Redundant feature: Two non-selected features say (f_i) and (f_j) are defined as redundant with respect to a given selected feature say (f_k, if, feature_feature_MI(f_i,f_k)=feature_feature_MI(f_j,f_k)).
Clicking on a feature selects it so it will remain highlighted until another feature is selected; clicking an already selected feature will deselect it.
With this step, one strives for a set of relevant but non-redundant ligand features [4, 5]: "relevant" in the sense that there is a strong association between the selected features and the bioactivity, and "non-redundant" in the sense that these features are conditionally independent given the bioactivity.
Once the features are identified, validation and test accuracies are computed using a standard (non-sparse) SVM built from the training data using the selected features, and evaluated on the validation or test data.
The robustness of the selected features was investigated by dividing the datasets in five non-overlapping subsets, and performing feature selection in each of them, ensuring five completely independent feature selections.
The TFBS families represented by the selected features were then regarded as the most important, relevant, and non-redundant features of all the 35 families.
From the features listed in Table 2, we can see selected features including both geometry features and texture features, which indicate that both geometry and texture features are useful to separate MC from non-MC.
3/un-select the previouly selected features.
Once, selected features have been never removed.
The selected features are given by (22).
The mRMR feature list also consists of the 500 preferentially selected features, where a small index of a feature implies that the feature is very important for separating the hydroxylated sites and the non-hydroxylated sites.
