Evaluation Method for Feature Rankings and their Aggregations for Biomarker Discovery
Proceedings of the third International Workshop on Machine Learning in Systems Biology, PMLR 8:122-135, 2009.
In this paper we investigate the problem of evaluating ranked lists of biomarkers, which are typically an output of the analysis of high-throughput data. This can be a list of probes from microarray experiments, which are ordered by the strength of their correlation to a disease. Usually, the ordering of the biomarkers in the ranked lists varies a lot if they are a result of different studies or methods. Our work consists of two parts. First, we propose a method for evaluating the “correctness” of the ranked lists. Second, we conduct a preliminary study of different aggregation approaches of the feature rankings, like aggregating rankings produced from different ranking algorithms and different datasets. We perform experiments on multiple public Neuroblastoma microarray studies. Our results show that there is a generally beneficial effect of aggregating feature rankings as compared to the ones produced by a single study or single method.