Multiobjective evolutionary-based multi-kernel learner for realizing transfer learning in the prediction of HIV-1 protease cleavage sites

Due to the unavailability of adequate patients and expensive labeling cost, many real-world biomedical cases have scarcity in the annotated data. This holds very true for HIV-1 protease specificity problem where only a few experimentally verified cleavage sites are present. The challenge then is to exploit the auxiliary data. However, the problem becomes more complicated when the underlying train and test data are generated from different distributions. To deal with the challenges, we formulate the HIV-1 protease cleavage site prediction problem into a bi-objective optimization problem and solving it by introducing a multiobjective evolutionary-based multi-kernel model. A solution for the optimization problem will lead us to decide the optimal number of base kernels with the best pairing of features. The bi-objective criteria encourage different individual kernels in the ensemble to mitigate the effect of distribution difference in training and test data with the ideal number of base kernels. In this paper, we considered eight different feature descriptors and three different kernel variants of support vector machines to generate the optimal multi-kernel learning model. Non-dominated sorting genetic algorithm-II is employed with bi-objective of achieving a maximum area under the receiver operating characteristic curve simultaneously with a minimum number of features. To validate the effectiveness of the model, the experiments were performed on four HIV-1 protease datasets. The performance comparison with fifteen state-of-the-art techniques on average accuracy and area under curve has been evaluated to justify the improvement of the proposed model. We then analyze Friedman and post hoc tests to demonstrate the significant improvement. The result obtained following the extensive experiment enumerates the bi-objective multi-kernel model performance enhancement on within and cross-learning over the other state-of-the-art techniques.