Statistical Analysis of Mascot Peptide Identification with Active Logistic Regression

We apply active learning and logistic regression to perform statistical analysis of Mascot peptide identification.Uncertainty sampling is used to select examples for labeling, and selected examples are labeled with reference data as the oracle. In each iteration of active learning, the penalized Newton-Raphson method is used to solve the logistic regression model. By testing the method on two datasets with known validity, the results have demonstrated that the proposed method can assign accurate probabilities to Mascot peptide identifications and have a high discrimination power to separate correct and incorrect peptide identifications. By use of active learning, superior classifiers have been achieved with a significantly reduced training dataset.