Default Priors for Neural Network Classification

Feedforward neural networks are a popular tool for classification, offering a method for fully flexible modeling. This paper looks at the underlying probability model, so as to understand statistically what is going on in order to facilitate an intelligent choice of prior for a fully Bayesian analysis. The parameters turn out to be difficult or impossible to interpret, and yet a coherent prior requires a quantification of this inherent uncertainty. Several approaches are discussed, including flat priors, Jeffreys priors and reference priors.     

Coilin levels and modifications influence artificial reporter splicing

Cajal bodies (CBs) and Gems are nuclear domains that contain factors responsible for spliceosomal small nuclear ribonucleoprotein (snRNP) biogenesis. The marker protein for CBs is coilin. In addition to snRNPs, coilin and other factors, canonical CBs contain the survivor of motor neuron protein (SMN). SMN can also localize to Gems. Considering the important role that coilin plays in the formation and composition of CBs, we tested the splicing efficiency of several cell lines that vary in regards to coilin level and modification using an artificial reporter substrate. We found that cells with both hypomethylated coilin and Gems are more efficient at reporter splicing compared to cells in which SMN localizes to CBs. In contrast, coilin reduction, which induces Gem formation, decreases cell proliferation and artificial reporter splicing. These findings demonstrate that coilin modifications or levels impact artificial reporter splicing, possibly by influencing snRNP biogenesis. Received 26 December 2007; received after revision 5 February 2008; accepted 7 February 2008