A Transfer Learning Approach for Predictive Modeling of Degenerate Biological Systems.

Modeling of a new domain can be challenging due to scarce data and high-dimensionality. Transfer learning aims to integrate data of the new domain with knowledge about some related old domains, in order to model the new domain better. This paper studies transfer learning for degenerate biological systems. Degeneracy refers to the phenomenon that structurally different elements of the system perform the same/similar function or yield the same/similar output. Degeneracy exits in various biological systems and contributes to the heterogeneity, complexity, and robustness of the systems. Modeling of degenerate biological systems is challenging and models enabling transfer learning in such systems have been little studied. In this paper, we propose a predictive model that integrates transfer learning and degeneracy under a Bayesian framework. Theoretical properties of the proposed model are studied. Finally, we present an application of modeling the predictive relationship between transcription factors and gene expression across multiple cell lines. The model achieves good prediction accuracy, and identifies known and possibly new degenerate mechanisms of the system.