Investigating topic models' capabilities in expression microarray data classification

In recent years a particular class of probabilistic graphicalmodels—called topic models—has proven to represent an useful and interpretabletool for understanding and mining microarray data. In this context, such modelshave been almost only applied in the clustering scenario, whereas theclassification task has been disregarded by researchers. In this paper, wethoroughly investigate the use of topic models for classification of microarray data,starting from ideas proposed in other fields (e.g., computer vision). A classificationscheme is proposed, based on highly interpretable features extracted from topicmodels, resulting in a hybrid generative-discriminative approach; an extensiveexperimental evaluation, involving 10 different literature benchmarks, confirms thesuitability of the topic models for classifying expression microarray data.