From: Comparison of evolutionary algorithms in gene regulatory network model inference
Criteria | Description |
---|---|
Goodness of data fit | Best/average Mean Squared Error (MSE) between data and model over a number of runs. This measures the ability of the model to reproduce the experimental data |
Identified interactions | Ability of algorithm to qualitatively identify interactions (Sensitivity/Specificity). An interaction is taken to be identified if the corresponding parameter has an absolute value larger than zero. , Average values over multiple runs are used for comparison purposes. |
Parameter quality | Best/average MSE between real parameters and algorithm solution over multiple runs. This measures the ability of the algorithm to find the exact parameters of a known model (important especially for underspecified systems.) |
Robustness over multiple runs | Average variance of kinetic orders/rate constants over multiple runs |
Robustness to noise | Performance of algorithm with noisy datasets: goodness of fit, identified interactions, parameter quality |
Performance for real microarray data | Sensitivity/Specificity and goodness of fit when applied to real microarray experiments rather than to synthetic data |
Scalability | Performance of algorithms with larger datasets, maximum dimensionality achieved, increase in running time and decrease in goodness of fit and identified parameter quality, (when moving from a smaller to a larger dataset) |
Average running time | Over a number of runs. |
Function calls | Average number of function calls required for the results obtained. |