Module: Parameter optimization

IMPI includes methods for parameter optimization and data revision. IMPI allows adjustable parameters. Using the implemented evolution strategy (ES) provides determined parameters which best fit the given dataset. Therefore, samples with pre-defined MAF are required (= y_est).

For applying an ES: Tools > Parameter Optimization > Evolution Strategy

Definition of expected MAFs for later-on parameter optimization for determining best fitting parameters for a given dataset using an evolution strategy (ES)

When all samples are defined, the implemented ES algorithm examines the best fitting parameters for the dataset. For this purpose, n solution candidates are generated with randomly defined parameter settings. These randomly set parameters are applied to all provided files. In order to identify the best set of parameters, a so-called fitness score is calculated, which provides the parameter values with which the expected MAFs can be achieved best. Each solution candidate is scored with the following fitness function:

\[\begin{split}fitness~score = mse(y_{est_{unclustered}}, y_{true_{unclustered}}) + \Delta MAF_{unclustered} + \\ mse(y_{est_{clustered}}, y_{true_{clustered}}) + \Delta MAF_{clustered}\end{split}\]

where the mean squared error (MSE) of y_est and y_true for unclustered and clustered results is calculated. y_est and y_true are vectors containing the estimated and expected MAFs. ΔMAF is calculated as the difference between the expected and actual MAF at the specified gene loci. The ES aims to find the one candidate with the lowest fitness score.

Resulting best fitting parameters for the given dataset will be displayed.