Abstract
Motivation Identification of biochemical processes that drive the transformation of a totipotent cell into various cell types is essential to our understanding of living systems. This complex machinery determines how tissues differ in terms of their anatomy, physiology, morphology, and, more importantly, how various cellular control mechanisms contribute to the observed similarities/ differences. Tissue-selective genes orchestrate various aspects of cellular machinery in different tissues, and are known to be implicated in a number of tissue-specific pathologies.
Results We propose a novel statistical approach that identifies and removes the effect of universally expressed genes in groups of tissues. This allows us to better characterize tissue similarities, as well as to identify tissue-selective genes. We use our method to construct a reliable hierarchy of tissue similarities. The groupings of tissues in this hierarchy are used to specify successively refined priors for identifying tissue-selective functions and their corresponding genes in the reduced subspace. We show that our refinement process enhances the signal-to-noise ratio in the identification of markers. Using case studies of immune cells and brain tissues, we show that our approach significantly outperforms the state-of-the-art methods, both in terms of coverage and reliability of the predicted tissue-selective genes.
Conclusions Our statistical approach provides a general framework for enhancing the sensitivity of marker detection methods, which can be used in conjunction with other techniques. Even in cases where the number of available expression datasets is limited, we show that our marker detection method outperforms existing techniques. We present detailed validation on immune cells and brain tissues in this paper. Our approach can be applied to construct similar datasets of other human tissues as well, for identifying tissue-specific genes. We demonstrate how these tissue-selective genes enhance our understanding of differentiating biochemical features of brain tissues, shed light on how tissue-selective pathologies progress, and help us identify specific biomarkers and targets for future interventions.
