Nonlinear Model-Based Method for Clustering Periodically Expressed Genes

Algorithm 2

The procedure for evaluating proposed clustering method.

(1) Repeat the following B times (where B is a preset integer number).

 (a) Randomly divide the original dataset into two nonoverlapping sets, a

learning set 𝐿, and a test set 𝑇.

 (b) Apply the evaluated method to the learning set 𝐿 to obtain a

partition𝑃(∘,𝐿).

 (c) Construct a predictor (classifier) 𝐶(∘,𝐿) using the cluster labels from the

partition𝑃(∘,𝐿).

 (d) Apply the predictor C(∘,𝐿) to the test set T to get the predicted partition

𝑃(∘,𝑇).

 (e) Apply the evaluated method to the test set 𝑇 to obtain a partition 𝑃(∘,𝑇).

 (f) Calculate the ARI of partitions 𝑃(∘,𝑇) and 𝑃(∘,𝑇).

(2) Calculate the average ARI (AARI) over the B times as the measure index of the

proposed clustering method.

(3) For the various number of clusters, 𝐾, repeat the procedure described in steps

(1) and (2) above to get AARI(𝐾), and then plot AARI(𝐾) with respect to 𝐾.