Page History

Definition

Using model ensembles is a popular technique that allows to improve prediction accuracy and obtain additional statistical measures for every prediction. Suppose we have an ensemble of M predictors {P₁ⁱ,P₂ⁱ,...,P_Mⁱ} for a molecule i. The ensemble average: 1/M∑_k P_kⁱ is frequently used as the ensemble prediction for the molecule.

CORREL measure uses ensemble predictions for molecules i,j to determine similarity of their similarityensemble responses. It is calculated as Pearson linear or Spearman rank correlation coefficients.

The main assumption of CORREL is that if the two molecules are under-predicted or over-predicted by the same ensemble models, then the ensemble response for these two molecules is "similar".

References

CORREL is used in the Associate Neural Network (ASNN) method [Tetko 2008] to provide bias correction of ensemble predictions.