Construction of precise support vector machine based models for predicting promoter strength

Abstract

Background: The prediction of the prokaryotic promoter strength based on its sequence is of great importance not only in the fundamental research of life sciences but also in the applied aspect of synthetic biology. Much advance has been made to build quantitative models for strength prediction, especially the introduction of machine learning methods such as artificial neural network (ANN) has significantly improve the prediction accuracy. As one of the most important machine learning methods, support vector machine (SVM) is more powerful to learn knowledge from small sample dataset and thus supposed to work in this problem. Methods: To confirm this, we constructed SVM based models to quantitatively predict the promoter strength. A library of 100 promoter sequences and strength values was randomly divided into two datasets, including a training set (⩾10 sequences) for model training and a test set (⩾10 sequences) for model test. Results: The results indicate that the prediction performance increases with an increase of the size of training set, and the best performance was achieved at the size of 90 sequences. After optimization of the model parameters, a high-performance model was finally trained, with a high squared correlation coefficient for fitting the training set (R2 > 0.99) and the test set (R2 > 0.98), both of which are better than that of ANN obtained by our previous work. Conclusions: Our results demonstrate the SVM-based models can be employed for the quantitative prediction of promoter strength.