Probabilistic Latent Tensor Factorization for 3-way Microarray Data Analysis with Missing Values

The recent advances in microarray technology enabled the measurement of gene expression levels of samples over a series of time points. Unlike the traditional 2D microarray data, such experiments generate 3D (gene-sample-time) microarray data, which require specialized methods for analysis. In this study, we propose a novel tensor factorization model for modeling 3D microarray data. The model assumes the existence of certain temporal patterns that are repeated over time. One main advantage of the model is that it handles the missing data implicitly, so that the estimation process is not effected by the existence of missing values, which commonly occur in microarray data. We evaluate our model on classification of the good or bad responders to Interferon beta (INFβ) treatments by using a real gene-sample-time microarray data set and achieve a promising prediction performance.