Using an optimal set of features with a machine learning-based approach to predict effector proteins for Legionella pneumophila

Abstract

Type IV secretion systems exist in a number of bacterial pathogens and are used to secrete effector proteins directly into host cells in order to change their environment making the environment hospitable for the bacteria. In recent years, several machine learning algorithms have been developed to predict effector proteins, potentially facilitating experimental verification. However, inconsistencies exist between their results. Previously we analysed the disparate sets of predictive features used in these algorithms to determine an optimal set of 370 features for effector prediction. This study focuses on the best way to use these optimal features by designing three machine learning classifiers, comparing our results with those of others, and obtaining de novo results. We chose the pathogen Legionella pneumophila strain Philadelphia-1, a cause of Legionnaires' disease, because it has many validated effector proteins and others have developed machine learning prediction tools for it. While all of our models give good results indicating that our optimal features are quite robust, Model 1, which uses all 370 features with a support vector machine, has slightly better accuracy. Moreover, Model 1 predicted 472 effector proteins that are deemed highly probable to be effectors and include 94% of known effectors. Although the results of our three models agree well with those of other researchers, their models only predicted 126 and 311 candidate effectors.

Fig 1. Workflow.

Fig 2. ROC curves for three designed classifiers for 10-fold, cross-validation results.

(a) Model 1, (b) Model 2, and (c) Model 3.

Fig 3. Venn diagram comparing predicted effector proteins for three methods.

The pink circle shows the results for Model 1, the yellow circle for the S4TE method, and the blue circle for the method by Burstein et al.