Then, we get a list of the number of the selected reference tag and calculate the frequency of selection for each reference tag number, then select the first K reference tags with the highest frequency as the final selected reference tags, i.e., f1, f2,…, f K.
The individuals are selected for interbreeding according to their frequency of selection, using one-point crossover method, and then mutation is applied to randomly selected ones.
Biomarker candidates were selected for inclusion in the model based on frequency of selection in four statistical learning approaches (for details see online Appendix B, available at http://care.diabetesjournals.org/cgi/content/full/dc08-1935/DC1).
Fig. 1 PROCESS TYPE (relative frequency of selection in the OCC-P and Matthiessen's (2015) corpus).
This two-step approach was adopted in order to ensure the convergence of the selection towards the same final subset with the highest frequency of selection.
The subsets of molecular descriptors yielding the highest model performance were selected at the end of the second round of GA (forward step selection based on decreased frequency of selection), and were used to fit and calibrate the final models.
The descriptors were then ranked based on frequency of selection during the GA runs, and the final set of descriptors encoding the most relevant structural information to the modelled endpoint was picked in a forward-selection manner.
If the algorithm did not converge during the first 100 runs, another round of 100 runs was performed on the top performing half of the initial set of descriptors, ranked by frequency of selection during the first 100 runs.
The initial 100 independent GA runs were conducted on the full list of the descriptors associated with each dataset, then a second set of 100 independent GA runs were conducted on the 50% of descriptors that showed the highest frequency of selection during the first round.
Also the frequency of selection of each gene over 50 runs of MOGASVM is reported.
