Enhancing prediction understandability for transmembane segments by BoostingFOIL

In recent years, many studies have focused on improving the accuracy of prediction of trans-membrane segments, and many significant results have been achieved. In spite of these considerable results, the existing methods lack the ability to explain the process of how a learning result is reached and why a prediction decision is made. The explanation of the decision process is important for acceptance of machine learning technology in bioinformatics applications such as protein structure prediction. Decision trees provide insightful interpretation, with form of the propositional IF-THEN rules. The decision tree algorithm produces a large number of rules which is not easy to read and difficult to express adequately complex characteristics of biological sequence. First-order rules with variables have outstanding representation capability, and they can be used to reduce the number of rules. Therefore, in this paper, we present an innovative approach to generate rules for understanding prediction of transmembrane segments. This new approach combines the first-order inductive learning (FOIL) with enhancing techniques of Boosting to produce a new algorithm called BoostingFOIL. The experimental results for prediction of transmembrane segments on 165 low-resolution test data set show that not only the comprehensibility of BoostingFOIL is much better than that of decision tree, but also the test accuracy of these rules is higher as well. The most important contribution of our work is that the first-order rules produced by BoostingFOIL can be easily applied to advanced deduction in inductive learning procedure.