Inductive Logic Programming for Gene Regulation Prediction

摘要

In Table 1, a summary of our experimental results with single, bagged and boosted Tilde decision trees can be found. In the first row, the baseline accuracy of 54.7% indicates that the models clearly improve upon random guessing. The second row shows the reference result by Middendorf et al. [1], where alternating decision trees (ADTs) were applied to a propositional version of the basic data (without the additional predicates). We included the results both with and without the use of FunCat terms. Since the predicates related to protein-protein interactions are not frequently used in the trees, we omitted them altogether in the experiments presented here. Given the same information, boosted decision trees are on par with ADTs (another boosting technique), whereas FunCat terms substantially improve the performance, both in predictive accuracy and compactness. Moreover, it is possible to extract the functional categories affected by theexperimental conditions, together with important transcription factor binding sites and transcription factors for these categories from the induced trees. For a detailed description of the data and a qualitative discussion of the results, we have to refer to the long version of the paper [6].Summing up, we propose a systems biology application of ILP, where the goal is to predict the regulation of a gene under a certain condition from binding site information, the state of regulators, and additional information. We believe that decoding the regulation mechanisms of genes is an exciting new application of learning in logic, requiring data integration from various sources and potentially contributing to a better understanding on a system level.