An interpretable semi-supervised classifier with bioinformatics applications

(IB)² seminar    
Isel Grau, Artificial Intelligence Research Group – VUB                                                                              

In the context of medical or bioinformatic applications of machine learning, obtaining data instances is a relatively easy process but labeling them could become quite expensive or tedious. Such scenarios lead to datasets with few labeled instances and a larger number of unlabeled ones. Semi-supervised classification techniques combine labeled and unlabeled data during the learning phase in order to increase classifier’s generalization capability. Regrettably, most successful semi-supervised classifiers do not allow explaining their outcome, thus behaving like black boxes. However, experts in the bioinformatics and medical domains demand a clear understanding of the decision process in order to trust the results of the machine learning algorithms. We present an interpretable self-labeling grey-box classifier that uses a black box to estimate the missing class labels and a white box to make the final predictions. We propose two different approaches for amending the self-labeling process: a first one based on the confidence of the black box and the latter one based on measures from Rough Set Theory. With this approach we aim to achieve interpretability by means of transparency and simplicity, while attaining superior prediction rates when compared with most prominent self-labeling classifiers reported in the literature. We will show our preliminary results on two case studies: the prediction of early folding in proteins and the classification of pathogenicity of Brugada Syndrome variants.