

    end for

   end for

  end for

 end for

end while