Ranking of Gene Regulators through Differential Equations and Gaussian Processes
Antti Honkela, Marta Milo, Matthew Holley, Magnus Rattray and Neil Lawrence
In: 2010 IEEE International Workshop on Machine Learning for Signal Processing (MLSP 2010), 29 Aug - 1 Sep 2010, Kittilä, Finland.
Gene regulation is controlled by transcription factor proteins which themselves are encoded as genes. This gives a network of interacting genes which control the functioning of a cell. With the advent of genome wide expression measurements the targets of given transcription factor have been sought through techniques such as clustering. In this paper we consider the harder problem of finding a gene's regulator instead of its targets. We use a model-based differential equation approach combined with a Gaussian process prior distribution for unobserved continuous-time regulator expression profile. Candidate regulators can then be ranked according to model likelihood. This idea, that we refer to as ranked regulator prediction (RRP), is then applied to finding the regulators of Gata3, an important developmental transcription factor, in the development of ear hair cells.