Untitled

Abstract

BackgroundPrediction of the binding ability of antigen peptides to major histocompatibility complex (MHC) class II molecules is important in vaccine development. The variable length of each binding peptide complicates this prediction. Motivated by a text mining model designed for building a classifier from labeled and unlabeled examples, we have developed an iterative supervised learning model for the prediction of MHC class II binding peptides.ResultsA linear programming (LP) model was employed for the learning task at each iteration, since it is fast and can re-optimize the previous classifier when the training sets are altered. The performance of the new model has been evaluated with benchmark datasets. The outcome demonstrates that the model achieves an accuracy of prediction that is competitive compared to the advanced predictors (the Gibbs sampler and TEPITOPE). The average areas under the ROC curve obtained from one variant of our model are 0.753 and 0.715 for the original and homology reduced benchmark sets, respectively. The corresponding values are respectively 0.744 and 0.673 for the Gibbs sampler and 0.702 and 0.667 for TEPITOPE.ConclusionThe iterative learning procedure appears to be effective in prediction of MHC class II binders. It offers an alternative approach to this important predictionproblem.

Predicting MHC-II Binding Affinity Using Multiple Instance Regression

“…Specifically, we have introduced a novel formulation of the problem of learning to predict variable length MHC-II binding peptides as an instance of a multiple instance learning problem. The proposed method shares an attractive feature of some of the recently developed MHC-II binding peptide prediction methods [23], [31] in that it does not require that the 9-mer cores in each binding peptide be identified prior to training the predictor. The 9-mer binding cores are identified by the learning algorithm based on the features of MHC-II binders and non-binders so as to optimize the predictive performance of the learned model.…”

“…The iterative approach for predicting MHC-II peptides [23] can be seen as an exemplar of this class of MIL and MIR methods.…”

“…Qualitative MHC-II binding prediction methods that simply classify MHC peptides into binders and non-binders; Examples of such methods include: (i) methods that use a position weight matrix to model ungapped multiple sequence alignment of MHC binding peptides [10], [11], [12], [13], [14], or rely on Hidden Markov Models (HMMs) [15], [16]; (ii) supervised machine learning methods based on Artificial Neural Networks (ANN) [17], [18] or Support Vector Machines (SVMs) [19], [20], [21], [22]; and (iii) semi-supervised machine learning methods [23], [24]. …”

“…For example, MEME [25], Gibbs sampling [26], matrix optimization techniques (MOTs) (Singh and Raghava, unpublished data), evolutionary algorithms [27], Monte Carlo (MC) search [28], and linear programming [29] form the basis of MHC-II binding peptide prediction methods RankPEP [11], Gibbs [13], HLA-DR4Pred [20], MOEA [14], NetMHCII [10], and LP [23], respectively. The success of these MHC-II prediction methods in identifying MHC-II peptides relies on the effectiveness of the corresponding motif-finding methods in recognizing the motif that characterizes the 9-mer core of MHC-II binding peptides.…”

“…The label associated with each bag could be either binary label indicating whether the corresponding peptide is an MHC-II binder or not or could be numeric label indicating the corresponding binding affinity of the peptide. An attractive feature of the proposed method (that is also shared by some of the recently developed MHC-II binding peptide prediction methods, e.g., [23], [31]) is that it does not require the 9-mer cores in each binding peptide to be identified prior to training the predictor. The 9-mer binding cores are identified by the learning algorithm based on the features of MHC-II binders and non-binders so as to optimize the predictive performance of the learned model.…”