CompScore | Optimization setup

In case a GA optimization is requested, the user must configure some GA parameters. This is done by filling the fields shown below. Default settings are also provided for each parameter and they are described in detail next.

Maximum Allowed Correlation Between Rankings: This parameter controls the maximum allowed correlation between two rankings of different scoring components. It can take values between 0 (correlation allowed) and 1 (no correlation allowed)
Minimum Number of Allowed Scoring Levels: This parameter controls the minimum number of allowed scores levels for a scoring component. It can take any integer value up to the number of compounds in the Data File. Scoring components spanning up to the specified number of unique values will be excluded from the calculations. For example, if set to 1, then only constant variables will be removed from the dataset.

  • Metric to Maximize: Here the user can select which one of the BEDROC or EF metrics are going to me maximized by the GA.
  • Alpha for BEDROC: If BEDROC is selected as the metric to maximize by the GA, the user must specify the value of the α parameter for BEDROC computation. This must be a value greater than 0.
  • Fraction of Screened Data for EF: In case EF is selected for maximization in the GA search, the user must provide a fraction of screened data at which EF should be maximum. This parameter takes values between 0 and 1.
  • Population Size for GA: Number of individuals in the population for GA evolution. Currently, up to 100 individuals are allowed.
  • Generations for GA: Number of generations that the initial population will evolve. Currently, up to 1000 generations are allowed by the CompScore Web Service.
  • CrossOver Probability for GA: Probability parameter for the cross-over operator of the GA. The user must provide a value between 0 and 1.
  • Mutation Probability for GA: Probability parameter for the mutation operator of the GA. The user must provide a value between 0 and 1.
  • Bootstrap: With this parameter the user can choose to perform a bootstrap cross-validation of the best performing consensus scoring solution found by the GA.
  • Number of BootStrap Iterations: If a bootstrap cross-validation of the best solution is requested, the user must provide its number of iterations. Currently, up to 1000 bootstrap re-samplings are allowed.